CN118827775B - Intelligent robot-assisted short message push management system and method - Google Patents
Intelligent robot-assisted short message push management system and method Download PDFInfo
- Publication number
- CN118827775B CN118827775B CN202411299746.7A CN202411299746A CN118827775B CN 118827775 B CN118827775 B CN 118827775B CN 202411299746 A CN202411299746 A CN 202411299746A CN 118827775 B CN118827775 B CN 118827775B
- Authority
- CN
- China
- Prior art keywords
- historical
- sequence
- push
- pushing
- feature
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/50—Network services
- H04L67/55—Push-based network services
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0251—Targeted advertisements
- G06Q30/0255—Targeted advertisements based on user history
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
- H04W4/12—Messaging; Mailboxes; Announcements
- H04W4/14—Short messaging services, e.g. short message services [SMS] or unstructured supplementary service data [USSD]
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Business, Economics & Management (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Development Economics (AREA)
- Evolutionary Biology (AREA)
- Signal Processing (AREA)
- Computer Networks & Wireless Communication (AREA)
- Strategic Management (AREA)
- Finance (AREA)
- Accounting & Taxation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Biomedical Technology (AREA)
- Game Theory and Decision Science (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Entrepreneurship & Innovation (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Economics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides an intelligent robot-assisted short message pushing management system and method, relating to the technical field of short message pushing management, comprising the steps of collecting a pushing characteristic set of a target pushing service; the method comprises the steps of acquiring a user historical behavior data sequence set, carrying out multi-scale feature extraction on historical push feedback and service use data by using K self-adaptive convolution network layers to construct a three-dimensional frame sequence, carrying out adjacency matrix fusion analysis to obtain a mapping feature sequence for feedback and service use, carrying out feature fusion analysis through a full-connection layer, reordering user feature sequences, determining personalized push time nodes, and generating a personalized push scheme. The technical problem that the traditional short message pushing generally adopts a general pushing strategy, individual difference and behavior patterns of users are not fully combined, so that the pushing effect is poor is solved, and personalized pushing management is realized through deep analysis of the behavior patterns of the users, so that the technical effect of improving the pushing effect is achieved.
Description
Technical Field
The invention relates to the technical field of short message pushing management, in particular to an intelligent robot-assisted short message pushing management system and method.
Background
The accurate short message pushing can remarkably improve marketing effect and customer satisfaction, however, the traditional short message pushing usually makes a pushing strategy based on simple rules, such as basic information or historical purchasing records of users, dynamic changes of user behaviors are ignored, even if the same customer receives the same service in different time periods, the method cannot adapt to the changes of user demands, and pushing contents cannot be optimized by utilizing specific behavior data of the users in different time periods, so that the pushing contents may not match with interests and demands of the users, and the pushing effect is affected.
Disclosure of Invention
The application provides an intelligent robot-assisted short message pushing management system and method, and aims to solve the technical problems that the conventional short message pushing generally adopts a general pushing strategy, and individual differences and behavior modes of users cannot be fully combined, so that individual requirements of the users cannot be met, and the pushing effect is poor.
The application discloses a first aspect, which provides an intelligent robot-assisted short message pushing management system, comprising a pushing feature acquisition module, a message sending module and a message sending module, wherein the pushing feature acquisition module is used for executing pushing feature acquisition and acquiring a pushing feature set of a target pushing service; the system comprises a history behavior data acquisition module, a multi-scale feature extraction module, a frame sequence acquisition module, a frame sequence analysis module, a three-dimensional frame sequence analysis module and a history analysis module, wherein the history behavior data acquisition module is used for acquiring a history behavior data sequence set of a target pushing user set of the target pushing business collected in advance by an interactive intelligent robot, the history behavior data sequence set comprises a history pushing feedback sub-data sequence set and a history business use sub-data sequence set of the target pushing user set in a preset history collection window, the multi-scale feature extraction module is used for respectively carrying out multi-scale feature extraction on the history pushing feedback sub-data sequence set and the history business use sub-data sequence set by utilizing K self-adaptive convolution network layers according to K receptive fields to obtain K history pushing feedback sub-data feature cluster sequence sets and K history business use sub-data feature cluster sets, the K receptive fields are inconsistent in size, the frame sequence acquisition module is used for constructing an initial three-dimensional frame sequence, the K history pushing feedback sub-data feature cluster sets and the K history business use sub-data feature sets are input into the initial three-dimensional frame sequence, the K frame sequence acquisition module is used for generating a three-dimensional frame sequence and a history feedback frame sequence adjacent to the history pushing feedback sub-data feature sequence, the system comprises a K historical push feedback frame sequence set, a K historical service usage frame sequence set, a characteristic fusion analysis module, a push time node determination module, a short message push management module and a short message push management module, wherein the K historical push feedback frame sequence set and the K historical service usage frame sequence set are subjected to intra-sequence adjacent matrix fusion analysis to obtain K historical push feedback mapping characteristic sequence sets and K historical service usage mapping characteristic sequence sets, the characteristic fusion analysis module is used for transmitting the K historical push feedback mapping characteristic sequence sets and the K historical service usage mapping characteristic sequence sets to a full-connection layer to perform characteristic fusion analysis to obtain a user fusion characteristic sequence set, the push time node determination module is used for reordering the user fusion characteristic sequence set according to the acceptance degree to obtain a user push fusion characteristic sequence set, determining a user push time node sequence set according to the user push fusion characteristic sequence set, and the short message push management module is used for performing user personalized push scheme identification of the target push service according to the user push time node sequence set to obtain a personalized push scheme and performing personalized push service management on the target push service set based on the user personalized push scheme.
The second aspect of the application discloses a short message pushing management method assisted by an intelligent robot, which is implemented by the short message pushing management system assisted by the intelligent robot, and comprises the steps of executing pushing characteristic collection, acquiring a pushing characteristic set of a target pushing service, acquiring a historical behavior data sequence set of a target pushing user set of the target pushing service, which is collected in advance, by the intelligent robot, wherein the historical behavior data sequence set comprises a historical pushing feedback sub-data sequence set and a historical service using sub-data sequence set of the target pushing user set in a preset historical collection window, respectively carrying out multi-scale characteristic extraction on the historical pushing feedback sub-data sequence set and the historical service using sub-data sequence set according to K experience fields by utilizing K adaptive convolution network layers, acquiring a K historical pushing feedback sub-data characteristic cluster set and a K historical service using sub-data characteristic cluster set, wherein the K experience sets are inconsistent in size, constructing an initial three-dimensional frame sequence, respectively carrying out three-dimensional analysis on the K experience feedback sub-data sequence set and the historical feedback sub-data cluster sequence set by using K adaptive convolution network layers, respectively carrying out three-dimensional feature extraction on the historical pushing feedback sub-data cluster sequence set and the historical service using the historical feedback sub-data cluster sequence set, the method comprises the steps of obtaining K historical push feedback mapping feature sequence sets and K historical service use mapping feature sequence sets, transmitting the K historical push feedback mapping feature sequence sets and the K historical service use mapping feature sequence sets to a full-connection layer for feature fusion analysis to obtain a user fusion feature sequence set, reordering the user fusion feature sequence sets from large to small according to acceptance to obtain a user push fusion feature sequence set, determining a user push time node sequence set according to the user push fusion feature sequence set, identifying a user personalized push scheme of a target push service according to the user push time node sequence set, obtaining a user personalized push scheme set, and carrying out short message push management of the target push service based on the user personalized push scheme set.
One or more technical schemes provided by the application have at least the following technical effects or advantages:
The method comprises the steps of acquiring a push feature collection of a target push service, acquiring a push feature set of the target push service, ensuring acquisition of basic data of the push service, including features of service types, sub-service quantity, sub-service types and the like, wherein the features provide a data basis for subsequent data analysis and push strategy optimization, acquiring historical behavior data of the target push user set of the target push service by an interactive intelligent robot, including a historical push feedback sub-data sequence and a historical service use sub-data sequence, providing comprehensive user behavior data, enabling a system to capture push feedback and service use conditions of a user in different time periods, providing basic data for personalized push, utilizing K self-adaptive convolutional network layers to extract multi-scale features, each convolutional layer extracting features of the historical data through different wild experiences, and obtaining features of the historical push feedback sub-data feature sequence and the historical service use sub-data feature sequence cluster, realizing capture of different levels and scales in the data, thereby more comprehensively determining the user behavior and service use conditions, enabling the multi-feature extraction to be beneficial to improve the expression capacity and the initial performance, enabling a user to capture the feature sequence, integrating the three-dimensional data sequence and the analysis sequence through the frame, integrating the feature sequence with the time-scale analysis sequence and the historical data sequence, enabling the feature sequence to be adjacent to be integrated by the time-scale analysis sequence, enabling the feature sequence to be formed by the frame to be adjacent to the time-scale analysis sequence, and the feature sequence to be integrated, and the feature sequence to be adjacent to the data sequence to be formed by a frame with a frame of the feature analysis sequence, the method comprises the steps of obtaining a user fusion characteristic sequence set, carrying out feature fusion analysis on a full-connection layer, obtaining a user fusion characteristic sequence set, carrying out comprehensive analysis on different characteristics on the full-connection layer, generating comprehensive behavior characteristics of users on different time nodes by comprehensively analyzing the different characteristics, realizing deep fusion of the characteristics, enabling the finally obtained user fusion characteristic sequence to comprehensively reflect the behavior mode and the acceptance of the users, providing more accurate data support for personalized pushing, determining a user pushing time node sequence set according to the acceptance ordering of the user fusion characteristic sequence, identifying the optimal pushing time point of the users, formulating proper pushing time for each user, optimizing the selection of pushing time, enabling pushing content to be sent at the time most acceptable by the users, improving the response rate and the user satisfaction degree of pushing, carrying out personalized pushing scheme identification according to the user pushing time node sequence set, carrying out short message management based on the identification, and meeting the requirements of personalized pushing content, and improving the user preference and the user precision.
The foregoing description is only an overview of the present application, and is intended to be implemented in accordance with the teachings of the present application in order that the same may be more clearly understood and to make the same and other objects, features and advantages of the present application more readily apparent.
Drawings
Fig. 1 is a schematic structural diagram of an intelligent robot-assisted short message pushing management system according to an embodiment of the present application;
Fig. 2 is a schematic flow chart of a method for intelligent robot-assisted short message push management according to an embodiment of the present application.
Reference numerals illustrate the push feature acquisition module 10, the historical behavior data acquisition module 20, the multi-scale feature extraction module 30, the frame sequence acquisition module 40, the adjacency matrix fusion analysis module 50, the feature fusion analysis module 60, the push time node determination module 70 and the short message push management module 80.
Detailed Description
According to the embodiment of the application, through providing the intelligent robot-assisted short message pushing management system and method, the technical problem that the pushing effect is poor due to the fact that the conventional short message pushing generally adopts a general pushing strategy and the individual difference and the behavior mode of the user cannot be fully combined, and the personalized pushing management is realized through deep analysis of the behavior mode of the user, so that the technical effect of improving the pushing effect is achieved.
Having described the basic principles of the present application, various non-limiting embodiments of the present application will now be described in detail with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the application.
In a first embodiment, as shown in fig. 1, an embodiment of the present application provides an intelligent robot-assisted sms push management system, where the system includes:
The push feature collection module 10, wherein the push feature collection module 10 is used for executing push feature collection and obtaining a push feature set of a target push service; the system comprises a historical behavior data acquisition module 20, a multi-scale feature extraction module 30, a frame sequence acquisition module 40, a fusion module 50, a frame sequence analysis module and a fusion module, wherein the historical behavior data acquisition module 20 is used for acquiring a historical behavior data sequence set of a target pushing user set of the target pushing service collected in advance by an interactive intelligent robot, the historical behavior data sequence set comprises a historical pushing feedback sub-data sequence set of the target pushing user set and a historical service using sub-data sequence set in a preset historical acquisition window, the multi-scale feature extraction module 30 is used for respectively carrying out multi-scale feature extraction on the historical pushing feedback sub-data sequence set and the historical service using sub-data sequence set by utilizing K adaptive convolution network layers according to K receptive fields to obtain K historical pushing feedback sub-data feature cluster sequence sets and K historical service using sub-data feature cluster sets, the K receptive fields are inconsistent in size, the frame sequence acquisition module 40 is used for constructing an initial three-dimensional frame sequence, the K pushing feedback sub-data feature cluster set and the K-frame sequence using sub-data sequence are used for fusion of the historical feedback sequence, the K-frame sequence acquisition module is used for analyzing the historical sequence of the adjacent frame sequence of the historical pushing sequence array, the fusion module 50 is used for analyzing the fusion module and analyzing the fusion module is used for analyzing the fusion module, the system comprises a user fusion feature sequence set, a feature fusion analysis module 60, a push time node determination module 70, a short message push management module 80 and a short message push management module 80, wherein the user fusion feature sequence set is obtained by acquiring K historical push feedback mapping feature sequence sets and K historical service usage mapping feature sequence sets, the feature fusion analysis module 60 is used for transmitting the K historical push feedback mapping feature sequence sets and the K historical service usage mapping feature sequence sets to a full-connection layer to perform feature fusion analysis to obtain the user fusion feature sequence set, the push time node determination module 70 is used for reordering the user fusion feature sequence sets from high to low according to acceptance, the user fusion feature sequence set is obtained, the user push fusion feature sequence set is determined according to the user push fusion feature sequence set, the short message push management module 80 is used for performing user personalized push scheme identification of the target push service according to the user push time node sequence set, obtaining user personalized push scheme set and performing short message push management of the target push service based on the user personalized push scheme set.
Further, the multi-scale feature extraction module 30 includes the following operation steps:
The method comprises the steps of respectively carrying out information entropy analysis on a history push feedback sub-data sequence set and a history service use sub-data sequence set to obtain a history push information entropy set and a history service use information entropy set, carrying out centralized analysis on the history push information entropy set and the history service use information entropy set to obtain K centralized information entropies, obtaining a plurality of sample information entropies and a plurality of sample receptive fields as sample data to carry out receptive field identifier training until the training is converged to obtain a receptive field identifier after the training is completed, and sending the K centralized information entropies to the receptive field identifier to carry out analysis to obtain the K receptive fields.
Further, the multi-scale feature extraction module 30 includes the following operation steps:
The method comprises the steps of constructing an information entropy distribution space for the history push information entropy set and the history service use information entropy set, wherein the information entropy distribution space comprises a plurality of space points, each space point corresponds to one information entropy, randomly extracting K space points from the plurality of space points to serve as K initial space points, respectively carrying out iterative global search on the K initial space points in the information entropy distribution space according to a preset iteration step length to obtain K iteration space point sets, and respectively calculating information entropy average values of the K iteration space point sets to obtain the K concentrated information entropies.
Further, the frame sequence obtaining module 40 includes the following operation steps:
The method comprises the steps of respectively carrying out feature extraction on a history push feedback sub-data sequence set and a history service use sub-data sequence set according to a preset maximum receptive field by utilizing a central self-adaptive convolution network layer to obtain a history push feedback sub-data comprehensive feature cluster and a history service use sub-data comprehensive feature cluster, respectively constructing an initial three-dimensional frame based on feature numbers in the history push feedback sub-data comprehensive feature cluster and the history service use sub-data comprehensive feature cluster, copying and sequencing the initial three-dimensional frame according to the number of acquisition time nodes in a preset history acquisition window, and obtaining the initial three-dimensional frame sequence.
Further, the adjacency matrix fusion analysis module 50 includes the following operation steps:
The method comprises the steps of randomly extracting a first historical push feedback frame sequence from the K historical push feedback frame sequence sets, extracting a first historical push feedback frame in the first historical push feedback frame sequence as a first historical push feedback mapping feature, extracting a second first historical push feedback frame in the first historical push feedback frame sequence and the first historical push feedback frame to perform inner product calculation to obtain a second adjacent matrix, performing convolution operation on the second adjacent matrix and the second first historical push feedback frame to obtain a second first historical push feedback mapping feature, constructing fusion analysis through multiple adjacent matrices to obtain a first historical push feedback mapping feature sequence, and performing intra-sequence adjacent fusion analysis on the K historical push feedback frame sequence sets to obtain K historical push feedback mapping feature sequence sets.
Further, the K receptive fields are the characteristic extraction widths of the K adaptive convolution network layers on the input data.
Further, the user fusion feature sequence is a sequence formed by the acceptability of the users in the target push user set to the target push service at different time nodes.
The following detailed description of the intelligent robot-assisted short message push management method will clearly know to those skilled in the art, and the intelligent robot-assisted short message push management system in this embodiment is relatively simple to describe because it corresponds to the method disclosed in the embodiment, and the relevant parts refer to the method part for description.
In a second embodiment, based on the same inventive concept as the intelligent robot-assisted sms push management system in the foregoing embodiment, as shown in fig. 2, an embodiment of the present application provides an intelligent robot-assisted sms push management method, where the method includes:
and executing push feature collection to acquire a push feature set of the target push service.
First, a target push service is defined, the type of the push service is determined, such as marketing promotion, service update, notification reminding and the like, a plurality of sub-services are included in the push service, for example, in a marketing campaign, the sub-services may be different products or service promotions, and the number of the sub-services is determined, for example, one push campaign includes a plurality of sub-services, and each sub-service may be pushed for different user groups or at different time nodes. Relevant service characteristic information is extracted from different data sources according to the target push service, including but not limited to service types, sub-service numbers, sub-service types, target user group characteristics, historical push effects and the like, and the extracted characteristics are integrated to form a push characteristic set.
The interactive intelligent robot acquires a historical behavior data sequence set of a target pushing user set of the target pushing service, wherein the historical behavior data sequence set comprises a historical pushing feedback sub-data sequence set and a historical service use sub-data sequence set of the target pushing user set in a preset historical acquisition window.
The intelligent robot is a data extraction tool, and has the main task of automatically retrieving and extracting corresponding historical behavior data according to a target push user set, and firstly, the user set aimed by a target push service, namely the target push user set, is defined, wherein the users can be groups which have performed a certain behavior or participated in similar push service, such as users who have purchased specific goods and subscribed to related services.
The preset history collection window is a preset time range, and can be set according to service requirements or pushing frequency, and the historical behavior data of the user can be collected in the time range, for example, a collection window of 24 hours can be defined, and historical pushing feedback sub-data and historical service usage sub-data of each hour of the user in the 24 hours can be collected. The historical push feedback sub-data refers to historical push feedback data of a target push user, for example, a series of feedback actions such as clicking, neglecting, replying and the like possibly generated after the user receives a previous push message, wherein the historical action data is service condition data of the target push user on related services, for example, service use frequency, service duration, function preference and the like of the user in a period of time, and the service use sub-data is formed by the data.
And carrying out serialization processing on the extracted historical behavior data, namely organizing original discrete data points into data sequences according to a time sequence, integrating the data sequences of all users to obtain a historical push feedback sub-data sequence set and a historical service use sub-data sequence set, wherein the sequences reflect the time evolution of the user behavior and provide a basis for the subsequent multi-scale feature extraction.
And respectively carrying out multi-scale feature extraction on the historical push feedback sub-data sequence set and the historical service using sub-data sequence set according to K receptive fields by using K self-adaptive convolution network layers to obtain K historical push feedback sub-data feature cluster sequence sets and K historical service using sub-data feature sequence cluster sets, wherein the K receptive fields are inconsistent in size.
The adaptive convolution network layer refers to a convolution layer which can dynamically adjust self parameters according to input data in the characteristic extraction process, and the adaptability enables the network to better capture key characteristics in the input data.
The receptive field refers to an input data range which can be captured by a neuron of a certain layer in the network, in the convolutional neural network, the size of the receptive field is determined by factors such as the size of a convolution kernel, the step length, parameters of a pooling layer and the like, the size of the receptive field directly influences the understanding depth and detail capturing capability of the network on data, K receptive fields with different sizes are adopted, meaning that each convolution layer processes the data by using different convolution kernel sizes and step lengths, and some layers pay more attention to local details and other layers pay more attention to overall trends.
The method comprises the steps that a history push feedback sub-data sequence set and a history service use sub-data sequence set are used as input data, K adaptive convolution network layers are input, K receptive fields are respectively adopted for multi-scale feature extraction, namely, each convolution layer uses different convolution kernel sizes and step sizes to process data, the arrangement allows the network to capture features on different scales, for example, small receptive fields can capture tiny details in the data, and large receptive fields can extract global modes. This inconsistency enables the network to understand the input data from multiple angles, thereby extracting richer features.
After the processing of the convolution layers of K different receptive fields, the historical push feedback sub-data is refined into K different characteristic cluster sequence sets, each set corresponds to one receptive field and represents the characteristics extracted on the scale, and similarly, the historical service use sub-data is also processed by the K convolution layers to generate K different characteristic cluster sequence sets which reflect the service use modes of users in different time periods and use situations.
An initial three-dimensional frame sequence is constructed, the K historical push feedback sub-data characteristic cluster sequence sets and the K historical service usage sub-data characteristic cluster sequence sets are input into the initial three-dimensional frame sequence, and K historical push feedback frame sequence sets and K historical service usage frame sequence sets are generated.
An initial three-dimensional frame sequence is constructed, the three-dimensional frame is a multi-dimensional structure, not only characteristic data per se is researched, but also a time dimension and a relation dimension between different characteristics are introduced, and the characteristic sequences can be represented and organized in a unified space through the three-dimensional frame, so that a time sequence mode and a relation in the data can be captured better.
The method comprises the steps of mapping a K historical push feedback sub-data feature cluster sequence set and a K historical service use sub-data feature cluster sequence set to corresponding positions of an initial three-dimensional frame respectively, wherein each feature cluster sequence corresponds to one position in the three-dimensional frame and is arranged according to time sequence, a frame diagram is a slice in the three-dimensional frame sequence and corresponds to a specific time point or feature type, through the frame diagrams, the change condition of user behaviors or service use features at the specific time point can be intuitively displayed, and each frame diagram corresponds to one feature sequence, so that time sequence connection is constructed for the behaviors of users.
K historical push feedback sub-data feature cluster sequences are arranged according to time and feature dimensions to generate K historical push feedback frame sequence sets, each frame sequence is a set of frame graphs, the frame graphs are connected through a time sequence and reflect a behavior mode of a user in historical push, and similarly, K historical service use frame sequence sets are generated by mapping K historical service use sub-data feature cluster sequences into a three-dimensional frame, and the sequences show service use conditions of the user in different time periods and evolution processes of the use conditions.
And carrying out intra-sequence adjacency matrix fusion analysis on the K historical push feedback frame sequence sets and the K historical service usage frame sequence sets to obtain K historical push feedback mapping feature sequence sets and K historical service usage mapping feature sequence sets.
In graph theory, the adjacency matrix is used to represent the connection relation between nodes in the graph, in this step, the adjacency matrix is used to represent the relation of each feature point in the feature sequence in time, and by constructing the adjacency matrix in the sequence, the dependency relation and the association mode inside the feature sequence can be captured, that is, if two feature points are adjacent in the time sequence, the connection can be established between them.
By fusion analysis of the adjacency matrix in each frame sequence, feature points with strong relevance in the same sequence can be integrated to generate relevant feature representation, so that potential modes of user behaviors and service use features can be captured better. Generating K historical push feedback mapping feature sequence sets by carrying out adjacency matrix fusion analysis on the K historical push feedback frame sequence sets, wherein the mapping feature sequences are further abstractions of original features and comprise a time association relation; similarly, the adjacency matrix fusion analysis is carried out on the K historical service use frame sequence sets, K historical service use mapping feature sequence sets are generated, and the feature sequences can better reflect the behavior patterns of users in different service use scenes.
And transmitting the K historical push feedback mapping feature sequence sets and the K historical services to a full-connection layer by using the mapping feature sequence sets to perform feature fusion analysis to obtain a user fusion feature sequence set.
The fully connected layer is a base layer in the neural network for combining different features into a higher level representation, in which each input node is connected to an output node, so that the input features can generate new output features by a weighted or nonlinear activation function. Feature fusion refers to combining features of different sources or different types to generate a unified feature representation.
And transmitting the K historical push feedback mapping feature sequence sets and the K historical service using the mapping feature sequence sets as inputs to a full-connection layer, wherein in the full-connection layer, a weight matrix is used for carrying out weighting processing on the input feature sequences, each input feature point is multiplied by a corresponding weight value to generate a weighted feature value, a bias value is added on the basis of the weighted feature value, and feature output is further adjusted. The connection layer generates a new feature representation through weighting, biasing and activating function processing of the input feature sequence, and outputs a user fusion feature sequence set which represents the behavior feature of the user in multiple dimensions and is a comprehensive feature sequence set, wherein the user fusion feature sequence is a sequence formed by the acceptability of the user in the target push user set to the target push service at different time nodes.
And reordering the user fusion characteristic sequence set from large to small according to the acceptance degree to obtain a user push fusion characteristic sequence set, and determining a user push time node sequence set according to the user push fusion characteristic sequence set.
In the push scene, the acceptance is an index for measuring the preference degree of the user on the push content or the opportunity, the acceptance can be calculated based on factors such as historical data, a behavior mode, user feedback and the like, the corresponding acceptance score of each user fusion feature sequence is calculated, the score can be obtained based on a weighted combination of a plurality of factors, such as the click rate, the reading duration, the interaction frequency and the like in the historical behavior of the user, the user fusion feature sequence set is reordered according to the calculated acceptance score from large to small, the user push fusion feature sequence set is obtained, and the ordered sequence reflects the acceptance degree of the user at different time points.
According to the high-acceptance time period in the user pushing fusion characteristic sequence set, a proper pushing time node is selected, namely, a time point with higher acceptance is selected for pushing, so that pushing content is guaranteed to be delivered to a user at the best opportunity, the selected pushing time nodes are combined into a user pushing time node sequence set, the time node sequence set is used as a reference basis of a final pushing strategy, a system is guided to push at the proper opportunity, and therefore the acceptance rate and the interaction rate of the user are improved.
And identifying the user personalized pushing scheme of the target pushing service according to the user pushing time node sequence set, obtaining a user personalized pushing scheme set, and carrying out short message pushing management of the target pushing service based on the user personalized pushing scheme set.
According to the historical behaviors and preferences of the user, the push content which is most suitable for the user's needs is identified, the push content is matched with the time node, and a user personalized push scheme is formed, for example, if the acceptance of a certain user at 9 pm is highest, the content related to the user is selected to be pushed in the time period. And generating a personalized pushing scheme set according to the pushing content and the optimal time node of each user, wherein each scheme covers the pushing content, the pushing time and the pushing mode.
According to the personalized pushing scheme set of the user, optimal pushing time and content are selected and sent to the user through the short message channel, and the experience of the user can be remarkably improved, disturbance is reduced, and the acceptance of the user to the pushing content is improved through the accurate pushing scheme and the optimized pushing time.
Further, the method comprises the steps of:
The method comprises the steps of respectively carrying out information entropy analysis on a history push feedback sub-data sequence set and a history service use sub-data sequence set to obtain a history push information entropy set and a history service use information entropy set, carrying out centralized analysis on the history push information entropy set and the history service use information entropy set to obtain K centralized information entropies, obtaining a plurality of sample information entropies and a plurality of sample receptive fields as sample data to carry out receptive field identifier training until the training is converged to obtain a receptive field identifier after the training is completed, and sending the K centralized information entropies to the receptive field identifier to carry out analysis to obtain the K receptive fields.
The information entropy is an index for measuring uncertainty of random variables, and the higher the information entropy is, the larger the uncertainty is, the more the information quantity is rich, and in pushing and service use data, the information entropy is used for identifying parts with higher uncertainty and important information.
First, the history push feedback sub-data sequence set and the history service use sub-data sequence set are preprocessed, such as denoising and normalization processing, so as to ensure that the data is suitable for information entropy calculation, the information entropy of each data sequence is calculated, the common calculation method is that the probability of each possible value is counted, then the entropy value is calculated by utilizing an information entropy formula, and finally a history push information entropy set and a history service use information entropy set are generated, so that the information quantity and uncertainty in history push feedback data and service use data are respectively reflected.
The centralized information entropy is an index obtained by performing aggregation analysis on a plurality of information entropy sets and is used for identifying a main uncertainty source. Summarizing the history push information entropy set and the history service use information entropy set, constructing an information entropy distribution space, performing iterative global search of K initial space points according to the information entropy distribution space, and performing information entropy average calculation according to global search results to obtain K concentrated information entropies.
The receptive field identifier is a machine learning model for identifying the characteristics of receptive fields, in convolutional neural networks, receptive fields refer to the size of the input area perceived by each neuron in the network, and by training the receptive field identifier, it can be determined which receptive fields are most effective for the extraction of data characteristics.
And extracting a plurality of sample receptive fields from different convolution network models to represent the perception range of the different receptive fields on the data. And taking the plurality of sample information entropies and the plurality of sample receptive fields as training data sets, carrying out standardized processing on the data so as to facilitate model training, designing a receptive field identifier model based on a neural network, training by utilizing the sample information entropies and the sample receptive fields, and during the training process, determining how to identify and determine the most effective receptive fields according to the characteristics of the information entropies by model learning, and judging whether the model is trained to be converged or not by monitoring training errors and verification errors, for example, the errors are not obviously reduced any more. When the training converges, an optimized receptive field identifier is obtained that can accurately identify important receptive field features in the data.
The K pieces of concentrated information entropy are input into a trained receptive field identifier, the receptive field identifier analyzes the input data, the most effective receptive field is identified and determined by using the learned model parameters, and finally the K pieces of receptive fields are output to represent the receptive field characteristics most suitable under different concentrated information entropy and are used for extracting key characteristics in the data.
Further, the method comprises the steps of:
The method comprises the steps of constructing an information entropy distribution space for the history push information entropy set and the history service use information entropy set, wherein the information entropy distribution space comprises a plurality of space points, each space point corresponds to one information entropy, randomly extracting K space points from the plurality of space points to serve as K initial space points, respectively carrying out iterative global search on the K initial space points in the information entropy distribution space according to a preset iteration step length to obtain K iteration space point sets, and respectively calculating information entropy average values of the K iteration space point sets to obtain the K concentrated information entropies.
The information entropy distribution space is a space representing information entropy value distribution, in the distribution space, space points are specific examples of information entropy values, each point represents one data point in a history push information entropy set or a history service use information entropy set, and the information quantity and uncertainty of the point in the data are reflected. Specifically, information entropy values are extracted from a history push information entropy set and a history service use information entropy set, the extracted information entropy values are mapped into a multidimensional space, each information entropy value is used as a point in the space, and the whole space is composed of the points, namely the information entropy distribution space.
K space points are randomly selected from a plurality of space points in the constructed distribution space, wherein the purpose of random selection is to ensure that the initial points cover different areas in the space, so as to provide diversified initial conditions, and the K space points which are randomly selected are taken as K initial space points, wherein the initial space points are points for algorithm initialization.
The preset iteration step length is the adjustment amplitude of each step in the iteration process, the moving range of the space point in the distribution space is determined, and the selection of the step length influences the searching efficiency and effect. The iterative global search refers to iterating the initial space point in the information entropy distribution space to find the space point with the best characteristic, and continuously adjusting the position of the space point to enable the space point to reach a global optimal or near optimal state in space.
Specifically, using K initial spatial points extracted randomly as a starting point, calculating information entropy difference values between the K initial spatial points and surrounding spatial points, adding spatial points with difference values within a preset threshold value to an iterative spatial point set, adjusting positions of the spatial points according to preset iteration step sizes to explore different areas of the information entropy distribution space, repeating the steps until iteration stop conditions are met, such as the maximum iteration times are reached or the change of the information entropy difference values is smaller than the preset threshold value, and outputting K iterative spatial point sets.
Accumulating the information entropy values of each space point, dividing the total information entropy value obtained by accumulation by the number of the space points to obtain concentrated information entropy, sequentially calculating the average value of each iteration space point set to obtain final K concentrated information entropies, and representing the overall information entropy characteristics of the points in the information entropy distribution space.
Further, the method comprises the steps of:
The method comprises the steps of respectively carrying out feature extraction on a history push feedback sub-data sequence set and a history service use sub-data sequence set according to a preset maximum receptive field by utilizing a central self-adaptive convolution network layer to obtain a history push feedback sub-data comprehensive feature cluster and a history service use sub-data comprehensive feature cluster, respectively constructing an initial three-dimensional frame based on feature numbers in the history push feedback sub-data comprehensive feature cluster and the history service use sub-data comprehensive feature cluster, copying and sequencing the initial three-dimensional frame according to the number of acquisition time nodes in a preset history acquisition window, and obtaining the initial three-dimensional frame sequence.
The preset maximum receptive field refers to the maximum range of convolution kernels in the convolution network layer, and larger receptive fields can capture a wider range of features. And respectively carrying out feature extraction on the historical push feedback sub-data sequence set and the historical service using sub-data sequence set according to a preset maximum receptive field by utilizing a central self-adaptive convolution network layer, namely adopting a convolution kernel to slide on input data, executing convolution operation to extract features, dynamically adjusting the size of the convolution kernel according to the features of the input data so as to adapt to the features of different data, and outputting a feature map after convolution operation to represent the features extracted under different receptive fields. And through convolution operation, acquiring a comprehensive characteristic cluster of the historical push feedback sub-data and a comprehensive characteristic cluster of the historical service usage sub-data, and comprehensively representing various characteristics in the historical push feedback data and various characteristics in the historical service usage data.
A three-dimensional frame is a spatial representation that maps feature data into a three-dimensional coordinate system, which is used to visualize the different features in the data and their relationships. The method comprises the steps of respectively counting the number of all features in the comprehensive feature cluster of the history push feedback sub-data and the comprehensive feature cluster of the history service using sub-data, drawing a three-dimensional frame in a three-dimensional space, mapping the number of the features into the three-dimensional frame, and constructing an initial three-dimensional frame, so that the relation and distribution among the features can be analyzed more intuitively.
Copying the initial three-dimensional frames according to the number of acquisition time nodes in a preset history acquisition window, wherein each copied frame represents data of a time point, sequencing the copied three-dimensional frames according to a time sequence to form a time sequence, wherein each time node is allocated with a three-dimensional frame to generate an initial three-dimensional frame sequence, and the sequence shows the change condition of history push feedback and service use data along with the time.
Further, the method comprises the steps of:
The method comprises the steps of randomly extracting a first historical push feedback frame sequence from the K historical push feedback frame sequence sets, extracting a first historical push feedback frame in the first historical push feedback frame sequence as a first historical push feedback mapping feature, extracting a second first historical push feedback frame in the first historical push feedback frame sequence and the first historical push feedback frame to perform inner product calculation to obtain a second adjacent matrix, performing convolution operation on the second adjacent matrix and the second first historical push feedback frame to obtain a second first historical push feedback mapping feature, constructing fusion analysis through multiple adjacent matrices to obtain a first historical push feedback mapping feature sequence, and performing intra-sequence adjacent fusion analysis on the K historical push feedback frame sequence sets to obtain K historical push feedback mapping feature sequence sets.
Randomly extracting a frame sequence from the K historical push feedback frame sequence sets, and taking the frame sequence as an analysis object and a first historical push feedback frame sequence.
The method comprises the steps of extracting a first frame from a first historical push feedback frame sequence, wherein the first frame in the sequence refers to the frame of the earliest data point in the time sequence, the frame represents the state at the beginning of the sequence, and the first frame is used as a first historical push feedback mapping characteristic.
A second frame is extracted from the first sequence of historical push feedback frames, the second frame in the sequence referring to a frame that is arranged after the first frame in the time sequence.
The inner product is a standard method for calculating the similarity of two vectors, the inner product calculation is carried out on the second first historical push feedback frame and the first historical push feedback frame to obtain the similarity between the two historical push feedback frames, a second adjacent matrix is generated according to the result of the inner product calculation, the adjacent matrix is a matrix representing the relationship between data points, and in the step, the adjacent matrix is used for representing the association relationship between the first frame and the second frame.
The convolution operation is to set the size and weight of the convolution kernel according to the requirement through the process of carrying out weighted summation on the input data by sliding the convolution kernel, slide the convolution kernel on a second adjacent matrix and a second frame, calculate the weighted summation to obtain a convolution result, and serve as a second first historical push feedback mapping feature to represent the feature extracted in the convolution operation.
And similarly, sequentially constructing adjacent matrixes of the plurality of historical push feedback frame sequences, and performing fusion analysis on the matrixes to obtain a more comprehensive characteristic sequence and a first historical push feedback mapping characteristic sequence.
And similarly, sequentially carrying out intra-sequence adjacency matrix fusion analysis on the K historical push feedback frame sequence sets to obtain K historical push feedback mapping feature sequence sets.
Further, the K receptive fields are the characteristic extraction widths of the K adaptive convolution network layers on the input data.
K receptive fields refer to extracting features in K adaptive convolutional networks using K different receptive field sizes that can accommodate different input data features to capture local information of different scales, each adaptive convolutional network layer having different receptive field widths that enable the network to extract local features of data from different scales, e.g., one layer extracts local details using a 3x3 convolution kernel and another layer extracts larger regional features using a 7x7 convolution kernel, by fusing features from different receptive fields, a multi-scale understanding of the data can be obtained for processing tasks with different spatial scales.
Further, the user fusion feature sequence is a sequence formed by the acceptability of the users in the target push user set to the target push service at different time nodes.
The user fusion feature sequence is aimed at the record of the acceptability change of each user in the target push service, the behavior mode of the user, such as the click rate, response time, interaction frequency and the like of the user, is determined according to the fusion feature sequence at different time nodes, the fusion feature of each user on each time node is obtained through fusion analysis of a full-connection layer, and a time sequence is formed, and the sequences reflect the acceptability change of the user on the push content and help the system to identify the behavior trend and preference of the user.
In summary, the intelligent robot-assisted short message push management method provided by the embodiment of the application has the following technical effects:
The method comprises the steps of acquiring a push feature collection of a target push service, acquiring a push feature set of the target push service, ensuring acquisition of basic data of the push service, including features of service types, sub-service quantity, sub-service types and the like, wherein the features provide a data basis for subsequent data analysis and push strategy optimization, acquiring historical behavior data of the target push user set of the target push service by an interactive intelligent robot, including a historical push feedback sub-data sequence and a historical service use sub-data sequence, providing comprehensive user behavior data, enabling a system to capture push feedback and service use conditions of a user in different time periods, providing basic data for personalized push, utilizing K self-adaptive convolutional network layers to extract multi-scale features, each convolutional layer extracting features of the historical data through different wild experiences, and obtaining features of the historical push feedback sub-data feature sequence and the historical service use sub-data feature sequence cluster, realizing capture of different levels and scales in the data, thereby more comprehensively determining the user behavior and service use conditions, enabling the multi-feature extraction to be beneficial to improve the expression capacity and the initial performance, enabling a user to capture the feature sequence, integrating the three-dimensional data sequence and the analysis sequence through the frame, integrating the feature sequence with the time-scale analysis sequence and the historical data sequence, enabling the feature sequence to be adjacent to be integrated by the time-scale analysis sequence, enabling the feature sequence to be formed by the frame to be adjacent to the time-scale analysis sequence, and the feature sequence to be integrated, and the feature sequence to be adjacent to the data sequence to be formed by a frame with a frame of the feature analysis sequence, the method comprises the steps of obtaining a user fusion characteristic sequence set, carrying out feature fusion analysis on a full-connection layer, obtaining a user fusion characteristic sequence set, carrying out comprehensive analysis on different characteristics on the full-connection layer, generating comprehensive behavior characteristics of users on different time nodes by comprehensively analyzing the different characteristics, realizing deep fusion of the characteristics, enabling the finally obtained user fusion characteristic sequence to comprehensively reflect the behavior mode and the acceptance of the users, providing more accurate data support for personalized pushing, determining a user pushing time node sequence set according to the acceptance ordering of the user fusion characteristic sequence, identifying the optimal pushing time point of the users, formulating proper pushing time for each user, optimizing the selection of pushing time, enabling pushing content to be sent at the time most acceptable by the users, improving the response rate and the user satisfaction degree of pushing, carrying out personalized pushing scheme identification according to the user pushing time node sequence set, carrying out short message management based on the identification, and meeting the requirements of personalized pushing content, and improving the user preference and the user precision.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (7)
1. Intelligent robot assisted short message pushing management system, which is characterized in that the system comprises:
the pushing feature acquisition module is used for executing pushing feature acquisition and acquiring a pushing feature set of a target pushing service;
The historical behavior data acquisition module is used for acquiring a historical behavior data sequence set of a target pushing user set of the target pushing service, which is collected in advance, by the interactive intelligent robot, wherein the historical behavior data sequence set comprises a historical pushing feedback sub-data sequence set and a historical service use sub-data sequence set of the target pushing user set in a preset historical acquisition window;
The multi-scale feature extraction module is used for respectively carrying out multi-scale feature extraction on the historical push feedback sub-data sequence set and the historical service using sub-data sequence set according to K receptive fields by using K self-adaptive convolution network layers to obtain K historical push feedback sub-data feature cluster sequence sets and K historical service using sub-data feature cluster sets, wherein the K receptive fields are inconsistent in size;
The frame sequence acquisition module is used for constructing an initial three-dimensional frame sequence, inputting the K historical push feedback sub-data characteristic cluster sequence sets and the K historical service use sub-data characteristic cluster sequence sets into the initial three-dimensional frame sequence, and generating K historical push feedback frame sequence sets and K historical service use frame sequence sets;
The adjacency matrix fusion analysis module is used for carrying out in-sequence adjacency matrix fusion analysis on the K historical push feedback frame sequence sets and the K historical service usage frame sequence sets to obtain K historical push feedback mapping feature sequence sets and K historical service usage mapping feature sequence sets;
The feature fusion analysis module is used for transmitting the K historical push feedback mapping feature sequence sets and the K historical services to a full-connection layer by using the mapping feature sequence sets to perform feature fusion analysis to obtain a user fusion feature sequence set;
The pushing time node determining module is used for reordering the user fusion characteristic sequence set according to the acceptance degree from large to small to obtain a user pushing fusion characteristic sequence set, and determining a user pushing time node sequence set according to the user pushing fusion characteristic sequence set;
The short message pushing management module is used for identifying a user personalized pushing scheme of the target pushing service according to the user pushing time node sequence set, obtaining a user personalized pushing scheme set and carrying out short message pushing management of the target pushing service based on the user personalized pushing scheme set;
the adjacency matrix fusion analysis module is further used for executing the following steps:
Randomly extracting a first historical push feedback frame sequence from the K historical push feedback frame sequence sets;
extracting a first history push feedback frame in the first history push feedback frame sequence as a first history push feedback mapping feature;
Extracting a second first history push feedback frame in the first history push feedback frame sequence and performing inner product calculation on the second first history push feedback frame and the first history push feedback frame to obtain a second adjacency matrix;
Performing convolution operation on the second adjacency matrix and the second first history push feedback frame to obtain a second first history push feedback mapping characteristic;
the method comprises the steps of constructing fusion analysis through a plurality of adjacency matrixes to obtain a first historical push feedback mapping characteristic sequence;
and carrying out intra-sequence adjacency matrix fusion analysis on the K historical push feedback frame sequence sets to obtain K historical push feedback mapping feature sequence sets.
2. The intelligent robot-assisted text message push management system of claim 1, wherein the multi-scale feature extraction module comprises the following operation steps:
Respectively carrying out information entropy analysis on the historical push feedback sub-data sequence set and the historical service use sub-data sequence set to obtain a historical push information entropy set and a historical service use information entropy set;
carrying out centralized analysis on the history push information entropy set and the history service use information entropy set to obtain K centralized information entropies;
Acquiring a plurality of sample information entropies and a plurality of sample receptive fields as sample data, and performing receptive field identifier training until the training is converged, so as to obtain a receptive field identifier after the training is completed;
and sending the K concentrated information entropies to the receptive field identifier for analysis to obtain the K receptive fields.
3. The intelligent robot-assisted text message push management system of claim 2, wherein the multi-scale feature extraction module comprises the following operation steps:
An information entropy distribution space is constructed for the history push information entropy set and the history service use information entropy set, wherein the information entropy distribution space comprises a plurality of space points, and each space point corresponds to one information entropy;
Randomly extracting K space points from the plurality of space points to serve as K initial space points;
respectively carrying out iterative global search on the K initial space points in the information entropy distribution space according to a preset iterative step length to obtain K iterative space point sets;
And respectively calculating the information entropy average value of the K iteration space point sets to obtain the K concentrated information entropies.
4. The intelligent robot-assisted message push management system of claim 3, wherein the frame sequence acquisition module comprises the following operation steps:
The central self-adaptive convolution network layer is utilized to respectively conduct feature extraction on the historical push feedback sub-data sequence set and the historical service using sub-data sequence set according to a preset maximum receptive field, and a historical push feedback sub-data comprehensive feature cluster and a historical service using sub-data comprehensive feature cluster are obtained;
respectively constructing an initial three-dimensional frame based on the characteristic quantity in the history push feedback sub-data comprehensive characteristic cluster and the history service usage sub-data comprehensive characteristic cluster;
copying and sequencing the initial three-dimensional frame according to the number of acquisition time nodes in a preset history acquisition window to obtain the initial three-dimensional frame sequence.
5. The intelligent robot-assisted text message push management system of claim 1, wherein the K receptive fields are the feature extraction widths of the K adaptive convolutional network layers on the input data.
6. The intelligent robot-assisted sms message delivery management system of claim 1, wherein the user fusion feature sequence is a sequence of receptivity of a user to the target delivery service at different time nodes in the target delivery user set.
7. The intelligent robot-assisted short message push management method is characterized by being implemented based on the intelligent robot-assisted short message push management system according to any one of claims 1-6, and comprises the following steps:
executing push feature collection to obtain a push feature set of a target push service;
The interactive intelligent robot acquires a historical behavior data sequence set of a target pushing user set of the target pushing service, wherein the historical behavior data sequence set comprises a historical pushing feedback sub-data sequence set and a historical service use sub-data sequence set of the target pushing user set in a preset historical acquisition window;
Respectively carrying out multi-scale feature extraction on the historical push feedback sub-data sequence set and the historical service using sub-data sequence set according to K receptive fields by using K self-adaptive convolution network layers to obtain K historical push feedback sub-data feature cluster sequence sets and K historical service using sub-data feature sequence cluster sets, wherein the K receptive fields are inconsistent in size;
Constructing an initial three-dimensional frame sequence, inputting the K historical push feedback sub-data characteristic cluster sequence sets and the K historical service usage sub-data characteristic cluster sequence sets into the initial three-dimensional frame sequence, and generating K historical push feedback frame sequence sets and K historical service usage frame sequence sets;
Performing intra-sequence adjacency matrix fusion analysis on the K historical push feedback frame sequence sets and the K historical service usage frame sequence sets to obtain K historical push feedback mapping feature sequence sets and K historical service usage mapping feature sequence sets;
transmitting the K historical push feedback mapping feature sequence sets and the K historical services to a full-connection layer by using the mapping feature sequence sets to perform feature fusion analysis to obtain a user fusion feature sequence set;
Re-ordering the user fusion feature sequence set according to the acceptance degree from large to small to obtain a user push fusion feature sequence set, and determining a user push time node sequence set according to the user push fusion feature sequence set;
And identifying the user personalized pushing scheme of the target pushing service according to the user pushing time node sequence set, obtaining a user personalized pushing scheme set, and carrying out short message pushing management of the target pushing service based on the user personalized pushing scheme set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202411299746.7A CN118827775B (en) | 2024-09-18 | 2024-09-18 | Intelligent robot-assisted short message push management system and method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202411299746.7A CN118827775B (en) | 2024-09-18 | 2024-09-18 | Intelligent robot-assisted short message push management system and method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN118827775A CN118827775A (en) | 2024-10-22 |
| CN118827775B true CN118827775B (en) | 2024-11-29 |
Family
ID=93075166
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202411299746.7A Active CN118827775B (en) | 2024-09-18 | 2024-09-18 | Intelligent robot-assisted short message push management system and method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118827775B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110825968A (en) * | 2019-11-04 | 2020-02-21 | 腾讯科技（深圳）有限公司 | Information push method, device, storage medium and computer equipment |
| CN111444428A (en) * | 2020-03-27 | 2020-07-24 | 腾讯科技（深圳）有限公司 | Information recommendation method and device based on artificial intelligence, electronic equipment and storage medium |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104462594A (en) * | 2014-12-29 | 2015-03-25 | 北京奇虎科技有限公司 | Method and device for providing user personalized resource message pushing |
| CN111669612B (en) * | 2019-03-08 | 2023-02-28 | 腾讯科技（深圳）有限公司 | Live broadcast-based information delivery method and device and computer-readable storage medium |
| CN111177575B (en) * | 2020-04-07 | 2020-07-24 | 腾讯科技（深圳）有限公司 | A content recommendation method, apparatus, electronic device and storage medium |
| CN115827977B (en) * | 2022-12-12 | 2025-07-18 | 中国工商银行股份有限公司 | Training method and device for user preference prediction model and computer equipment |
-
2024
- 2024-09-18 CN CN202411299746.7A patent/CN118827775B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110825968A (en) * | 2019-11-04 | 2020-02-21 | 腾讯科技（深圳）有限公司 | Information push method, device, storage medium and computer equipment |
| CN111444428A (en) * | 2020-03-27 | 2020-07-24 | 腾讯科技（深圳）有限公司 | Information recommendation method and device based on artificial intelligence, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN118827775A (en) | 2024-10-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11574139B2 (en) | Information pushing method, storage medium and server | |
| US20090006363A1 (en) | Granular Data for Behavioral Targeting | |
| EP3596689A1 (en) | Mixed-initiative machine learning systems and methods for determining segmentations | |
| CN112765480A (en) | Information pushing method and device and computer readable storage medium | |
| CN110956224A (en) | Evaluation model generation method, evaluation data processing method, evaluation model generation device, evaluation data processing equipment and medium | |
| CN108665311B (en) | A time-varying feature similarity calculation and recommendation method for e-commerce users based on deep neural network | |
| CN113469730A (en) | Customer repurchase prediction method and device based on RF-LightGBM fusion model under non-contract scene | |
| EP2578006A1 (en) | Classification of network users based on corresponding social network behavior | |
| CN112258223B (en) | Marketing advertisement click prediction method based on decision tree | |
| CN105468598A (en) | Friend recommendation method and device | |
| CN119539883B (en) | Digital advertising delivery method based on big data | |
| EP4341815A1 (en) | Using machine learning model to make action recommendation to improve performance of client application | |
| CN116431931A (en) | Real-time incremental data statistical analysis method | |
| CN119323452A (en) | Cloud computing-based customer data integrated management system and method | |
| CN112132209B (en) | Attribute prediction method based on biasing characteristics | |
| CN114511016B (en) | User attribution prediction method, device and equipment based on model adaptive adjustment | |
| CN118827775B (en) | Intelligent robot-assisted short message push management system and method | |
| CN118982256B (en) | A BI decision management system and method based on big data | |
| CN117271905B (en) | Crowd image-based lateral demand analysis method and system | |
| Sergue | Customer churn analysis and prediction using machine learning for a B2B SaaS company | |
| CN118364308A (en) | Data processing methods | |
| CN118194022A (en) | Data management method and system based on SaaS | |
| CN116756387A (en) | A time-aware graph neural network session recommendation method and system | |
| CN117523601A (en) | Livestock behavior recognition method and device based on global and local information enhancement | |
| CN114155016A (en) | Click rate estimation method, device, equipment and readable storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |