[go: up one dir, main page]

CN117056593B - Space-time multi-scale interest point recommendation method - Google Patents

Space-time multi-scale interest point recommendation method

Info

Publication number
CN117056593B
CN117056593B CN202310936429.0A CN202310936429A CN117056593B CN 117056593 B CN117056593 B CN 117056593B CN 202310936429 A CN202310936429 A CN 202310936429A CN 117056593 B CN117056593 B CN 117056593B
Authority
CN
China
Prior art keywords
period
time period
user
vector
interest
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
Application number
CN202310936429.0A
Other languages
Chinese (zh)
Other versions
CN117056593A (en
Inventor
温雯
邓峰颖
郝志峰
蔡瑞初
陈炳丰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangdong University of Technology
Original Assignee
Guangdong University of Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Guangdong University of Technology filed Critical Guangdong University of Technology
Priority to CN202310936429.0A priority Critical patent/CN117056593B/en
Publication of CN117056593A publication Critical patent/CN117056593A/en
Application granted granted Critical
Publication of CN117056593B publication Critical patent/CN117056593B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9537—Spatial or temporal dependent retrieval, e.g. spatiotemporal queries

Landscapes

  • Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

本发明涉及数据挖掘和推荐系统技术领域，具体涉及一种时空多尺度的兴趣点推荐方法；本发明将用户行为的学习转换成了潜在状态的学习，并以一种结合距离信息的方式引入空间信息，有效地捕捉了用户的移动特征，且时空多尺度的推荐方式增加了兴趣点推荐的多样性。

This invention relates to the field of data mining and recommendation system technology, specifically to a spatiotemporal multi-scale interest point recommendation method. This invention transforms the learning of user behavior into the learning of latent states, and introduces spatial information in a way that combines distance information, effectively capturing the user's movement characteristics. Furthermore, the spatiotemporal multi-scale recommendation method increases the diversity of interest point recommendations.

Description

Space-time multi-scale interest point recommendation method
Technical Field
The invention relates to the technical field of data mining and recommendation systems, in particular to a space-time multi-scale interest point recommendation method.
Background
In recent years, point-of-interest (POI) recommendation is rapidly developed, which provides accurate service and personalized recommendation for users by mining user sign-in data to model the behavior mode of the users, increases the use viscosity of the users on a platform, and is a technology for intelligently matching interest points conforming to user preference or habit through big data and recommending the interest points to a demander. The existing interest point recommendation technology mainly comprises the steps of recommending interest points to users by analyzing original content (User generated content, UGC) of the users, or recommending the interest points to the users according to user information and the interest point information, wherein the problem of inaccurate interest point recommendation caused by high data operation cost and objectivity of original content of the users exists in the technology of recommending the interest points by analyzing the original content of the users, and the problem of inaccurate interest point recommendation caused by sparse user information exists in the technology of recommending the interest points by the user information and the interest point information;
The traditional interest point recommendation method based on collaborative filtering generally regards the behavior preference of a user as fixed, however, the behavior preference of the user often changes along with time, the sign-in behavior often presents a periodicity, the interest point recommendation method based on the space-time information is a main trend of the current interest point recommendation, but the challenges of sparse sign-in data, difficult extraction of space-time sequence features, difficult capture of personalized differences of the user and the like are still faced, and in the prior art, the interest point recommendation is mainly carried out through the following three schemes:
The method utilizes the spatial information such as the position of the user and the position of the POI to establish the spatial relationship between the user and the interest points, thereby improving the accuracy and diversity of recommendation. Such schemes typically consider the user's behavioral preferences as fixed, however, the user's behavioral preferences tend to vary over time, and there is a dependency and periodicity in the movement behavior.
The method utilizes the check-in time of the user, the access period of the interest point and the like to establish the time relationship between the user and the interest point, thereby improving the accuracy and the versatility of recommendation. Such schemes can mine the periodicity of the user's movement behavior to establish a temporal relationship between the user and the points of interest, but ignore important information about geographic locations, and then the user tends to sign in at surrounding points of interest and there is a spatial clustering site.
The method is characterized in that the space-time relationship between the user and the interest points is established by utilizing the time and the position of check-in of the user and the space-time information such as the access period and the geographic distribution of the interest points, so that the accuracy and the interpretability of the recommendation are improved.
Disclosure of Invention
The invention provides a time-space multi-scale interest point recommendation method and device for overcoming the problems that the behavior preference of a user in the prior art often changes along with time, the movement behavior has important information of dependency and periodicity and ignores geographic positions, check-in data are sparse, space-time sequence feature extraction is difficult, and user individuation difference is difficult to capture.
In order to solve the technical problems, the technical scheme of the invention is as follows:
A time-space multi-scale interest point recommendation method comprises the following steps:
s1, preprocessing an interest point sign-in sequence of each period of a user to obtain potential state vectors and distance offset vectors of each period of the next period;
S2, inputting the potential state vector and the distance offset vector of each period of the next period into the decoding multi-layer perceptron together to generate interest points of each period of the next period, wherein the interest points accord with user preference;
and S3, further screening the interest points which are obtained in the step S2 and accord with the preference of the user according to the time range and the space range of the user, and returning the screened interest points to the user.
Further, in step S1, the preprocessing is performed on the interest point sign-in sequence of each period of the user to obtain a potential state vector and a distance offset vector of each period of the next period, which specifically are:
S1.1, acquiring a sign-in sequence sent by a user, and encoding the sign-in sequence of the user into a Multi-hot vector;
S1.2, calculating Multi-hot vectors of each period of the current period of the user to obtain distance offset vectors of each period of the next period;
s1.3, calculating Multi-hot vectors of each period of the current period of the user to obtain potential state vectors of each period of the current period of the user;
S1.4, calculating potential state vectors of each period of the current period to obtain state transition quantity of each period of the current period;
And S1.5, adding the potential state vector of each period of the current period with the corresponding state transition quantity to obtain the potential state vector of each period of the next period.
Further, in step S1.2, the Multi-hot vector of each period of the current period of the user is calculated to obtain a distance offset vector of each period of the next period:
step S1.21, radial basis function values between each pair of interest points are calculated in advance by using a radial basis function kernel, and a radial basis function matrix between the interest points is obtained:
K(poiA,poiB)＝exp(-γ||poiA-poiB||2)
Wherein poi A represents the longitude and latitude of the point of interest A, poi B represents the longitude and latitude of the point of interest B, gamma represents the hyper-parameter of the correlation level of the point of interest, and K represents the radial basis function matrix;
step S1.22, calculating importance scores of sign-in interest points of each period in the previous period by using an attention mechanism;
Step S1.23, calculating a distance offset vector of each period of the next period through a radial basis function matrix and an importance score:
wherein W (4) is POI embedded matrix, For a set of POIs that the user checked in during the j-th period of the t-1 period,For the user to embed the matrix in the j-th period of the t-1 period, w α is a parameter.
Further, the importance score of the sign-in interest point of each period in the previous period is calculated, specifically:
is an importance score.
Further, in step S1.4, the potential state vector of each period of the current period is calculated to obtain the state transition amount of each period of the current period, which specifically includes:
s1.41, modeling a behavior state of a user to obtain each state node in the model, wherein each state node represents a potential state vector of the user in each period and a long-term preference vector of the user;
s1.42, obtaining the dependent influence quantity of the j-th period of the previous period by using a full connection layer;
And S1.43, inputting the potential state vector of each period, the user long-term preference vector and the dependence influence quantity obtained in the step S1.42 into a fully connected network to obtain the state transition quantity of each period in the period.
Further, in step S1.42, a full connection layer is used to obtain the dependent influence of the j-th period of the previous period, specifically:
f inf is a fully-connected layer, For the dependent influence of the jth period of the previous cycle,For a potential state vector of t-1 cycle i periods,Is a potential state vector for j periods of the t-1 cycle.
Further, in step S1.43, the potential state vector of each period, the user long-term preference vector, and the dependency influence obtained in step S3.2 are input into the fully connected network together to obtain the state transition amount of each period in the period
Where sigma is the activation function and,AndAs a learnable parameter, f shift is a nonlinear fully-connected network, and the state transition quantityFor the dependent influence of the jth period of the previous cycle,For t-1 cycle j periods, e u is a user long-term preference vector.
Further, in step S1.5, the potential state vector of each period is added to the corresponding state transition amount to obtain a potential state vector of each period of the next period, where:
further, in step S2, the potential state vector and the distance offset of each period of the next period are input into the decoding multi-layer perceptron together, so as to obtain a predicted value of each period of the next period, and interest points of each period of the next period are generated according to the predicted value, wherein:
S2.1 the distance offset vector of each period of the next period obtained in the step S1.2 And calculating the potential state vector of each period of the next period obtained in the step S1.5 to obtain a predicted value of each period of the next period;
and S2.2, carrying out descending order sequencing on the probability of the predicted value obtained in the step S2.1, and generating interest points of each period of the next period.
Further, in step S2.1, the distance offset vector of each period of the next cycle obtained in step S2 and the potential state vector of each period of the next cycle obtained in step S1.5 are calculated to obtain a predicted value of each period of the next cycle:
where f θ (·) is the decoding multi-layer perceptron, As the distance offset vector,Is the potential state vector for the jth period of the t cycles.
Compared with the prior art, the technical scheme of the invention has the beneficial effects that:
in order to solve the above-mentioned deficiency, a space-time multi-scale interest point recommendation method is provided, the learning of user behavior is converted into the learning of potential states, and the spatial information is introduced in a mode of combining distance information, so that the movement characteristics of the user are effectively captured, the accuracy of the interest point recommendation is improved, and the recommendation method aims at predicting the sign-in interest points of the user in the next period through giving the user-interest point interaction matrix of the user in the previous period.
Drawings
Fig. 1 is a flowchart of a method for recommending interest points in a space-time multi-scale manner according to the present invention.
Detailed Description
The drawings are for illustrative purposes only and are not to be construed as limiting the present patent;
For the purpose of better illustrating the embodiments, certain elements of the drawings may be omitted, enlarged or reduced and do not represent the actual product dimensions;
it will be appreciated by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.
The technical scheme of the invention is further described below with reference to the accompanying drawings and examples.
Example 1
The embodiment provides a method for recommending interest points in a space-time multi-scale manner, as shown in fig. 1, comprising the following steps:
s1, preprocessing an interest point sign-in sequence of each period of a user to obtain potential state vectors and distance offset vectors of each period of the next period;
S2, inputting the potential state vector and the distance offset vector of each period of the next period into the decoding multi-layer perceptron together to generate interest points of each period of the next period, wherein the interest points accord with user preference;
and S3, further screening the interest points which are obtained in the step S2 and accord with the preference of the user according to the time range and the space range of the user, and returning the screened interest points to the user.
In the implementation process, inputting the interest point sign-in sequence of the user in each period of the previous week, coding the sign-in sequence of the user, representing the sign-in sequence as a Multi-hot vector identifiable by a model, for each user, if the user has sign-in behaviors in the current period and the next period, reserving the set of sign-in records, coding the reorganized sign-in records, taking t-1 week as an example, defining the coded sign-in matrix of the current period as X t-1,Where τ represents the number of time periods in one cycle and the coded check-in matrix for the next cycle is defined as X t,The method comprises the steps of inputting Multi-hot vectors of all time periods of a user into a neighborhood perception offset calculator to obtain distance offset of each time period of a next period, inputting Multi-hot vectors of all time periods of the user into a time sequence state encoder to obtain potential state vectors of each time period of the user, inputting the potential state vectors of each time period into a state dependent network to obtain state transition quantity of each time period, adding the potential state vectors of each time period with corresponding state transition quantity to obtain potential state vectors of each time period of the next period, inputting the potential state vectors of each time period of the next period and the distance offset into a decoding Multi-layer perception machine together to generate interest points of each time period of the next period, further dividing and screening the interest points according to a time range and a space range selected by the user, and sending the screened interest points to be recommended to a user terminal.
In the aspect of time, the application assumes that the time sequence movement behavior of the user is generated by different states in different stages, thereby converting the learning problem of the time sequence preference of the user into the dynamic change of the time sequence state and the dependency learning problem; in terms of space, inspired by Skip-Gram model in word2vec, the application uses similar technology to model the influence exerted by signed POI on candidate POI, combines Gaussian radial basis function kernel and attention mechanism, and introduces space information in a mode based on combined distance information.
In step S1, the preprocessing is performed on the point of interest sign-in sequence of each period of the user to obtain a potential state vector and a distance offset vector of each period of the next period, which specifically are:
S1.1, acquiring a sign-in sequence sent by a user, and encoding the sign-in sequence of the user into a Multi-hot vector;
S1.2, calculating Multi-hot vectors of each period of the current period of the user to obtain distance offset vectors of each period of the next period;
s1.3, calculating Multi-hot vectors of each period of the current period of the user to obtain potential state vectors of each period of the current period of the user;
S1.4, calculating potential state vectors of each period of the current period to obtain state transition quantity of each period of the current period;
And S1.5, adding the potential state vector of each period of the current period with the corresponding state transition quantity to obtain the potential state vector of each period of the next period.
In step S1.2, a Multi-hot vector of each period of the current period of the user is calculated to obtain a distance offset vector of each period of the next period:
step S1.21, radial basis function values between each pair of interest points are calculated in advance by using a radial basis function kernel, and a radial basis function matrix between the interest points is obtained:
K(poiA,poiB)＝exp(-γ||poiA-poiB||2)
Wherein poi A represents the longitude and latitude of the point of interest A, poi B represents the longitude and latitude of the point of interest B, gamma represents the hyper-parameter of the correlation level of the point of interest, and K represents the radial basis function matrix;
step S1.22, calculating importance scores of sign-in interest points of each period in the previous period by using an attention mechanism;
Step S1.23, calculating a distance offset vector of each period of the next period through a radial basis function matrix and an importance score:
wherein W (1) is POI embedded matrix, For a set of POIs that the user checked in during the j-th period of the t-1 period,For the user to embed the matrix in the j-th period of the t-1 period, w α is a parameter.
The importance score of the sign-in interest point of each period in the previous period is calculated, specifically:
is an importance score.
In step S1.4, the potential state vector of each period of the current period is calculated to obtain the state transition quantity of each period of the current period, which specifically includes:
S1.41, in order to model the complex dependence of the user behavior in different time periods, it is assumed that the behavior state of the user can be represented by a graph, the graph is provided with tau state nodes, each node represents the potential state of the user in each time period, and dependency edges and time offsets exist among the nodes;
s1.42, in order to aggregate the dependence of adjacent edges of each state node, a full connection layer f inf is used for learning the dependence influence of each adjacent edge to obtain the dependence influence quantity of the jth period of the previous period
S1.43 potential status of each periodDependency impact derived in user long-term preference vector e u And the state transition quantity of each period in the period is obtained by inputting the state transition quantity into the nonlinear fully-connected network f shift.
In step S1.42, a full connection layer is used to obtain the dependent influence of the jth period of the previous period, specifically:
f inf is a fully-connected layer, For the dependent influence of the jth period of the previous cycle,For a potential state vector of t-1 cycle i periods,Is a potential state vector for j periods of the t-1 cycle.
In step S1.43, the potential state vector of each period, the user long-term preference vector and the dependence influence quantity obtained in step S3.2 are input into the fully connected network together to obtain the state transition quantity of each period in the period
Where sigma is the activation function and,AndAs a learnable parameter, f shift is a nonlinear fully-connected network, and the state transition quantityFor the dependent influence of the jth period of the previous cycle,For potential state vectors of j periods of the t-1 period, e u is a user long-term preference vector, and in order to better capture the mode and long-term preference of user state change, a parameter is added to promote the method to learn the periodic dependence of user check-in.
In step S1.5, the potential state vector of each period is added to the corresponding state transition amount to obtain the potential state vector of each period of the next period, where:
in step S2, the potential state vector and the distance offset vector of each period of the next period are input into the decoding multi-layer perceptron together, and interest points of each period of the next period according with user preference are generated, wherein:
S2.1 the distance offset vector of each period of the next period obtained in the step S1.2 And calculating the potential state vector of each period of the next period obtained in the step S1.5 to obtain a predicted value of each period of the next period;
and S2.2, carrying out descending order sequencing on the probability of the predicted value obtained in the step S2.1, and generating interest points of each period of the next period.
In step S2.1, the distance offset vector of each period of the next cycle obtained in step S2 and the potential state vector of each period of the next cycle obtained in step S1.5 are calculated to obtain the predicted value of each period of the next cycle:
where f θ (·) is the decoding multi-layer perceptron, As the distance offset vector,Is the potential state vector for the jth period of the t cycles.
The same or similar reference numerals correspond to the same or similar components;
The terms describing the positional relationship in the drawings are merely illustrative, and are not to be construed as limiting the present patent;
It is to be understood that the above examples of the present invention are provided by way of illustration only and not by way of limitation of the embodiments of the present invention. Other variations or modifications of the above teachings will be apparent to those of ordinary skill in the art. It is not necessary here nor is it exhaustive of all embodiments. Any modification, equivalent replacement, improvement, etc. which come within the spirit and principles of the invention are desired to be protected by the following claims.

Claims (7)

1.一种时空多尺度的兴趣点推荐方法，其特征在于，包括以下步骤：1. A spatiotemporal multi-scale point of interest recommendation method, characterized by comprising the following steps: S1：将用户各个时段的兴趣点签到序列进行预处理得到下一周期每个时段的潜在状态向量和距离偏置向量；S1: Preprocess the user's interest point check-in sequence for each time period to obtain the potential state vector and distance bias vector for each time period in the next cycle; 所述步骤S1，包括以下步骤：Step S1 includes the following steps: S1.1：获取用户发送的签到序列，对用户的签到序列编码成Multi-hot向量；S1.1: Obtain the check-in sequence sent by the user and encode the user's check-in sequence into a Multi-hot vector; S1.2：将用户当前周期每个时段的Multi-hot向量进行计算，得到下一周期每个时段的距离偏置向量；S1.2: Calculate the Multi-hot vector of each time period in the current period to obtain the distance bias vector of each time period in the next period; 所述步骤S1.2，包括以下步骤：Step S1.2 includes the following steps: 步骤S1.21：使用径向基函数核提前将每一对兴趣点之间的径向基函数值计算，得到兴趣点之间的径向基函数矩阵：Step S1.21: Calculate the radial basis function values between each pair of interest points in advance using the radial basis function kernel, thus obtaining the radial basis function matrix between interest points: K(poiA,poiB)＝exp(-γ||poiA-poiB||2)K(poi A ,poi B )=exp(-γ||poi A -poi B || 2 ) 其中，poiA代表兴趣点A的经纬度，poiB代表兴趣点B的经纬度，γ代表兴趣点相关性水平的超参数，K代表径向基函数矩阵；Where poi A represents the latitude and longitude of point of interest A, poi B represents the latitude and longitude of point of interest B, γ represents the hyperparameter of the correlation level of the points of interest, and K represents the radial basis function matrix. 步骤S1.22：使用注意力机制计算出前一个周期中每个时段签到兴趣点的重要性得分；Step S1.22: Use the attention mechanism to calculate the importance score of the check-in interest points for each time period in the previous cycle; 步骤S1.23：通过径向基函数矩阵、重要性得分计算得到下一个周期每个时段的距离偏置向量：Step S1.23: Calculate the distance bias vector for each time period of the next cycle using the radial basis function matrix and importance score: 其中，W(4)为POI嵌入矩阵，为用户在t-1周期第j个时段的签到的POI集合，为用户在t-1周期第j个时段的嵌入矩阵，wα为参数为距离偏置向量，为用户在t-1周期第j个时段的径向基函数矩阵；Where W (4) is the POI embedding matrix, Let be the set of POIs that the user checked in during the j-th time period of period t-1. Let w<sub>α</sub> be the embedding matrix of the user in the j-th time period of period t-1, and w <sub>α </sub> be the parameter. This is the distance bias vector. The radial basis function matrix for the user in the j-th time period of period t-1; S1.3：将用户当前周期每个时段的Multi-hot向量进行计算，得到用户在当前周期每个时段的潜在状态向量；S1.3: Calculate the Multi-hot vector of the user in each time period of the current cycle to obtain the potential state vector of the user in each time period of the current cycle; S1.4：将当前周期每个时段的潜在状态向量进行计算，得到当前周期每个时段的状态转移量；S1.4: Calculate the potential state vector for each time period of the current cycle to obtain the state transition amount for each time period of the current cycle; S1.5：将当前周期每个时段的潜在状态向量与对应的状态转移量相加，得到下一周期每个时段的潜在状态向量；S1.5: Add the potential state vector of each time period in the current cycle to the corresponding state transition quantity to obtain the potential state vector of each time period in the next cycle; S2：将下一周期每个时段的潜在状态向量和距离偏置向量一并输入到解码多层感知机中，生成下一周期每个时段符合用户偏好的兴趣点；S2: Input the potential state vector and distance bias vector of each time period in the next cycle into the decoding multilayer perceptron to generate interest points that conform to user preferences for each time period in the next cycle; 所述步骤S2，包括以下步骤：Step S2 includes the following steps: S2.1：将步骤S1.2中得到的下一周期每个时段的距离偏置向量和步骤S1.5中得到的下一个周期每个时段的潜在状态向量进行计算，得到下一周期每个时段的预测值；S2.1: The distance bias vector for each time period of the next cycle obtained in step S1.2 The predicted value for each time period of the next cycle is calculated by combining the potential state vector obtained in step S1.5 with the potential state vector for each time period of the next cycle. S2.2：对步骤S2.1得到的预测值的概率进行降序排序，生成下一周期每个时段兴趣点；S2.2: Sort the probabilities of the predicted values obtained in step S2.1 in descending order to generate points of interest for each time period in the next cycle; S3：根据用户的时间范围和空间范围，对步骤S2中得到符合用户偏好的兴趣点进一步筛选，将筛选后的兴趣点返回至用户。S3: Based on the user's time and space range, further filter the points of interest obtained in step S2 that match the user's preferences, and return the filtered points of interest to the user. 2.根据权利要求1所述的一种时空多尺度的兴趣点推荐方法，其特征在于，计算出前一个周期中每个时段签到兴趣点的重要性得分，具体为：2. The spatiotemporal multi-scale interest point recommendation method according to claim 1, characterized in that, the importance score of the interest points checked in for each time period in the previous period is calculated, specifically as follows: 为重要性得分。 The importance score is calculated based on the score. 3.根据权利要求1所述的一种时空多尺度的兴趣点推荐方法，其特征在于，步骤S1.4中，将当前周期每个时段的潜在状态向量进行计算，得到当前周期每个时段的状态转移量，具体为：3. The spatiotemporal multi-scale interest point recommendation method according to claim 1, characterized in that, in step S1.4, the potential state vector of each time period in the current period is calculated to obtain the state transition amount of each time period in the current period, specifically as follows: S1.41：对用户的行为状态进行建模，获取模型里的每个状态节点，每个状态节点代表用户在各个时段的潜在状态向量、用户长期偏好向量；S1.41: Model the user's behavioral state and obtain each state node in the model. Each state node represents the user's potential state vector and long-term preference vector at various time periods. S1.42：使用一个全连接层得到前一个周期第j时段的依赖影响量；S1.42: Use a fully connected layer to obtain the dependency impact of the j-th time period of the previous cycle; S1.43：将各个时段的潜在状态向量、用户长期偏好向量及步骤S1.42中得到的依赖影响量一并输入到全连接网络中，得到周期内每一时段的状态转移量。S1.43: Input the potential state vectors of each time period, the user's long-term preference vectors, and the dependency influence quantities obtained in step S1.42 into the fully connected network to obtain the state transition quantities for each time period within the cycle. 4.根据权利要求3所述的一种时空多尺度的兴趣点推荐方法，其特征在于，步骤S1.42中，使用一个全连接层得到前一个周期第j时段的依赖影响量，具体为：4. The spatiotemporal multi-scale interest point recommendation method according to claim 3, characterized in that, in step S1.42, a fully connected layer is used to obtain the dependency influence of the j-th time period of the previous cycle, specifically as follows: finf为全连接层，为前一个周期第j时段的依赖影响量，为t-1周期i个时段的潜在状态向量，为t-1周期j个时段的潜在状态向量。f inf is a fully connected layer. This represents the dependency impact in the j-th time period of the previous cycle. Let i be the potential state vector for period i in period t-1. Let be the potential state vector for j time periods in period t-1. 5.根据权利要求4所述的一种时空多尺度的兴趣点推荐方法，其特征在于，步骤S1.43中，将各个时段的潜在状态向量、用户长期偏好向量及步骤S3.2中得到的依赖影响量一并输入到全连接网络中，得到周期内每一时段的状态转移量：5. The spatiotemporal multi-scale interest point recommendation method according to claim 4, characterized in that, in step S1.43, the potential state vector of each time period, the user's long-term preference vector, and the dependency influence quantity obtained in step S3.2 are input together into the fully connected network to obtain the state transition quantity of each time period within the cycle: 其中σ为激活函数，和为可学习的参数，fshift为非线性全连接网络，为状态转移量，为前一个周期第j时段的依赖影响量，为t-1周期j个时段的潜在状态向量，eu为用户长期偏好向量，为周期内每一时段的状态转移量。Where σ is the activation function. and The parameters are learnable, and f- shift represents a non-linear fully connected network. For state transition quantities, This represents the dependency impact in the j-th time period of the previous cycle. Let $ u$ be the latent state vector for $j$ time intervals within period $t-1$, and $e$ be the user's long-term preference vector. This represents the state transition amount for each time period within the cycle. 6.根据权利要求1所述的一种时空多尺度的兴趣点推荐方法，其特征在于，步骤S1.5中，将每个时段的潜在状态向量与对应的状态转移量相加，得到下一周期每个时段的潜在状态向量，其中：6. The spatiotemporal multi-scale interest point recommendation method according to claim 1, characterized in that, in step S1.5, the potential state vector of each time period is added to the corresponding state transition quantity to obtain the potential state vector of each time period in the next cycle, wherein: 7.根据权利要求1所述的一种时空多尺度的兴趣点推荐方法，其特征在于，步骤S2.1中，将步骤S2中得到的下一周期每个时段的距离偏置向量和步骤S1.5中得到的下一个周期每个时段的潜在状态向量进行计算，得到下一周期每个时段的预测值：7. The spatiotemporal multi-scale interest point recommendation method according to claim 1, characterized in that, in step S2.1, the distance bias vector for each time period of the next period obtained in step S2 and the potential state vector for each time period of the next period obtained in step S1.5 are calculated to obtain the predicted value for each time period of the next period: 其中fθ(·)是解码多层感知机，为距离偏置向量，为t个周期第j个时段的潜在状态向量。Where f θ (·) is the decoding multilayer perceptron. This is the distance bias vector. Let be the potential state vector for the j-th time period of t periods.
CN202310936429.0A 2023-07-27 2023-07-27 Space-time multi-scale interest point recommendation method Active CN117056593B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310936429.0A CN117056593B (en) 2023-07-27 2023-07-27 Space-time multi-scale interest point recommendation method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310936429.0A CN117056593B (en) 2023-07-27 2023-07-27 Space-time multi-scale interest point recommendation method

Publications (2)

Publication Number Publication Date
CN117056593A CN117056593A (en) 2023-11-14
CN117056593B true CN117056593B (en) 2025-11-07

Family

ID=88661676

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310936429.0A Active CN117056593B (en) 2023-07-27 2023-07-27 Space-time multi-scale interest point recommendation method

Country Status (1)

Country Link
CN (1) CN117056593B (en)

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113590971A (en) * 2021-08-13 2021-11-02 浙江大学 Interest point recommendation method and system based on brain-like space-time perception characterization

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9461876B2 (en) * 2012-08-29 2016-10-04 Loci System and method for fuzzy concept mapping, voting ontology crowd sourcing, and technology prediction
CA3177530A1 (en) * 2021-07-14 2023-01-14 Strong Force TX Portfolio 2018, LLC Systems and methods with integrated gaming engines and smart contracts

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113590971A (en) * 2021-08-13 2021-11-02 浙江大学 Interest point recommendation method and system based on brain-like space-time perception characterization

Also Published As

Publication number Publication date
CN117056593A (en) 2023-11-14

Similar Documents

Publication Publication Date Title
CN116091152A (en) Recommendation method and system based on multi-level comparison learning and multi-mode knowledge graph
Jiao et al. A novel next new point-of-interest recommendation system based on simulated user travel decision-making process
CN114637911B (en) A next point of interest recommendation method based on attention fusion perception network
CN111400620A (en) User Trajectory Position Prediction Method Based on Spatio-temporal Embedding Self-Attention
CN108804646B (en) Point of interest sign-in prediction method integrating deep learning and factorization machine
CN112541129B (en) Method and device for processing interaction event
CN117194763A (en) Next POI recommendation method based on user preferences and spatiotemporal context information
CN112632404B (en) Next point of interest recommendation method based on multi-granularity self-attention
CN117474072A (en) Method, device, equipment and medium for training graph convolution neural network model
CN117743678A (en) Web API recommendation method based on correlation and compatibility fusion
CN117874342A (en) A method for recommending the next point of interest based on users&#39; long-term and short-term preferences
CN115114462B (en) Model training method, device, multimedia recommendation method, equipment and storage medium
CN117056593B (en) Space-time multi-scale interest point recommendation method
CN119277322B (en) User position complementing method and device, terminal equipment and storage medium
CN119415772A (en) An adaptive recommendation method integrating graph neural network and knowledge graph
Gong et al. STCDM: Spatio-Temporal Contrastive Diffusion Model for Check-In Sequence Generation
Zhang et al. Minimally Supervised Contextual Inference from Human Mobility: An Iterative Collaborative Distillation Framework.
Zeng et al. BMAM: complete the missing POI in the incomplete trajectory via mask and bidirectional attention model
CN116976992A (en) Target object group determination method and device and computer equipment
CN117763300B (en) Intelligent program recommendation method based on temporal graph converter and preference fluctuation
CN117421059B (en) Management system and method for public component library
Lee et al. Deep learning hybrid model for equitable data representation: Physics model incorporating ramp traffic dynamics & data-driven enhancements
CN118014004B (en) Session recommendation method for double-channel graph fusion for processing information loss
CN117094762B (en) User rights and interests recommending method and system based on AI artificial intelligence
CN120744446A (en) Model training method, device, electronic equipment, medium and program product

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