WO2019184217A1 - Hotspot event classification method and apparatus, and storage medium - Google Patents

Abstract

Provided are a hotspot event classification method and apparatus, and a storage medium. The method comprises: acquiring, in real time from a pre-determined social server, information text issued by a first pre-set number of users, and performing word segmentation on the information text according to a pre-determined word segmentation rule to obtain segmented words corresponding to each piece of information text. Then, the method comprises: extracting a feature word in the segmented words, using a pre-determined probability algorithm to determine an event subject corresponding to the feature word, then calculating, according to a pre-set calculation formula, a hotspot event index value corresponding to the feature word, and judging whether the hotspot event index value is greater than a pre-set threshold; and if the hotspot event index value is greater than the pre-set threshold, then acquiring an information vector of the information text corresponding to the feature word in a pre-set vectorization manner, and inputting the information vector into a pre-trained event classification model so as to determine an event type corresponding to the information text. By means of the present application, an event type of a hotspot event can be rapidly and accurately analyzed.

Description

Hot event classification method, device and storage medium

The present application claims priority to Chinese Patent Application No. 201100252849.6, entitled "Hot Spot Event Classification Method, Apparatus and Storage Medium", which is filed on March 26, 2018, the entire contents of which are incorporated by reference. The way it is combined with this application.

Technical field

The present application relates to the field of information technology, and in particular, to a hot event classification method, apparatus, and computer readable storage medium.

Background technique

With the development of network technology, the use of social media has become more widespread, and the number of events in social media has also increased. In the face of the number of incidents, how to quickly identify the types of events in social media, understand the areas of social media users and hot topics, and make corresponding decisions has become a difficult problem for managers.

At present, the existing classification methods of social media hotspot events are not perfect, and a classification method is needed to accurately and quickly analyze the event types of hot events in the early stage of hot event development.

Summary of the invention

In view of the above, the present application provides a hot event classification method, apparatus, and computer readable storage medium, the main purpose of which is to improve the speed and accuracy of hot event classification on social media.

To achieve the above objective, the present application provides a method for classifying hotspot events, including:

Obtaining step: obtaining, in real time, a first preset number of information texts published by the user from a predetermined server;

Word segmentation step: segmenting the above information text by using a predetermined word segmentation rule to obtain a word segment corresponding to each information text;

Determining step: extracting a feature word preset in the word segment, and determining a event theme corresponding to the feature word by using a predetermined probability algorithm;

Calculating step: calculating a hot event indicator value corresponding to the feature word according to a preset calculation formula;

The classification step is: determining whether the hot event indicator value is greater than a preset threshold, and if the hot event indicator value is greater than a preset threshold, acquiring an information vector of the information text corresponding to the feature word by using a preset vectorization manner, and inputting the information vector In the pre-trained event classification model, the event type corresponding to the information text is determined.

In addition, the present application further provides an electronic device, including: a memory and a processor, the hotspot event classification program is stored on the memory, and the hotspot event classification program is executed by the processor, and the following steps can be implemented:

Obtaining step: obtaining, in real time, a first preset number of information texts published by the user from a predetermined server;

Word segmentation step: segmenting the above information text by using a predetermined word segmentation rule to obtain a word segment corresponding to each information text;

Determining step: extracting a feature word preset in the word segment, and determining a event theme corresponding to the feature word by using a predetermined probability algorithm;

Calculating step: calculating a hotspot event index value corresponding to the feature word according to a preset calculation formula;

The classification step is: determining whether the hot event indicator value is greater than a preset threshold, and if the hot event indicator value is greater than a preset threshold, acquiring an information vector of the information text corresponding to the feature word by using a preset vectorization manner, and inputting the information vector In the pre-trained event classification model, the event type corresponding to the information text is determined.

In addition, in order to achieve the above object, the present application further provides a computer readable storage medium, where the computer readable storage medium includes a hotspot event classification program, and when the hotspot event classification program is executed by a processor, the foregoing can be implemented as described above. Any step in the hot event classification method.

The hot event classification method, the electronic device and the computer readable storage medium proposed by the application obtain the information texts published by the social account in the server, and segment the information text to extract the feature words, and then calculate the maximum corresponding to the feature words. Probabilistic event theme, and using the preset calculation formula to calculate the event index value corresponding to the feature word, and finally vectorizing the information text corresponding to the feature word whose event index value is greater than the preset threshold value, and input the event classification model, thereby accurately Determine the event type of the information text and improve the event classification speed.

DRAWINGS

1 is a schematic diagram of a preferred embodiment of an electronic device of the present application;

2 is a block diagram showing a preferred embodiment of the hot event classification procedure of FIG. 1;

3 is a flowchart of a preferred embodiment of a hot event classification method according to the present application;

FIG. 4 is a flowchart of the training of the event classification model of the present application.

The implementation, functional features and advantages of the present application will be further described with reference to the accompanying drawings.

detailed description

It is understood that the specific embodiments described herein are merely illustrative of the application and are not intended to be limiting.

FIG. 1 is a schematic diagram of a preferred embodiment of an electronic device 1 of the present application.

In this embodiment, the electronic device 1 may be a server, a smart phone, a tablet computer, a personal computer, a portable computer, and other electronic devices having computing functions.

The electronic device 1 includes a memory 11, a processor 12, a network interface 13, a communication bus 14, and a thesaurus 15. The network interface 13 can optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). Communication bus 14 is used to implement connection communication between these components.

The memory 11 includes at least one type of readable storage medium. The at least one type of readable storage medium may be a non-volatile storage medium such as a flash memory, a hard disk, a multimedia card, a card type memory, or the like. In some embodiments, the memory 11 may be an internal storage unit of the electronic device 1, such as a hard disk of the electronic device 1. In other embodiments, the memory 11 may also be an external storage unit of the electronic device 1, such as a plug-in hard disk equipped on the electronic device 1, a smart memory card (SMC), and security. Digital (Secure Digital, SD) card, flash card (Flash Card), etc.

In this embodiment, the memory 11 can be used not only for storing application software and various types of data installed in the electronic device 1, such as the hot event classification program 10, the vocabulary 15, and the like. The vocabulary 15 is used to store all the words and words involved in the word segmentation process and the feature words of the annotation.

The processor 12, in some embodiments, may be a Central Processing Unit (CPU), microprocessor or other data processing chip for running program code or processing data stored in the memory 11, such as performing hotspot event classification. The computer program code of the program 10, the training of the event classification model, and the like.

1 shows only the electronic device 1 having the components 11-15 and the hotspot event classification program 10, but it should be understood that not all illustrated components may be implemented, and more or fewer components may be implemented instead.

Optionally, the electronic device 1 may further include a display, which may be referred to as a display screen or a display unit. In some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, and an Organic Light-Emitting Diode (OLED) touch sensor. The display is used to display information processed in the electronic device 1 and a work interface for displaying visualizations, such as event types of information text.

Optionally, the electronic device 1 may further include a user interface, and the user interface may include an input unit such as a keyboard, a voice output device such as an audio, a headphone, etc., optionally, the user interface may further include a standard wired interface and a wireless interface. .

The electronic device 1 may further include a radio frequency (RF) circuit, a sensor, an audio circuit, and the like, and details are not described herein.

In the embodiment of the electronic device 1 shown in FIG. 1, the program code of the hotspot event classification program 10 is stored in the memory 11 as a computer storage medium, and when the processor 12 executes the program code of the hotspot event classification program 10, the following steps are implemented. :

Obtaining step: obtaining, in real time, a first preset number of information texts published by the user from a predetermined server;

Word segmentation step: segmenting the above information text by using a predetermined word segmentation rule to obtain a word segment corresponding to each information text;

Determining step: extracting a feature word preset in the word segment, and determining a event theme corresponding to the feature word by using a predetermined probability algorithm;

Calculating step: calculating a hot event indicator value corresponding to the feature word according to a preset calculation formula;

The classification step is: determining whether the hot event indicator value is greater than a preset threshold, and if the hot event indicator value is greater than a preset threshold, acquiring an information vector of the information text corresponding to the feature word by using a preset vectorization manner, and inputting the information vector In the pre-trained event classification model, the event type corresponding to the information text is determined.

For specific principles, please refer to the following FIG. 2 for a schematic diagram of a preferred embodiment of the hotspot event classification procedure 10 and FIG. 3 for an overview of a flow chart of a preferred embodiment of the hot event classification method.

As shown in FIG. 2, it is a block diagram of a preferred embodiment of the hotspot event classification program 10 of FIG. A module as referred to in this application refers to a series of computer program instructions that are capable of performing a particular function.

In this embodiment, the hotspot event classification program 10 includes: an acquisition module 110, a word segmentation module 120, a determination module 130, a calculation module 140, a determination module 150, and a classification module 160, and functions or operation steps implemented by the modules 110-160 Both are similar to the above, and will not be described in detail here, exemplarily, for example:

The obtaining module 110 is configured to obtain, in real time, a first preset number of user-published information texts from a predetermined server. The predetermined server may be a social server such as a WeChat server, a Weibo server, or a QQ server. The user refers to a social account of the social server, and the first preset number of users may refer to a part of the social account of the social server, and may also refer to all social accounts of the social server.

The word segmentation module 120 is configured to segment the information text by using a predetermined word segmentation rule to obtain a word segment corresponding to each information text. The predetermined word segmentation rule includes: splitting the obtained information text according to preset type punctuation marks, such as “,”, “.”, “!”, “;”, “?”, and the like. Short sentence. According to the words stored in the thesaurus 15, each short sentence is segmented using the long word priority principle. The long word priority principle refers to finding the longest word from the lexicon 15 that is the same as the short sentence as a participle of the short sentence.

The determining module 130 is configured to extract a feature word preset in the word segment, and determine a event topic corresponding to the feature word by using a predetermined probability algorithm. The feature words are pre-labeled and stored in the thesaurus 15. The predetermined probability algorithm includes calculating a final probability P ₃ according to the first selection probability P ₁ and the second selection probability P ₂ . A second predetermined number of implicit event topics are added between the feature words and the event topic text, the hidden event topics being virtual and having no real meaning. The calculation method of the first selection probability P ₁ : determining the first quantity X ₁ of the feature words contained in each implicit event topic and the hiddenness of each feature word according to the mapping relationship between the predetermined implicit event topic and the feature words A second number X ₂ containing event subjects, the first selection probability P ₁ =1/(X ₁ *X ₂ ) of each feature word for each implicit event subject is determined according to X ₁ and X ₂ . The second selection probability P ₂ is calculated according to a predetermined mapping relationship between the implicit event topic and the event topic, determining a third quantity X ₃ of the implicit event topic contained in each event topic and each implicit event topic belongs to The fourth number X ₄ of event subjects, based on X ₃ and X _{4 ,} determines a second selection probability P ₂ =1/(X ₃ *X ₄ ) for each event topic for each event topic. Substituting P ₁ and P ₂ into a predetermined probability calculation formula, the final probability P ₃ of each feature word for each event subject is calculated. The predetermined probability is calculated as P ₃ =P ₁ *P ₂ .

The calculation module 140 is configured to calculate a hot event indicator value corresponding to the feature word according to a preset calculation formula. Wherein, the preset calculation formula is as follows:

Where v represents the speed of event development, a represents the hot event indicator value, ie “accelerated” of event development, t represents time point, T represents time interval, i is an integer, and t _i represents the time when the i-th feature word appears. Point, X _i represents the number of occurrences of the i-th feature word.

The determining module 150 is configured to determine whether the hot event indicator value is greater than a preset threshold. The preset threshold is preset. When the hot event indicator value is greater than the preset threshold, it indicates that the “acceleration” of the event development of the event subject has exceeded a certain range, and the type of the event should be analyzed immediately.

The classification module 160 is configured to acquire an information vector of the information text corresponding to the feature word by using a preset vectorization manner when the hot event indicator value is greater than a preset threshold, and input the information vector into the pre-trained event classification model. Determine the type of event corresponding to the information text. The preset vectorization manner includes: encoding an user information of the information text by using an automatic encoder to generate a user information vector; and performing word vector coding on the information text by using a predetermined word vector model to generate the information text. The text information vector; the user information vector is combined with the text information vector to generate an information vector corresponding to the information text.

The event classification model is a long-term and short-term memory network model, as shown in FIG. 4, which is a flowchart of the training of the event classification model of the present application, and the training steps of the event classification model are as follows:

Obtaining a third preset number of information texts, and generating an information vector corresponding to each information text, determining an event type corresponding to each information vector according to a predetermined mapping relationship between the information text and the event type, and determining the information vector and the event type Mapping relational data as sample data;

Dividing the sample data into a training set of a first ratio and a verification set of a second ratio, wherein the first ratio is greater than the second ratio;

The event classification model is trained by using sample data in the training set, and the accuracy of the event classification model is verified by using sample data in the verification set after training;

If the accuracy is greater than the preset value, the training is completed. If the accuracy is less than or equal to the preset value, the number of sample data is increased, and then the step of dividing the sample data into the training set and the verification set is returned.

As shown in FIG. 3, it is a flowchart of a preferred embodiment of the hot event classification method of the present application.

In this embodiment, when the processor 12 executes the computer program of the hotspot event classification program 10 stored in the memory 11, the hotspot event classification method includes: Step S10 - Step S60:

In step S10, the obtaining module 110 acquires the information text of the first preset number of users from the predetermined server in real time. The predetermined server may be a social server such as a WeChat server, a Weibo server, or a QQ server. The user refers to a social account of the social server, and the first preset number of users may refer to a part of the social account of the social server, and may also refer to all social accounts of the social server. For example, the information text of the WeChat account of the sales clerk A ₁ in the circle of friends or the group of friends is obtained from the WeChat server.

In step S20, according to the acquired information text, the word segmentation module 120 performs segmentation on the information text by using a predetermined word segmentation rule to obtain a word segment corresponding to each information text. The word segmentation refers to dividing the information text into words or words. For example, the message text is “B ₁ successfully developed the C ₁ product”, and the results after the word segmentation are “B ₁ ”, “success”, “development”, “out”, “to”, “C ₁ ”, “product” ", where B ₁ can be a company or department, and C ₁ can be a product name. The predetermined word segmentation rule includes: splitting the obtained information text according to preset type punctuation marks, such as “,”, “.”, “!”, “;”, “?”, and the like. Short sentence. For example, the information from the start position of the message text (the first word) to the first preset type punctuation mark is a short sentence, the first preset type punctuation mark to the second preset type punctuation mark The information between the two is a short sentence, ..., the information between each of the two preset types of punctuation marks is a short sentence until the information text is completely split into short sentences. However, it should be understood that if there is no preset type punctuation at the end position of the information, the information from the last preset type punctuation mark to the information end position (the last word) is a short sentence. According to the words stored in the thesaurus 15, each short sentence is segmented using the long word priority principle. The long word priority principle refers to finding the longest word from the lexicon 15 that is the same as the short sentence as a participle of the short sentence. Suppose, first required a word phrase is a word T1, starting with a first word begins, starting from a find a longest word in lexicon R ₁ 15, R ₁ and T ₁ part of the same and T ₂ from _a rest portion T1 excluded R, T ₂ again in the same manner until all words and identify the words in the thesaurus from T ₁ as 15, the result is "R ₁ / R ₂ ...... ".

In step S30, if the word segmentation of the information text contains the feature word stored in the thesaurus 15, the determining module 130 determines the event subject corresponding to the feature word by using a predetermined probability algorithm. However, it should be understood that the word segmentation of the information text may not contain feature words or one or more feature words. The feature words are pre-labeled and stored in the lexicon 15.

The predetermined probability algorithm includes: adding a second preset number of implicit event topics between the feature words and the event topic text, the hidden event topics being virtual and having no real meaning. For example, add 50 implicit event themes between the feature word and the event topic text: k ₁ , k ₂ , ..., k ₅₀ . Determining, according to a predetermined mapping relationship between the subject matter of the implicit event and the feature word, a first quantity X ₁ of the feature words contained in each implicit event topic and a second quantity X ₂ of the implicit event subject to which each feature word belongs, A first selection probability P ₁ =1/(X ₁ *X ₂ ) of each feature word for each implied event subject is determined according to the first quantity X ₁ and the second quantity X ₂ . For example, the second number of implicit event topics to which the feature word Y belongs is 5, wherein the first number of feature words contained in one hidden event topic k ₇ is 7, and the feature word Y is the implicit event topic k ₇ The first selection probability is 1/35. Determining, according to a predetermined mapping relationship between the implicit event topic and the event topic, a third quantity X ₃ of the implicit event topic included in each event topic and a fourth number X _{4 of the} event topic to which each implicit event topic belongs, A second selection probability P ₂ =1/(X ₃ *X ₄ ) for each event subject for each event subject is determined according to the third number X ₃ and the fourth number X ₄ . For example, the fourth number of event topics to which the implicit event topic k ₇ belongs is 4, wherein the third number of implicit event topics contained in one event topic Z is 5, then the implicit event topic k _{7 is} for the event topic Z The second selection probability is 1/20. The first selection probability P ₁ and the second selection probability P _{2 are} substituted into a predetermined probability calculation formula, and the distribution of the final probability P ₃ of each feature word to each event topic is calculated. The predetermined probability is calculated as P ₃ =P ₁ *P ₂ . For example, the first selection probability P ₁ of the feature word Y to the implicit event topic k ₇ is 1/35, and the second selection probability P ₂ of the event theme topic k ₇ to the event subject text Z is 1/20, then the feature word The final probability P ₃ of Y to the event subject text Z is 1/700. Similarly, the final probability P3 of the feature word Y to other event subject texts and the final probability P ₃ of each event topic text of other feature words of the information text are calculated. Finally, the event theme of the maximum probability corresponding to each feature word is used as the event topic corresponding to the feature word.

In step S40, the calculation module 140 calculates a hot event indicator value corresponding to each feature word according to a preset calculation formula. Wherein, the preset calculation formula is as follows:

Where v represents the speed of event development, a represents the hot event indicator value, ie “accelerated” of event development, t represents time point, T represents time interval, i is an integer, and t _i represents the time when the i-th feature word appears. Point, X _i represents the number of occurrences of the i-th feature word. Therefore, the hot event indicator value of the event topic corresponding to all feature words is calculated, and the hotspot index value is larger, and the event development trend representing the event topic is faster.

In step S50, the determining module 150 determines whether the hot event indicator value is greater than a preset threshold. The preset threshold is preset. When the hot event indicator value is greater than the preset threshold, it indicates that the “acceleration” of the event development of the event subject has exceeded a certain range, and the type of the event should be analyzed immediately.

Step S60: If the hot event indicator value is greater than the preset threshold, the classification module 150 acquires the information vector of the information text corresponding to the feature word by using a preset vectorization manner, and inputs the information vector into the pre-trained event classification model. Determine the type of event corresponding to the information text. The preset vectorization manner includes: using an automatic encoder, such as an Auto-Encoder, encoding user information of the information text to generate a user information vector. Further, the Auto-Encoder is an unsupervised learning algorithm mainly used for data dimensionality reduction or feature extraction. The information text is then subjected to word vector coding using a predetermined word vector model to generate a text information vector of the information text. The predetermined word vector model may be a Word2Vec model or a Doc2Vec model. For example, the information text is coded using the Word2Vec model to generate a text information vector of the information text. Finally, the user information vector and the text information vector are spliced together to generate an information vector corresponding to the information text.

The event classification model is an LSTM model, as shown in FIG. 4, which is a flowchart of the event classification model training of the present application. The training steps of the event classification model are as follows:

Obtaining a third preset number of information texts, and generating an information vector corresponding to each information text, determining an event type corresponding to each information vector according to a predetermined mapping relationship between the information text and the event type, and determining the information vector and the event type Map relational data as sample data. For example, 100,000 pieces of information text are obtained from the microblog server, the event type of the information text is marked, and the information text is generated into 100,000 corresponding information vectors, and each information is determined according to a predetermined mapping relationship between the information text and the event type. The event type of the text, and the mapping relationship between the information vector and the corresponding event type is taken as sample data.

The sample data is divided into a training set of a first ratio and a verification set of a second ratio, wherein the first ratio is greater than the second ratio. For example, 80% of the sample data, that is, 80,000 sample data are randomly used as the training set, and the remaining 20% of the sample data, that is, 20,000 sample data, is used as the verification set.

The event classification model is trained by using the sample data in the training set, and the accuracy of the event classification model is verified by using the sample data in the verification set after the training. For example, the sample data of 80,000 users in the training set is input into the LSTM model for training, an event classification model is generated, and sample data of 20,000 users in the verification set is input into the generated event classification model for accuracy verification.

If the accuracy is greater than the preset value, the training is completed. If the accuracy is less than or equal to the preset value, the number of sample data is increased, and then the step of dividing the sample data into the training set and the verification set is returned. Assume that the default value is 98%. If the verification accuracy is greater than 98%, the training is completed. If the accuracy is less than 98%, then 20,000 sample data is added, and then the steps of dividing the sample data into the training set and the verification set are returned.

The hot event classification method proposed by the above embodiment obtains the information text published by the user from the server, performs word segmentation processing on the information text, extracts feature words in the word segmentation, and then calculates a maximum probability event of the feature word by using a predetermined probability algorithm. The theme, and using the preset calculation formula to calculate the hot event indicator value of the feature word, vectorizing the information text corresponding to the feature word whose hot spot event index value is greater than the preset value, input the event classification model to determine the event type, and improve the event classification. Efficiency and shorten analysis time.

In addition, the embodiment of the present application further provides a computer readable storage medium, where the computer readable storage medium includes a hotspot event classification program 10, and when the hotspot event classification program 10 is executed by the processor, the following operations are implemented:

Obtaining step: obtaining, in real time, a first preset number of information texts published by the user from a predetermined server;

Word segmentation step: segmenting the above information text by using a predetermined word segmentation rule to obtain a word segment corresponding to each information text;

Determining step: extracting a feature word preset in the word segment, and determining a event theme corresponding to the feature word by using a predetermined probability algorithm;

Calculating step: calculating a hot event indicator value corresponding to the feature word according to a preset calculation formula;

The classification step is: determining whether the hot event indicator value is greater than a preset threshold, and if the hot event indicator value is greater than a preset threshold, acquiring an information vector of the information text corresponding to the feature word by using a preset vectorization manner, and inputting the information vector In the pre-trained event classification model, the event type corresponding to the information text is determined.

Preferably, the preset calculation formula is as follows:

Where v represents the speed of event development, a represents the hot event indicator value, t represents the time point, T represents the time interval, i is an integer, t _i represents the time point at which the i-th feature word appears, and X _i represents the i-th feature The number of times a word appears.

Preferably, the predetermined word segmentation rule comprises:

According to the preset type punctuation marks, the obtained information texts are divided into short sentences;

According to the words stored in the thesaurus, each short sentence is segmented using the long word priority principle.

Preferably, the predetermined probability algorithm comprises:

Adding a second preset number of implicit event topics between the feature word and the event topic text;

Determining, according to a predetermined mapping relationship between the subject matter of the implicit event and the feature word, a first quantity X ₁ of the feature words contained in each implicit event topic and a second quantity X ₂ of the implicit event subject to which each feature word belongs, Determining, according to the first quantity X ₁ and the second quantity X _{2 ,} a first selection probability P ₁ =1/(X ₁ *X ₂ ) of each feature word to each implicit event topic;

Determining, according to a predetermined mapping relationship between the implicit event topic and the event topic, a third quantity X ₃ of the implicit event topic included in each event topic and a fourth number X _{4 of the} event topic to which each implicit event topic belongs, Determining, according to the third quantity X ₃ and the fourth quantity X _{4 ,} a second selection probability P ₂ =1/(X ₃ *X ₄ ) of each implicit event topic for each event topic;

The first selection probability P ₁ and the second selection probability P _{2 are} substituted into a predetermined probability calculation formula, and the distribution of the final probability P ₃ of each feature word to each event topic is calculated.

Preferably, the predetermined probability calculation formula is as follows:

P ₃ =P ₁ *P ₂

Where P ₁ represents the first selection probability, P ₂ represents the second selection probability, and P ₃ represents the final probability.

Preferably, the preset vectorization manner includes:

Encoding the user information of the information text using an automatic encoder to generate a user information vector;

Generating a word vector of the information text using a predetermined word vector model to generate a text information vector of the information text;

The user information vector and the text information vector are spliced together to generate an information vector corresponding to the information text.

Preferably, the event classification model is a long-term and short-term memory network model, and the training steps of the event classification model are as follows:

Obtaining a third preset number of information texts, and generating an information vector corresponding to each information text, determining an event type corresponding to each information vector according to a predetermined mapping relationship between the information text and the event type, and determining the information vector and the event type Mapping relational data as sample data;

Dividing the sample data into a training set of a first ratio and a verification set of a second ratio, wherein the first ratio is greater than the second ratio;

The event classification model is trained by using sample data in the training set, and the accuracy of the event classification model is verified by using sample data in the verification set after training;

If the accuracy is greater than the preset value, the training is completed. If the accuracy is less than or equal to the preset value, the number of sample data is increased, and then the step of dividing the sample data into the training set and the verification set is returned.

The specific implementation manner of the computer readable storage medium of the present application is substantially the same as the specific implementation manner of the foregoing hot event classification method, and details are not described herein again.

The serial numbers of the embodiments of the present application are merely for the description, and do not represent the advantages and disadvantages of the embodiments.

Through the description of the above embodiments, those skilled in the art can clearly understand that the foregoing embodiment method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be through hardware, but in many cases, the former is better. Implementation. Based on such understanding, the technical solution of the present application, which is essential or contributes to the prior art, may be embodied in the form of a software product stored in a storage medium (such as ROM/RAM as described above). , a disk, an optical disk, including a number of instructions for causing a terminal device (which may be a mobile phone, a computer, a server, or a network device, etc.) to perform the methods described in the various embodiments of the present application.

The above is only a preferred embodiment of the present application, and thus does not limit the scope of the patent application, and the equivalent structure or equivalent process transformation made by the specification and the drawings of the present application, or directly or indirectly applied to other related technical fields. The same is included in the scope of patent protection of this application.

Claims (20)

A hotspot event classification method is applied to an electronic device, and the method includes: Obtaining step: obtaining, in real time, a first preset number of information texts published by the user from a predetermined server; Word segmentation step: segmenting the above information text by using a predetermined word segmentation rule to obtain a word segment corresponding to each information text; Determining step: extracting a feature word preset in the word segment, and determining a event theme corresponding to the feature word by using a predetermined probability algorithm; Calculating step: calculating a hot event indicator value corresponding to the feature word according to a preset calculation formula; The classification step is: determining whether the hot event indicator value is greater than a preset threshold, and if the hot event indicator value is greater than a preset threshold, acquiring an information vector of the information text corresponding to the feature word by using a preset vectorization manner, and inputting the information vector In the pre-trained event classification model, the event type corresponding to the information text is determined. The hotspot event classification method according to claim 1, wherein the preset calculation formula is as follows:

Where v represents the speed of event development, a represents the hot event indicator value, t represents the time point, T represents the time interval, i is an integer, t _i represents the time point at which the i-th feature word appears, and X _i represents the i-th feature The number of times a word appears. The hotspot event classification method according to claim 1, wherein the predetermined word segmentation rule comprises: According to the preset type punctuation marks, the obtained information texts are divided into short sentences; According to the words stored in the thesaurus, each short sentence is segmented using the long word priority principle. The hotspot event classification method according to claim 1, wherein the predetermined probability algorithm comprises: Adding a second preset number of implicit event topics between the feature word and the event topic text; Determining, according to a predetermined mapping relationship between the subject matter of the implicit event and the feature word, a first quantity X ₁ of the feature words contained in each implicit event topic and a second quantity X ₂ of the implicit event subject to which each feature word belongs, Determining, according to the first quantity X ₁ and the second quantity X _{2 ,} a first selection probability P ₁ =1/(X ₁ *X ₂ ) of each feature word to each implicit event topic; Determining, according to a predetermined mapping relationship between the implicit event topic and the event topic, a third quantity X ₃ of the implicit event topic included in each event topic and a fourth number X _{4 of the} event topic to which each implicit event topic belongs, Determining, according to the third quantity X ₃ and the fourth quantity X _{4 ,} a second selection probability P ₂ =1/(X ₃ *X ₄ ) of each implicit event topic for each event topic; The first selection probability P ₁ and the second selection probability P _{2 are} substituted into a predetermined probability calculation formula, and the distribution of the final probability P ₃ of each feature word to each event topic is calculated. The hotspot event classification method according to claim 4, wherein the predetermined probability calculation formula is as follows: P ₃ =P ₁ *P ₂ Where P ₁ represents the first selection probability, P ₂ represents the second selection probability, and P ₃ represents the final probability. The hotspot event classification method according to claim 1, wherein the preset vectorization manner comprises: Encoding the user information of the information text using an automatic encoder to generate a user information vector; Generating a word vector of the information text using a predetermined word vector model to generate a text information vector of the information text; The user information vector and the text information vector are spliced together to generate an information vector corresponding to the information text. The hotspot event classification method according to claim 1, wherein the event classification model is a long-term and short-term memory network model, and the training steps of the event classification model are as follows: Obtaining a third preset number of information texts, and generating an information vector corresponding to each information text, determining an event type corresponding to each information vector according to a predetermined mapping relationship between the information text and the event type, and determining the information vector and the event type Mapping relational data as sample data; Dividing the sample data into a training set of a first ratio and a verification set of a second ratio, wherein the first ratio is greater than the second ratio; The event classification model is trained by using sample data in the training set, and the accuracy of the event classification model is verified by using sample data in the verification set after training; If the accuracy is greater than the preset value, the training is completed. If the accuracy is less than or equal to the preset value, the number of sample data is increased, and then the step of dividing the sample data into the training set and the verification set is returned. An electronic device, comprising: a memory and a processor, wherein the memory stores a hotspot event classification program, and the hotspot event classification program is executed by the processor, and the following steps can be implemented: Obtaining step: acquiring, in real time, a first preset number of information texts published by the user from a predetermined social server; Word segmentation step: segmenting the above information text by using a predetermined word segmentation rule to obtain a word segment corresponding to each information text; Determining step: extracting a feature word preset in the word segment, and determining a event theme corresponding to the feature word by using a predetermined probability algorithm; Calculating step: calculating a hotspot event index value corresponding to the feature word according to a preset calculation formula; The classification step is: determining whether the hot event indicator value is greater than a preset threshold, and if the hot event indicator value is greater than a preset threshold, acquiring an information vector of the information text corresponding to the feature word by using a preset vectorization manner, and inputting the information vector In the pre-trained event classification model, the event type corresponding to the information text is determined. The electronic device according to claim 8, wherein the preset calculation formula is as follows:

Where v represents the speed of event development, a represents the hot event indicator value, t represents the time point, T represents the time interval, i is an integer, t _i represents the time point at which the i-th feature word appears, and X _i represents the i-th feature The number of times a word appears. The electronic device according to claim 8, wherein the predetermined word segmentation rule comprises: According to the preset type punctuation marks, the obtained information texts are divided into short sentences; According to the words stored in the thesaurus, each short sentence is segmented using the long word priority principle. The electronic device of claim 8, wherein the predetermined probability algorithm comprises: Adding a second preset number of implicit event topics between the feature word and the event topic text; Determining, according to a predetermined mapping relationship between the subject matter of the implicit event and the feature word, a first quantity X ₁ of the feature words contained in each implicit event topic and a second quantity X ₂ of the implicit event subject to which each feature word belongs, Determining, according to the first quantity X ₁ and the second quantity X _{2 ,} a first selection probability P ₁ =1/(X ₁ *X ₂ ) of each feature word to each implicit event topic; Determining, according to a predetermined mapping relationship between the implicit event topic and the event topic, a third quantity X ₃ of the implicit event topic included in each event topic and a fourth number X _{4 of the} event topic to which each implicit event topic belongs, Determining, according to the third quantity X ₃ and the fourth quantity X _{4 ,} a second selection probability P ₂ =1/(X ₃ *X ₄ ) of each implicit event topic for each event topic; The first selection probability P ₁ and the second selection probability P _{2 are} substituted into a predetermined probability calculation formula, and the distribution of the final probability P ₃ of each feature word to each event topic is calculated. The electronic device according to claim 11, wherein the predetermined probability calculation formula is as follows: P ₃ =P ₁ *P ₂ Where P ₁ represents the first selection probability, P ₂ represents the second selection probability, and P ₃ represents the final probability. The electronic device according to claim 8, wherein the preset vectorization manner comprises: Encoding the user information of the information text using an automatic encoder to generate a user information vector; Generating a word vector of the information text using a predetermined word vector model to generate a text information vector of the information text; The user information vector and the text information vector are spliced together to generate an information vector corresponding to the information text. The electronic device according to claim 8, wherein the event classification model is a long-term and short-term memory network model, and the training steps of the event classification model are as follows: Obtaining a third preset number of information texts, and generating an information vector corresponding to each information text, determining an event type corresponding to each information vector according to a predetermined mapping relationship between the information text and the event type, and determining the information vector and the event type Mapping relational data as sample data; Dividing the sample data into a training set of a first ratio and a verification set of a second ratio, wherein the first ratio is greater than the second ratio; The event classification model is trained by using sample data in the training set, and the accuracy of the event classification model is verified by using sample data in the verification set after training; If the accuracy is greater than the preset value, the training is completed. If the accuracy is less than or equal to the preset value, the number of sample data is increased, and then the step of dividing the sample data into the training set and the verification set is returned. A computer readable storage medium, characterized in that the computer readable storage medium includes a hotspot event classification program, and when the hotspot event classification program is executed by a processor, the following steps can be implemented: Obtaining step: acquiring, in real time, a first preset number of information texts published by the user from a predetermined social server; Word segmentation step: segmenting the above information text by using a predetermined word segmentation rule to obtain a word segment corresponding to each information text; Determining step: extracting a feature word preset in the word segment, and determining a event theme corresponding to the feature word by using a predetermined probability algorithm; Calculating step: calculating a hotspot event index value corresponding to the feature word according to a preset calculation formula; The classification step is: determining whether the hot event indicator value is greater than a preset threshold, and if the hot event indicator value is greater than a preset threshold, acquiring an information vector of the information text corresponding to the feature word by using a preset vectorization manner, and inputting the information vector In the pre-trained event classification model, the event type corresponding to the information text is determined. The computer readable storage medium of claim 15, wherein the predetermined calculation formula is as follows:

Where v represents the speed of event development, a represents the hot event indicator value, t represents the time point, T represents the time interval, i is an integer, t _i represents the time point at which the i-th feature word appears, and X _i represents the i-th feature The number of times a word appears. The computer readable storage medium of claim 15, wherein the predetermined word segmentation rules comprise: According to the preset type punctuation marks, the obtained information texts are divided into short sentences; According to the words stored in the thesaurus, each short sentence is segmented using the long word priority principle. The computer readable storage medium of claim 15, wherein the predetermined probability algorithm comprises: Adding a second preset number of implicit event topics between the feature word and the event topic text; Determining, according to a predetermined mapping relationship between the subject matter of the implicit event and the feature word, a first quantity X ₁ of the feature words contained in each implicit event topic and a second quantity X ₂ of the implicit event subject to which each feature word belongs, Determining, according to the first quantity X ₁ and the second quantity X _{2 ,} a first selection probability P ₁ =1/(X ₁ *X ₂ ) of each feature word to each implicit event topic; Determining, according to a predetermined mapping relationship between the implicit event topic and the event topic, a third quantity X ₃ of the implicit event topic included in each event topic and a fourth number X _{4 of the} event topic to which each implicit event topic belongs, Determining, according to the third quantity X ₃ and the fourth quantity X _{4 ,} a second selection probability P ₂ =1/(X ₃ *X ₄ ) of each implicit event topic for each event topic; The first selection probability P ₁ and the second selection probability P _{2 are} substituted into a predetermined probability calculation formula, and the distribution of the final probability P ₃ of each feature word to each event topic is calculated. The computer readable storage medium of claim 18, wherein the predetermined probability calculation formula is as follows: P ₃ =P ₁ *P ₂ Where P ₁ represents the first selection probability, P ₂ represents the second selection probability, and P ₃ represents the final probability. The computer readable storage medium according to claim 15, wherein the event classification model is a long-term and short-term memory network model, and the training steps of the event classification model are as follows: Obtaining a third preset number of information texts, and generating an information vector corresponding to each information text, determining an event type corresponding to each information vector according to a predetermined mapping relationship between the information text and the event type, and determining the information vector and the event type Mapping relational data as sample data; Dividing the sample data into a training set of a first ratio and a verification set of a second ratio, wherein the first ratio is greater than the second ratio; The event classification model is trained by using sample data in the training set, and the accuracy of the event classification model is verified by using sample data in the verification set after training; If the accuracy is greater than the preset value, the training is completed. If the accuracy is less than or equal to the preset value, the number of sample data is increased, and then the step of dividing the sample data into the training set and the verification set is returned.

Images