WO2022267460A1

WO2022267460A1 - Event-based sentiment analysis method and apparatus, and computer device and storage medium

Info

Publication number: WO2022267460A1
Application number: PCT/CN2022/072045
Authority: WO
Inventors: 周骏红; 彭琛
Original assignee: 平安科技（深圳）有限公司
Priority date: 2021-06-25
Filing date: 2022-01-14
Publication date: 2022-12-29
Also published as: CN113434631A; CN113434631B

Abstract

Disclosed in the embodiments of the present application are an event-based sentiment analysis method and apparatus, and a computer device and a storage medium. The method comprises: if a sentiment analysis request is received, parsing the sentiment analysis request to acquire initial text; pre-processing the acquired initial text to obtain a plurality of short pieces of text to be analyzed, which comprise event keywords; inputting all of said short pieces of text into a preset argument extraction model, so as to determine a sentiment trigger word, a main body and an event of each said short piece of text; and inputting the determined sentiment trigger word, the main body and the event of each said short piece of text into a preset sentiment determination model, so as to obtain a sentiment label corresponding to the main body of each said short piece of text. By means of the present application, an argument in text can be effectively extracted, so as to accurately determine a sentiment label, thereby improving the usage experience of a user; and the present application can also be applied to scenarios such as smart government affairs, thereby promoting the construction of a smart city.

Description

Event-based sentiment analysis method, device, computer equipment and storage medium

This application claims the priority of the Chinese patent application submitted to the China Patent Office on June 25, 2021 with the application number 202110712428.9 and the title of the invention is "Event-Based Sentiment Analysis Method, Device, Computer Equipment and Storage Medium", the entire content of which Incorporated in this application by reference.

technical field

The present application relates to the field of data processing, and in particular to an event-based sentiment analysis method, device, computer equipment and storage medium.

Background technique

At present, many techniques have been developed to judge the emotional tendency of a piece of text. For example, the original unsupervised learning method extracts the emotional trigger words in the sentence and combines the syntax and grammar to judge the emotional score. However, due to the complexity of the language, the accuracy and generalization ability of this method are limited. With the introduction of Word Embedding, the information contained in the text can be represented by a word vector matrix, which makes end-to-end supervised learning possible, such as learning with the text word vector matrix as the input of the neural network and the emotional score as the output. Neural network model, the effect of this method was initially limited by the ability of word vector matrix to extract text semantics, and with the emergence of models with strong ability to extract semantic information such as ELMo, GPT and Bert in recent years, supervised learning method has become The mainstream of sentiment analysis.

The initial sentiment analysis task refers to judging the sentiment score of a piece of text, but for texts involving multiple subjects and multiple events, the emotions of different events of different subjects may be different, so the details of different subjects and different events may be different. The Granular Sentiment Analysis Task (ABSA) was proposed, and one way to realize this task is to input a piece of text and output the subject, event and corresponding sentiment score at the same time. Analysis, the accuracy rate is not high; another way of thinking is to give a good event system in advance, input text, and output the emotional score of the text subject on different events in the event system. However, the inventor realized that there are two problems in the above two ways of thinking. First, it does not reflect the difference in the way of expressing the emotions of different events. For example, the emotional trigger words of "increasing profits/increasing liabilities" are all rising, but the emotional tendency is completely different. Different; second, in the ABSA task, it is not clear enough to find out the corresponding relationship between subject, event, and emotional trigger words, so the corresponding relationship output by the above two sentiment analysis ideas is relatively sloppy. Generally speaking, the existing sentiment analysis methods are not ideal in practice and need to be optimized and improved.

Contents of the invention

Embodiments of the present application provide an event-based sentiment analysis method, device, computer equipment, and storage medium, which can effectively extract arguments in text to accurately determine sentiment tags and improve user experience.

In the first aspect, the embodiment of the present application provides an event-based sentiment analysis method, which includes:

If a sentiment analysis request is received, parse the sentiment analysis request to obtain the initial text;

Preprocessing the acquired initial text to obtain a plurality of short texts to be analyzed including event keywords, wherein different event keywords are associated with corresponding events;

Input all the short texts to be analyzed into the preset argument extraction model to determine the emotional trigger words, subjects and events of each short text to be analyzed, wherein different events are associated with different emotional trigger words;

The determined emotional trigger words, subjects and events of each short text to be analyzed are input into a preset emotion determination model to obtain an emotional label corresponding to the subject of each short text to be analyzed.

In the second aspect, the embodiment of the present application also provides an event-based sentiment analysis device, which includes:

A request parsing unit, configured to parse the sentiment analysis request to obtain the initial text if the sentiment analysis request is received;

A preprocessing unit, configured to preprocess the acquired initial text to obtain a plurality of short texts to be analyzed including event keywords, wherein different event keywords are associated with corresponding events;

The argument extraction unit is used to input all the short texts to be analyzed into the preset argument extraction model to determine the emotional trigger words, subjects and events of each short text to be analyzed, wherein different events are associated with different emotions trigger word;

The emotion determination unit is configured to input the determined emotional trigger words, subjects and events of each short text to be analyzed into a preset emotion determination model, so as to obtain an emotion label corresponding to each subject of the short text to be analyzed.

In the third aspect, the embodiment of the present application also provides a computer device, which includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

In the fourth aspect, the embodiment of the present application also provides a computer-readable storage medium, the storage medium stores a computer program, and the computer program can implement the following steps when executed by a processor:

The embodiment of the present application provides an event-based sentiment analysis method, device, computer equipment, and storage medium. The embodiment of the application determines the emotional trigger word, subject and event through the argument extraction model, which can improve the determination of different subjects. Under the corresponding emotional trigger words, the accuracy of the emotional label of the event is improved, and the user experience is improved. At the same time, the method of this embodiment can also be applied to scenarios such as smart government affairs, thereby promoting the effect of smart city construction.

Description of drawings

In order to illustrate the technical solutions of the embodiments of the present application more clearly, the drawings that need to be used in the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of the present application. Ordinary technicians can also obtain other drawings based on these drawings on the premise of not paying creative work.

FIG. 1 is a schematic flow diagram of an event-based sentiment analysis method provided in an embodiment of the present application;

Fig. 1a is a schematic diagram of an application scenario of an event-based sentiment analysis method provided by an embodiment of the present application;

FIG. 2 is a schematic flowchart of an event-based sentiment analysis method provided by another embodiment of the present application;

FIG. 3 is a schematic sub-flow diagram of an event-based sentiment analysis method provided by another embodiment of the present application;

FIG. 4 is a schematic sub-flow diagram of an event-based sentiment analysis method provided by another embodiment of the present application;

FIG. 5 is a schematic sub-flow diagram of an event-based sentiment analysis method provided by another embodiment of the present application;

FIG. 6 is a schematic flowchart of an event-based sentiment analysis method provided by another embodiment of the present application;

FIG. 7 is a schematic block diagram of an event-based sentiment analysis device provided in an embodiment of the present application;

Fig. 8 is a schematic block diagram of an event-based sentiment analysis device provided by another embodiment of the present application;

Fig. 9 is a schematic block diagram of a data acquisition unit of an event-based sentiment analysis device provided by another embodiment of the present application;

FIG. 10 is a schematic block diagram of a text segmentation unit of an event-based sentiment analysis device provided by another embodiment of the present application;

Fig. 11 is a schematic block diagram of a first training unit of an event-based sentiment analysis device provided by another embodiment of the present application;

Fig. 12 is a schematic block diagram of an event-based sentiment analysis device provided by another embodiment of the present application;

FIG. 13 is a schematic diagram of the structure and composition of a computer device provided by an embodiment of the present application.

detailed description

The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of them. Based on the embodiments in this application, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the scope of protection of this application.

It should be understood that when used in this specification and the appended claims, the terms "comprising" and "comprises" indicate the presence of described features, integers, steps, operations, elements and/or components, but do not exclude one or Presence or addition of multiple other features, integers, steps, operations, elements, components and/or collections thereof.

It should also be understood that the terminology used in the specification of this application is for the purpose of describing particular embodiments only and is not intended to limit the application. As used in this specification and the appended claims, the singular forms "a", "an" and "the" are intended to include plural referents unless the context clearly dictates otherwise.

Please refer to FIG. 1 and FIG. 1a. FIG. 1 is a schematic flowchart of an event-based sentiment analysis method provided by an embodiment of the present application, and FIG. 1a is a schematic diagram of a scenario of an event-based sentiment analysis method in an embodiment of the present application. The event-based sentiment analysis method is applied in the management server 20 . The management server 20 preprocesses the initial text according to an event-based sentiment analysis method, thereby obtaining a plurality of short texts to be analyzed including event keywords, and inputting the short texts to be analyzed into the argument extraction model 10 to determine each to-be-analyzed text Analyze the emotional trigger words, subject and event of the short text, then determine the argument extraction model of the emotional trigger word, subject and event input emotion determination model 30, thereby obtain the corresponding emotional label with the text to be analyzed, the method of the present embodiment is extremely Greatly improved the accuracy of determining sentiment labels for text. The following will introduce in detail the steps of the event-based sentiment analysis method from the perspective of the management server 20.

As shown in FIG. 1 , the steps of the event-based sentiment analysis method may specifically include steps S101-S104.

Step S101, if a sentiment analysis request is received, parse the sentiment analysis request to obtain initial text.

In this embodiment, if the management server receives the sentiment analysis request, it may parse the sentiment analysis request to obtain the initial text. The initial text may be pre-stored in the database, or crawled from an external server by a web crawler. For example, the initial text can be the public opinion information of the enterprise, and the public opinion information can include related subjects, events, and emotional trigger words. Through the analysis of different subjects, events, and emotional keywords, it is possible to accurately distinguish the information corresponding to the same emotional trigger words. different emotions.

In order to perform sentiment analysis, it is usually necessary to determine sentiment keywords and events, and the management server may define the events. For example, for the event system, the public opinion events of an enterprise are generally closely related to the business operation logic, thus forming a complete event system. Aiming at the similarity, correlation, and similarity of events in corporate public opinion, it is possible to design events related to public opinion and emotion. A three-level event system. Among them, the first level can include finance, personnel, operation, capital events, personnel, compliance, credit, and other eight aspects. Each aspect contains a series of specific events to form a second-level label. The second-level label is further detailed. After the division, a total of 110 specific events can be formed.

For the sentiment system, sentiment labels such as major negative, general negative, neutral, general positive, and major positive can be included. Further targeting specific events, we can summarize the relevant emotional trigger words, and design the mapping relationship between emotional trigger words in different directions and the emotional label system. For example, for the event of "individual stock market performance", we can summarize the emotional trigger word set ", "skyrocketing", "falling", "slumping"...}, and at the same time, the related emotional label system can be mapped as {"rising": generally positive, "slumping": major negative...}.

Step S102, preprocessing the acquired initial text to obtain a plurality of short texts to be analyzed including event keywords, wherein different event keywords are associated with corresponding events.

In this embodiment, the management server can also perform preprocessing on the acquired initial text. Since the initial text obtained by the management server can be a news release related to corporate public opinion including multiple events, the content of the news release can be very long, and often involves a large number of sentences that are not related to the target event, so it is necessary to The initial text is preprocessed accordingly, such as dividing the initial text into several short texts, and through screening, the short texts that have no redundant information related to the event can be excluded, and only the short texts that include event keywords are determined to be Analyzing short texts, that is, all determined short texts to be analyzed may include event keywords and the like.

Usually, different events can be associated with different event keywords. Specifically, the event keywords contained in each event can be summarized according to the text information in the current network news. An event can be associated with an event keyword, or it can be associated with There are multiple event keywords.

Step S103, input all the short texts to be analyzed into the preset argument extraction model to determine the emotional trigger words, subjects and events of each short text to be analyzed, wherein different events are associated with different emotional trigger words.

In this embodiment, the management server can input all the short texts to be analyzed into the preset argument extraction model, and the preset argument extraction model can combine the emotional trigger words, subjects and events in the text to be analyzed Accurate identification, that is, through the analysis of the argument extraction model, can make the relevant key information in the short text to be analyzed be determined. Of course, as an optional embodiment, through the argument extraction model, it is also possible to distinguish and determine the time, place and other elements included in the short text to be analyzed, so as to more accurately determine the emotion of the short text to be analyzed Label.

Among them, the preset argument extraction model can be a model obtained by training the neural network through training data, which can improve the efficiency and accuracy of determining the emotional trigger words, subjects and events of the text to be analyzed, and can more effectively analyze The text to be analyzed for a subject and multiple events.

Step S104, input the determined emotional trigger words, subjects and events of each short text to be analyzed into the preset emotion determination model, so as to obtain the corresponding emotion tags of each subject of the short text to be analyzed.

In this embodiment, the management server can input the short text to be analyzed after determining the emotion trigger word, the subject and the event into the emotion determination model, so as to obtain the emotion tag corresponding to the subject of each short text to be analyzed. For example, the current stock price of company A is ** yuan, and the stock price has risen sharply. The main body of the text is company A, the stock price is an event, and the rise is an emotional trigger word. It can be seen that the emotional label corresponding to the text can be generally positive. For another example, the current annual loss of company B is ** yuan, and the annual loss decreases. The main body of the text is company B, and the annual loss is an event, and the decrease is generally positive. Wherein, the preset emotion determination model may be a model obtained by training a neural network with training data, which can more accurately determine the emotion label of the text to be analyzed.

Therefore, it can be seen that through the preset argument extraction model and the preset emotion determination model, the text to be analyzed can be accurately and efficiently marked with emotional tags, which improves the user experience.

As shown in FIG. 2 , in an embodiment, the method further includes steps S201 to S204 before step S101 .

Step S201, crawling the original text through a web crawler.

Among them, in order to realize the rapid standardization of emotional tags for texts, the management server can obtain and process relevant training data to train related neural networks to obtain related models for tagging emotional tags. At the same time, the management server can extract a large amount of news related to corporate public opinion from relevant news websites through web crawlers, that is, the obtained news can be the original text to be processed.

In step S202, the acquired original text is preprocessed to obtain a plurality of short texts to be trained including event keywords, and the obtained short texts to be trained are stored in a preset database as a training set.

Among them, the management server can also preprocess the obtained original text, that is, divide the original text to obtain multiple short texts, and determine whether the short texts contain event keywords by confirming whether the short texts contain event keywords. This is the short text to be trained. In order to be able to manage and utilize the short text to be trained, the short text to be trained can be stored in a preset database as a training set for easy recall.

As shown in FIG. 3 , in an embodiment, the step S202 may include: steps S301-S302.

Step S301: Segment the original text according to a preset text segmentation function and preset event keywords to obtain a plurality of short texts to be trained including event keywords.

Among them, the preset text segmentation function can be cut_text. Through the text segmentation function and the preset event keywords, the original text can be segmented so that relevant information can be divided into the same short text as much as possible. Short texts including preset event keywords are used as short texts to be trained.

As shown in FIG. 4, in an embodiment, the step S301 may include steps S401-S403.

Step S401: Segment the original text according to a preset text segmentation function to obtain multiple original subtexts.

Specifically, the management server can segment the original text according to a preset text segmentation function, so as to obtain multiple shorter original subtexts. Since the original text may be longer and contain more useless information, it is necessary to distinguish and identify the original sub-text.

Step S402, judging whether the original subtext includes preset event keywords.

Wherein, the management server can judge whether the original subtext includes preset event keywords, that is, to realize screening and determination of the original subtext.

Step S403, if the original subtext includes preset event keywords, determine the original subtext as the short text to be trained.

Wherein, the management server can determine the original subtext as the short text to be trained when the original subtext includes preset event keywords, so as to realize the classification and segmentation of the original text.

Furthermore, the management server can also determine the event described in the short text to be trained according to the event keyword, so as to facilitate subsequent processing.

In a further embodiment, the step S301 also includes the following steps:

Step S404, if the original subtext does not include preset event keywords, delete the original subtext.

Wherein, if the original subtext does not include preset event keywords, the original subtext can be deleted, so as to determine more accurate training data, that is, determine a more reasonable short text to be trained.

Step S302, storing the obtained short text to be trained as a training set in a preset database.

Wherein, in order to facilitate searching and utilization, the management server may store the obtained short text to be trained as a training set in a preset database.

Step S203, if a model training instruction is received, call the short text to be trained from the preset database for labeling, so as to determine the emotional trigger words, subjects and events included in each text to be trained.

Among them, if the management server receives the model training instruction sent by the user, it can call the short text to be trained from the preset database for labeling, so as to determine the emotional trigger words, subjects and events included in each text to be trained. The above labeling can be performed manually, or automatically according to relevant labeling instructions, which is not specifically limited in this embodiment.

Step S204, train the preset first neural network through the marked text to be trained to obtain an argument extraction model.

Wherein, the management server can obtain the marked text to be trained, and use the marked text to be trained to realize the training of the first neural network, thereby obtaining a trained argument extraction model.

As shown in FIG. 5 , in an embodiment, the step S204 may include steps S501 - S502 .

Step S501, using Bert coding to obtain the vector of the marked text to be trained.

Wherein, the management server can also use Bert encoding to obtain the vector of the marked text to be trained, so as to perform subsequent training steps.

Step S502 , input the vector of the marked text to be trained and the emotional trigger words, subjects and events included in the marked text to be trained into the preset first neural network for training to obtain an argument extraction model.

Wherein, the management server can input the vector of the marked text to be trained and the emotional trigger words, subjects and events included in the marked text to be trained into the preset first neural network to train the first neural network, An argument extraction model can be obtained by training the first neural network.

As shown in FIG. 6 , if each event is mapped with a corresponding emotion tag through an associated emotion trigger word in advance, the method further includes steps S601 - S603 before step S101 .

Step S601, obtaining the emotional trigger words, subjects and events included in the tagged text to be trained.

Wherein, the management server can obtain the emotional trigger words, subjects and events included in the tagged text to be trained.

Step S602, determine the emotion label of the text to be trained according to the acquired emotion trigger words and events.

Wherein, each event is pre-associated with emotional trigger words, and for different events, different emotional tags mapped to different emotional trigger words are different. Therefore, the management server can determine the emotion tags of the text to be trained according to the acquired emotion trigger words and events.

Step S603, train the second neural network by using the determined emotional labels and the labeled text to be trained to obtain an emotion determination model.

Wherein, the management server can realize the training of the second neural network through the determined emotion label and the labeled text to be trained, so as to obtain the emotion determination model.

To sum up, the embodiment of the present application can effectively extract the arguments in the text to accurately determine the emotional tags, improve the user experience, and can also be applied to scenarios such as smart government affairs, thereby promoting the construction of smart cities.

Those skilled in the art can understand that all or part of the processes in the methods of the above embodiments can be implemented through computer programs to instruct related hardware, and the programs can be stored in a computer-readable storage medium. , may include the flow of the embodiments of the above-mentioned methods. Wherein, the storage medium may be a magnetic disk, an optical disk, a read-only memory (Read-Only Memory, ROM), etc.

Please refer to Fig. 7, corresponding to the above-mentioned event-based sentiment analysis method, the embodiment of the present application also proposes an event-based sentiment analysis device, the device 100 includes: a request parsing unit 101, a preprocessing unit 102, and an argument extraction unit 103 and an emotion determining unit 104.

The request parsing unit 101 is configured to, if a sentiment analysis request is received, parse the sentiment analysis request to obtain an initial text.

In this embodiment, if the management server receives the sentiment analysis request, it may parse the sentiment analysis request to obtain the initial text. The initial text may be pre-stored in the database, or crawled from an external server by a web crawler.

The preprocessing unit 102 is configured to preprocess the acquired initial text to obtain a plurality of short texts to be analyzed including event keywords, wherein different event keywords are associated with corresponding events.

The argument extraction unit 103 is used to input all short texts to be analyzed into a preset argument extraction model to determine the emotional trigger words, subjects and events of each short text to be analyzed, wherein different events have different Emotional triggers.

The emotion determination unit 104 is configured to input the determined emotion trigger words, subjects and events of each short text to be analyzed into a preset emotion determination model, so as to obtain an emotion tag corresponding to the subject of each short text to be analyzed.

As shown in FIG. 8 , in an embodiment, before the request parsing unit 101 , the device 100 further includes a text crawling unit 201 , a data acquisition unit 202 , a data labeling unit 203 and a first training unit 204 .

Wherein, the text crawling unit 201 is configured to crawl the original text through a web crawler.

The data acquisition unit 202 is configured to preprocess the acquired original text to obtain a plurality of short texts to be trained including event keywords, and store the obtained short texts to be trained as a training set in a preset database.

As shown in FIG. 9 , in an embodiment, the data acquisition unit 202 may include: a text segmentation unit 301 and a text storage unit 302 .

The text segmentation unit 301 is configured to segment the original text according to a preset text segmentation function and preset event keywords, so as to obtain a plurality of short texts to be trained including event keywords.

As shown in FIG. 10 , in an embodiment, the text segmentation unit 301 may include a text processing unit 401 , a text judgment unit 402 and a text determination unit 403 .

The text processing unit 401 is configured to segment the original text according to a preset text segmentation function to obtain multiple original subtexts.

A text judging unit 402, configured to judge whether the original subtext includes preset event keywords.

A text determining unit 403, configured to determine the original subtext as the short text to be trained if the original subtext includes preset event keywords.

In a further embodiment, the text segmentation unit 301 also includes the following units:

A text deletion unit 404, configured to delete the original subtext if the original subtext does not include preset event keywords.

Wherein, if the original subtext does not include preset event keywords, the original subtext can be deleted to determine more accurate training data, that is, to determine a more reasonable short text to be trained.

The text storage unit 302 is configured to store the obtained short text to be trained as a training set in a preset database.

The data labeling unit 203 is configured to call the short text to be trained from the preset database for labeling if a model training instruction is received, so as to determine the emotional trigger words, subjects and events included in each text to be trained.

The first training unit 204 is configured to train a preset first neural network through the marked text to be trained to obtain an argument extraction model.

As shown in FIG. 11 , in an embodiment, the first training unit 204 may include a vector determination unit 501 and a first model training unit 502 .

The vector determining unit 501 is used to obtain the vector of the marked text to be trained by using Bert coding.

The first model training unit 502 is used to input the vector of the marked text to be trained and the emotional trigger words, subjects and events included in the marked text to be trained into the preset first neural network for training to obtain argument extraction Model.

As shown in Figure 12, if each event is mapped with a corresponding emotion tag through the associated emotional trigger word in advance, before the request parsing unit 101 of the device 100, it also includes a feature acquisition unit 601, a tag determination unit 602, a second Two training units 603.

The feature acquisition unit 601 is configured to acquire the emotional trigger words, subjects and events included in the tagged text to be trained.

A tag determining unit 602, configured to determine an emotional tag related to the text to be trained according to the acquired emotional trigger words and events.

Wherein, each event is pre-associated with emotional trigger words, and for different events, different emotional tags mapped to different emotional trigger words are different. Therefore, the management server can determine the emotional label of the text to be trained according to the acquired emotional trigger words and events.

The second training unit 603 is configured to train the second neural network by using the determined emotional label and the labeled text to be trained to obtain an emotion determination model.

It should be noted that those skilled in the art can clearly understand that the specific implementation process of the above-mentioned event-based sentiment analysis device 100 and each unit can refer to the corresponding descriptions in the foregoing method embodiments. For the convenience and brevity of description, I won't repeat them here.

It can be seen from the above that in terms of hardware implementation, the above request parsing unit 101, preprocessing unit 102, argument extraction unit 103, and sentiment determination unit 104 can be embedded in or independent of the event-based sentiment analysis device in the form of hardware. It can be stored in the memory of the event-based sentiment analysis device in the form of software, so that the processor can invoke and execute the operations corresponding to the above units. The processor may be a central processing unit (CPU), a microprocessor, a single-chip microcomputer, and the like.

The above event-based sentiment analysis apparatus can be implemented in the form of a computer program, and the computer program can be run on the computer device as shown in FIG. 13 .

FIG. 13 is a schematic diagram of the structural composition of a computer device of the present application. The device may be a server, where the server may be an independent server or a server cluster composed of multiple servers.

Referring to FIG. 13 , the computer device 700 includes a processor 702 connected through a system bus 701 , a memory, an internal memory 704 and a network interface 705 , wherein the memory may include a non-volatile storage medium 703 and an internal memory 704 .

The non-volatile storage medium 703 can store an operating system 7031 and a computer program 7032. When the computer program 7032 is executed, the processor 702 can execute an event-based sentiment analysis method.

The processor 702 is used to provide calculation and control capabilities and support the operation of the entire computer device 700 .

The internal memory 704 provides an environment for running the computer program 7032 in the non-volatile storage medium 703. When the computer program 7032 is executed by the processor 702, the processor 702 can execute an event-based sentiment analysis method.

The network interface 705 is used for network communication with other devices. Those skilled in the art can understand that the structure shown in FIG. 13 is only a block diagram of a partial structure related to the solution of this application, and does not constitute a limitation on the computer device 700 on which the solution of this application is applied. The specific computer device 700 may include more or fewer components than shown, or combine certain components, or have a different arrangement of components.

Wherein, the processor 702 is configured to execute the computer program 7032 stored in the memory, so as to realize the steps in the event-based sentiment analysis method described above.

It should be understood that in the embodiment of the present application, the processor 602 may be a central processing unit (Central Processing Unit, CPU), and the processor 602 may also be other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), Application Specific Integrated Circuit (ASIC), off-the-shelf programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Wherein, the general-purpose processor may be a microprocessor or the processor may be any conventional processor or the like.

Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware through computer programs. The computer program can be stored in a storage medium, which is a computer-readable storage medium. The computer program is executed by at least one processor in the computer system to implement the process steps of the above method embodiments.

Therefore, the present application also provides a storage medium. The storage medium may be a computer-readable storage medium, and the computer-readable storage medium may be non-volatile or volatile. The storage medium stores a computer program, and when the computer program is executed by the processor, the processor executes the steps in the event-based sentiment analysis method described above.

The storage medium is a physical, non-transitory storage medium, such as a U disk, a mobile hard disk, a read-only memory (Read-Only Memory, ROM), a magnetic disk or an optical disk, and other physical storages that can store program codes. medium,.

Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the relationship between hardware and software Interchangeability. In the above description, the composition and steps of each example have been generally described according to their functions. Whether these functions are executed by hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art may use different methods to implement the described functions for each specific application, but such implementation should not be regarded as exceeding the scope of the present application.

In the several embodiments provided in this application, it should be understood that the disclosed devices and methods may be implemented in other ways. For example, the device embodiments described above are illustrative only. For example, the division of each unit is only a logical function division, and there may be another division method in actual implementation. For example, several units or components may be combined or integrated into another system, or some features may be omitted, or not implemented.

The steps in the methods of the embodiments of the present application can be adjusted, combined and deleted according to actual needs. Units in the device in the embodiment of the present application may be combined, divided and deleted according to actual needs. In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, each unit may exist separately physically, or two or more units may be integrated into one unit.

If the integrated unit is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of software products, and the computer software products are stored in a storage medium In the above, several instructions are included to make a computer device (which may be a personal computer, a terminal, or a network device, etc.) execute all or part of the steps of the methods described in the various embodiments of the present application.

The above is only a specific embodiment of the application, but the scope of protection of the application is not limited thereto. Any person familiar with the technical field can easily think of various equivalents within the scope of the technology disclosed in the application. Modifications or replacements, these modifications or replacements shall be covered within the scope of protection of this application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

An event-based sentiment analysis method comprising:

If a sentiment analysis request is received, parse the sentiment analysis request to obtain the initial text;

Preprocessing the acquired initial text to obtain a plurality of short texts to be analyzed including event keywords, wherein different event keywords are associated with corresponding events;

Input all the short texts to be analyzed into the preset argument extraction model to determine the emotional trigger words, subjects and events of each short text to be analyzed, wherein different events are associated with different emotional trigger words;

The determined emotional trigger words, subjects and events of each short text to be analyzed are input into a preset emotion determination model to obtain an emotional label corresponding to the subject of each short text to be analyzed.
The method according to claim 1, wherein, if the sentiment analysis request is received, before the step of parsing the sentiment analysis request to obtain the initial text, further comprising:

Crawl the original text through a web crawler;

Preprocessing the acquired original text to obtain a plurality of short texts to be trained including event keywords, and storing the obtained short texts to be trained as a training set in a preset database;

If the model training instruction is received, the short text to be trained is called from the preset database for labeling, so as to determine the emotional trigger words, subjects and events included in each text to be trained;

The preset first neural network is trained through the labeled text to be trained to obtain an argument extraction model.
The method according to claim 2, wherein the obtained original text is preprocessed to obtain a plurality of short texts to be trained including event keywords, and the obtained short texts to be trained are stored as a training set Steps to the preset database, including:

Segment the original text according to a preset text segmentation function and preset event keywords to obtain a plurality of short texts to be trained including event keywords;

Store the obtained short text to be trained as a training set in a preset database.
The method according to claim 3, wherein said original text is segmented according to a preset text segmentation function and preset event keywords to obtain a plurality of short texts to be trained including event keywords ,include:

Segmenting the original text according to a preset text segmentation function to obtain multiple original subtexts;

judging whether the original subtext includes preset event keywords;

If the original subtext includes preset event keywords, the original subtext is determined as the short text to be trained.
The method of claim 4, wherein the method further comprises:

If the original subtext does not include preset event keywords, the original subtext is deleted.
The method according to claim 2, wherein the step of training the preset first neural network through the labeled text to be trained to obtain the argument extraction model includes:

Use Bert encoding to obtain the vector of the labeled text to be trained;

The vector of the marked text to be trained and the emotional trigger words, subjects and events included in the marked text to be trained are input into the preset first neural network for training to obtain an argument extraction model.
The method according to claim 2, wherein, if each event is mapped with a corresponding emotional label by the associated emotional trigger word in advance, the method also includes:

Obtain the emotional trigger words, subjects and events included in the tagged text to be trained;

Determine the emotional label of the text to be trained according to the acquired emotional trigger words and events;

The second neural network is trained by using the determined emotional labels and the labeled text to be trained to obtain an emotional determination model.
An event-based sentiment analysis device, comprising:

A request parsing unit, configured to parse the sentiment analysis request to obtain the initial text if the sentiment analysis request is received;

A preprocessing unit, configured to preprocess the acquired initial text to obtain a plurality of short texts to be analyzed including event keywords, wherein different event keywords are associated with corresponding events;

The argument extraction unit is used to input all the short texts to be analyzed into the preset argument extraction model to determine the emotional trigger words, subjects and events of each short text to be analyzed, wherein different events are associated with different emotions trigger word;

The emotion determination unit is configured to input the determined emotional trigger words, subjects and events of each short text to be analyzed into a preset emotion determination model, so as to obtain an emotion label corresponding to each subject of the short text to be analyzed.
A computer device, wherein the computer device includes a memory and a processor, and a computer program is stored on the memory, and the processor implements the following steps when executing the computer program:

If a sentiment analysis request is received, parse the sentiment analysis request to obtain the initial text;

Preprocessing the acquired initial text to obtain a plurality of short texts to be analyzed including event keywords, wherein different event keywords are associated with corresponding events;

Input all the short texts to be analyzed into the preset argument extraction model to determine the emotional trigger words, subjects and events of each short text to be analyzed, wherein different events are associated with different emotional trigger words;

The determined emotional trigger words, subjects and events of each short text to be analyzed are input into a preset emotion determination model to obtain an emotional label corresponding to the subject of each short text to be analyzed.
The computer device according to claim 9, wherein, before the step of parsing the sentiment analysis request to obtain the initial text if the sentiment analysis request is received, further comprising:

Crawl the original text through a web crawler;

Preprocessing the acquired original text to obtain a plurality of short texts to be trained including event keywords, and storing the obtained short texts to be trained as a training set in a preset database;

If the model training instruction is received, the short text to be trained is called from the preset database for labeling, so as to determine the emotional trigger words, subjects and events included in each text to be trained;

The preset first neural network is trained through the labeled text to be trained to obtain an argument extraction model.
The computer device according to claim 10, wherein said preprocessing the acquired original text to obtain a plurality of short texts to be trained including event keywords, and using the obtained short texts to be trained as a training set The steps to save to the preset database include:

Segment the original text according to a preset text segmentation function and preset event keywords to obtain a plurality of short texts to be trained including event keywords;

Store the obtained short text to be trained as a training set in a preset database.
The computer device according to claim 11, wherein said original text is segmented according to a preset text segmentation function and preset event keywords to obtain a plurality of short texts to be trained including event keywords steps, including:

Segmenting the original text according to a preset text segmentation function to obtain multiple original subtexts;

judging whether the original subtext includes preset event keywords;

If the original subtext includes preset event keywords, the original subtext is determined as the short text to be trained.
The computer device of claim 12, wherein the method further comprises:

If the original subtext does not include preset event keywords, the original subtext is deleted.
The computer device according to claim 10, wherein the step of training the preset first neural network through the labeled text to be trained to obtain the argument extraction model includes:

Use Bert encoding to obtain the vector of the labeled text to be trained;

The vector of the marked text to be trained and the emotional trigger words, subjects and events included in the marked text to be trained are input into the preset first neural network for training to obtain an argument extraction model.
The computer device according to claim 10, wherein, if each event is mapped with a corresponding emotional tag through the associated emotional trigger word in advance, the processor further implements:

Obtain the emotional trigger words, subjects and events included in the tagged text to be trained;

Determine the emotional label of the text to be trained according to the acquired emotional trigger words and events;

The second neural network is trained by using the determined emotional labels and the labeled text to be trained to obtain an emotional determination model.
A computer-readable storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the processor performs the following steps:

If a sentiment analysis request is received, parse the sentiment analysis request to obtain the initial text;

Preprocessing the acquired initial text to obtain a plurality of short texts to be analyzed including event keywords, wherein different event keywords are associated with corresponding events;

Input all the short texts to be analyzed into the preset argument extraction model to determine the emotional trigger words, subjects and events of each short text to be analyzed, wherein different events are associated with different emotional trigger words;

The determined emotional trigger words, subjects and events of each short text to be analyzed are input into a preset emotion determination model to obtain an emotional label corresponding to the subject of each short text to be analyzed.
The computer-readable storage medium according to claim 16, wherein, before the step of parsing the sentiment analysis request to obtain the initial text if the sentiment analysis request is received, further comprising:

Crawl the original text through a web crawler;

Preprocessing the acquired original text to obtain a plurality of short texts to be trained including event keywords, and storing the obtained short texts to be trained as a training set in a preset database;

If the model training instruction is received, the short text to be trained is called from the preset database for labeling, so as to determine the emotional trigger words, subjects and events included in each text to be trained;

The preset first neural network is trained through the labeled text to be trained to obtain an argument extraction model.
The computer-readable storage medium according to claim 17, wherein the preprocessing is performed on the acquired original text to obtain a plurality of short texts to be trained including event keywords, and the obtained short texts to be trained are The steps to store as a training set into a preset database include:

Segment the original text according to a preset text segmentation function and preset event keywords to obtain a plurality of short texts to be trained including event keywords;

Store the obtained short text to be trained as a training set in a preset database.
The computer-readable storage medium according to claim 18, wherein the original text is segmented according to a preset text segmentation function and preset event keywords to obtain a plurality of texts to be trained including event keywords Steps for a short text, including:

Segmenting the original text according to a preset text segmentation function to obtain multiple original subtexts;

judging whether the original subtext includes preset event keywords;

If the original subtext includes preset event keywords, the original subtext is determined as the short text to be trained.
The computer-readable storage medium of claim 19, wherein the processor further performs the steps of:

If the original subtext does not include preset event keywords, the original subtext is deleted.