GPT-based directional label public opinion analysis method and corresponding device

Abstract

The invention relates to a directional label public opinion analysis method and device based on GPT, and the method comprises the steps: obtaining first public opinion data of a target user for a data label in response to the data label configured for the target user; based on the GPT and the first public opinion data, obtaining a public opinion theme and an emotion category for the first public opinion data; and according to the public opinion theme, the emotion category and the user information of the target user, generating a public opinion analysis result. Therefore, according to the data label set by the user, the event in the target field can be monitored in a targeted manner, that is, the first public opinion data is acquired in a targeted manner and oriented public opinion analysis is performed, on one hand, the processing amount of the first public opinion data can be reduced, and the public opinion analysis efficiency can be improved. And on the other hand, the public opinion analysis result does not contain irrelevant data, so that the accuracy of the public opinion analysis result is improved, and the interference of the irrelevant data on the user is avoided.

Description

Technical Field

[0001] The present disclosure relates to the field of data processing technology, and in particular to a GPT-based directional label public opinion analysis method and a corresponding device. Background Art

[0002] Public opinion analysis refers to the collection and analysis of public data from social networks, news platforms, forums, etc. to identify the public's attitudes and emotions towards a topic, field or event. However, when conducting public opinion analysis, related technologies generally use keyword filtering technology to identify and extract keywords or key phrases from the collected public opinion data, and then generate public opinion reports or public opinion analysis results based on the extracted keywords or key phrases, which may result in inaccurate public opinion analysis results. Summary of the invention

[0003] The purpose of the present invention is to provide a GPT-based directional label public opinion analysis method and corresponding device to solve the above technical problems.

[0004] In order to achieve the above-mentioned object, the present disclosure provides a first aspect of a directional tag public opinion analysis method based on GPT, the method comprising: In response to a data tag configured for a target user, first public opinion data of the target user for the data tag is obtained, wherein the target user is a domain expert in a target domain, the data tag is a data tag used to identify a data category in the target domain, and the target user is determined based on a user influence value and a domain influence value of the user in the target domain, the user influence value is used to characterize the degree of influence of the user on other users, and the domain influence value is used to characterize the degree of influence of the user on the target domain; Based on the GPT and the first public opinion data, obtaining a public opinion theme and a sentiment category for the first public opinion data, wherein the GPT is used to output a corresponding public opinion theme and sentiment category according to the input first public opinion data; Generate a public opinion analysis result based on the public opinion topic, the emotion category and the user information of the target user.

[0005] Optionally, the target user includes multiple users, and correspondingly, the public opinion topics and the emotion categories include multiple users, and generating the public opinion analysis result according to the public opinion topics, the emotion categories and the user information of the target user includes: Taking the public opinion topic as a classification dimension, classifying the sentiment category corresponding to the first public opinion data into the public opinion topic corresponding to the first public opinion data, and obtaining the sentiment category under each public opinion topic; For each of the public opinion topics, the sentiment categories under the public opinion topic are aggregated to obtain the sentiment distribution information of the public opinion topic; A public opinion analysis result is generated based on the sentiment distribution information of each public opinion topic, multiple public opinion topics, and user information of multiple target users.

[0006] Optionally, the target user is obtained in the following manner: Obtaining second public opinion data of users in the target network platform for preset data tags through an application program interface of the target network platform, wherein the preset data tags are data tags in the target field; Determining, according to the second public opinion data, a public opinion influence value of a user corresponding to the second public opinion data in the target field; The target user is determined according to the public opinion influence value of the user corresponding to the second public opinion data in the target field.

[0007] Optionally, the second public opinion data includes content data published by the user in the target network platform, user data of the user in the target network platform, and interaction data of the user in the target network platform, and determining, based on the second public opinion data, a public opinion influence value of the user corresponding to the second public opinion data in the target field includes: Determine, according to the user data, a first influence value of the user on other users in the target network platform, and determine the first influence value as the user influence value; Determine, based on the content data and the interaction data, a second impact value of the content published by the user in the target network platform in the target domain, and determine the second impact value as the domain impact value; The public opinion influence value of the user corresponding to the second public opinion data in the target field is determined according to the user influence value and the field influence value.

[0008] Optionally, the user data includes first user data, second user data, and third user data, the first user data is used to indicate the credibility of the content published by the user in the target network platform, the second user data includes the fan information of the user in the target network platform, and the third user data includes the follower information of the user in the target network platform, and determining the first influence value of the user on other users in the target network platform according to the user data includes: Determining a third impact value based on a correspondence between a preset trust score and the credibility and the first user data; Based on the fan information and the follower information, construct a social network of the user in the target network platform, and determine the degree centrality, closeness centrality, betweenness centrality and eigenvector centrality of the user in the social network according to the social network, and determine a fourth influence value according to the degree centrality, the closeness centrality, the betweenness centrality and the eigenvector centrality; Taking the target constant as a base and the number of followers in the follower information as an index, a target value is obtained, and the target value is normalized to obtain a fifth influence value; A first influence value of the user on other users in the target network platform is determined according to the third influence value, the fourth influence value, and the fifth influence value.

[0009] Optionally, the interaction data includes first interaction data, second interaction data, third interaction data and fourth interaction data, the first interaction data includes the number of likes, the second interaction data includes the number of comments, the third interaction data includes the number of reposts, and the fourth interaction data includes the number of replies, and determining the second impact value of the content published by the user in the target network platform in the target field according to the content data and the interaction data includes: Acquire hot topics in the target field and target historical content data published by the user in the target network platform, wherein the hot topics are topics with a network popularity value greater than a preset network popularity value, and the target historical content data are content data published by the user in a preset time interval; Determining the semantic similarity between the historical content data and the hot topic based on the semantic recognition model, the historical content data and the hot topic; When the semantic similarity is greater than a preset semantic similarity threshold, the sum of the likes, the comments, the forwardings, and the replies is calculated, and the ratio of the sum to the duration corresponding to the preset time interval is calculated to obtain a content interaction value; A second influence value of the content published by the user in the target network platform in the target field is determined according to the semantic similarity and the content interaction value.

[0010] A second aspect of the present disclosure provides a directional tag public opinion analysis device based on GPT, the device comprising: A first acquisition module is used to obtain, in response to a data tag configured for a target user, first public opinion data of the target user for the data tag, wherein the target user is a domain expert in a target domain, the data tag is a data tag used to identify a data category in the target domain, and the target user is determined based on a user influence value and a domain influence value of the user in the target domain, the user influence value is used to characterize the degree of influence of the user on other users, and the domain influence value is used to characterize the degree of influence of the user on the target domain; A first processing module, configured to obtain a public opinion theme and a sentiment category for the first public opinion data based on the GPT and the first public opinion data, wherein the GPT is configured to output a corresponding public opinion theme and sentiment category according to the input first public opinion data; The second processing module is used to generate a public opinion analysis result according to the public opinion topic, the emotion category and the user information of the target user.

[0011] In a third aspect, the present disclosure provides a computer-readable medium having a computer program stored thereon, which, when executed by a processing device, implements the steps of any one of the methods described in the first aspect.

[0012] In a fourth aspect, the present disclosure provides an electronic device, including: a storage device having a computer program stored thereon; A processing device is used to execute the computer program in the storage device to implement the steps of any one of the methods in the first aspect.

[0013] In a fifth aspect, the present disclosure provides a computer program product, comprising a computer program, which implements the steps of any one of the methods described in the first aspect when executed by a processor.

[0014] Through the above technical solution, data tags can be configured for target users, so that the first public opinion data of the target users under the data tags can be obtained in a targeted manner, and then the corresponding public opinion analysis results can be generated according to the obtained first public opinion data. Therefore, events in the target field can be monitored in a targeted manner according to the data tags set by the user, that is, the first public opinion data can be obtained in a targeted manner and targeted public opinion analysis can be performed. On the one hand, the processing volume of the first public opinion data can be reduced and the efficiency of public opinion analysis can be improved. On the other hand, since the first public opinion data is the public opinion data under the data tag, when the public opinion analysis results are generated based on the first public opinion data, irrelevant data can be excluded from the public opinion analysis results, thereby improving the accuracy of the public opinion analysis results and avoiding interference of irrelevant data to users.

[0015] Other features and advantages of the present disclosure will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The accompanying drawings are used to provide a further understanding of the present disclosure and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the present disclosure but do not constitute a limitation of the present disclosure. In the accompanying drawings: Figure 1 It is a flow chart of a method for directional tag public opinion analysis based on GPT according to an exemplary embodiment of the present disclosure; Figure 2 is a flowchart of another method for directional tag public opinion analysis based on GPT according to an exemplary embodiment of the present disclosure; Figure 3 This is a schematic diagram showing part of the content of public opinion analysis results in a visual manner according to an exemplary embodiment of the present disclosure; Figure 4 This is a schematic diagram showing part of the content of public opinion analysis results in a visual manner according to an exemplary embodiment of the present disclosure; Figure 5 This is a schematic diagram showing part of the content of public opinion analysis results in a visual manner according to an exemplary embodiment of the present disclosure; Figure 6 It is a structural block diagram of a directional tag public opinion analysis device based on GPT according to an exemplary embodiment of the present disclosure; Figure 7 is a block diagram of an electronic device according to an exemplary embodiment; Figure 8 is a block diagram of another electronic device according to an exemplary embodiment. DETAILED DESCRIPTION

[0017] The specific implementation of the present disclosure is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described herein is only used to illustrate and explain the present disclosure, and is not used to limit the present disclosure.

[0018] As mentioned in the background technology, when conducting public opinion analysis, the relevant technology generally identifies and extracts keywords or key phrases from the collected public opinion data based on keyword filtering technology, and then generates a public opinion report or public opinion analysis results based on the extracted keywords or key phrases. However, since this public opinion analysis method is a general analysis method, it is impossible to distinguish the specific needs of different fields and cannot filter information in a targeted manner. On the one hand, the public opinion analysis results may contain a large amount of data that is irrelevant to the topics that users are concerned about, making it difficult for users to quickly find valuable information from the public opinion analysis results. On the other hand, for public opinion data in specific fields or labels, this general method may not be able to effectively identify it, resulting in the omission of important information, which in turn makes the public opinion analysis results unable to accurately reflect the attitudes and / or emotions of the public in this field towards a certain topic, field or event.

[0019] In view of this, the embodiments of the present disclosure provide a GPT-based directional label public opinion analysis method and a corresponding device to solve the above technical problems.

[0020] The embodiments of the present disclosure are further explained below with reference to the accompanying drawings.

[0021] Figure 1 and Figure 2 is a flowchart of a method for directional tag public opinion analysis based on GPT according to an exemplary embodiment of the present disclosure, referring to Figure 1 and Figure 2 , the public opinion analysis method may include the following steps: S101: In response to a data tag configured for a target user, first public opinion data of the target user for the data tag is obtained, wherein the target user is a domain expert in a target domain, the data tag is a data tag for identifying a data category in the target domain, and the target user is determined based on a user influence value and a domain influence value of the user in the target domain, the user influence value is used to characterize the degree of influence of the user on other users, and the domain influence value is used to characterize the degree of influence of the user on the target domain.

[0022] In this embodiment, the target user may be one or more than one. When there are multiple target users, the same data tag may be configured for the multiple target users to find out the areas of consensus or disagreement among the multiple target users. Different data tags may also be configured for the multiple target users to find out the focus of each target user.

[0023] In this embodiment, the data label and the first public opinion data can be set according to the actual situation, and the present disclosure embodiment does not impose any restrictions on this. For example, in the field of virtual assets, the data labels may include "Bitcoin", "Blockchain Supervision" or "Cryptocurrency", etc. In the field of image processing, the data labels may include "image enhancement", "feature extraction" or "semantic segmentation", etc. For example, the first public opinion data may include the content posted by the user in the target network platform, the comment content of the user in the target network platform, the like content of the user in the target network platform, and the user information of the user in the target network platform.

[0024] In this embodiment, the first public opinion data of the target user for the data tag can be obtained by automatically collecting the content related to the data tag posted by all users in the social network through the social network API (such as the API of the information publishing platform), and then filtering the content posted by the target user from the collected content according to the user ID of the target user, thereby obtaining the first public opinion data. Of course, the first public opinion data of the target user for the data tag can also be obtained by other methods, and the present embodiment does not impose any restrictions on this.

[0025] S102: Based on the GPT and the first public opinion data, obtain the public opinion theme and emotion category for the first public opinion data, wherein the GPT is used to output the corresponding public opinion theme and emotion category according to the input first public opinion data.

[0026] It should be understood that in the related art, public opinion topics are generally generated based on probability models (such as Latent Dirichlet Allocation (LDA, LDA) and Non-negative Matrix Factorization (NMF)) and models that consider time dynamics and topic relevance (such as Dynamic Topic Model (DTM) and Correlated Topic Model (CTM)). However, when these models process short texts or texts with complex contexts, the quality of the topics generated is often poor, which affects the accuracy of the analysis results. GPT (Generative Pre-trained Transformer) can better understand the text context. Therefore, this embodiment improves the accuracy of the public opinion analysis results by inputting the first public opinion data into GPT to generate high-quality topics using TopicGPT technology.

[0027] In addition, in the related art, the sentiment of the input text is generally determined by a dictionary-based method (i.e., using a pre-defined sentiment dictionary to evaluate the sentiment tendency of the text), a machine learning-based method (i.e., predicting sentiment by training a classifier, such as a support vector machine or a random forest), or a deep learning-based method (i.e., using a recurrent neural network, a convolutional neural network, or a Transformer model to process sequence data and capture long-term dependencies in the text, thereby obtaining the sentiment tendency of the text). Although these methods have achieved certain results in processing explicit emotional expressions, they are often difficult to accurately identify when faced with complex and implicit emotional expressions, such as irony, humor, or metaphors. Therefore, this embodiment captures the complex emotional changes in the first public opinion data and identifies the implicit emotions in the first public opinion data by inputting the first public opinion data into GPT, using the THOR framework of GPT, thereby improving the accuracy of the emotional categories and further improving the accuracy of the public opinion analysis results.

[0028] Among them, it is worth explaining that the method of generating public opinion topics and sentiment categories based on GPT and the first public opinion data in this embodiment is only a schematic illustration. In a possible manner, GPT may include two, one of which is used to output the corresponding public opinion topic based on the input first public opinion data, and the other GPT is used to output the corresponding sentiment category based on the input first public opinion data. In a possible manner, the first public opinion data can also be input into BERT (Bidirectional Encoder Representations from Transformers), RoBERTa (Robustly optimized BERT approach), T5 (Text-to-Text Transfer Transformer) or XLNet (eXtreme Language Model Pre-training) which are excellent in text understanding and semantic analysis, so as to obtain the public opinion topics and sentiment categories corresponding to the first public opinion data.

[0029] S103: Generate public opinion analysis results based on the public opinion theme, sentiment category, and user information of the target user.

[0030] For example, the public opinion topic, sentiment category, and user information of the target user can be input into a large language model, such as a T5 model, to generate the corresponding public opinion analysis results. The public opinion analysis results can be determined according to actual conditions, and the embodiments of the present disclosure do not impose any restrictions on this. For example, the public opinion analysis results can be used to indicate the public opinion topics that the target user is interested in in the target field, as well as the emotional distribution of the content under the public opinion topic, etc.

[0031] After obtaining the public opinion analysis results, in order to let users intuitively understand the current public opinion situation, the public opinion analysis results can be displayed in a variety of visual ways such as charts, word clouds, bar charts, etc. Figure 3-5 At the same time, in order to enable users to timely and accurately understand the current public opinion situation, the public opinion analysis results can also be updated when the first public opinion data under the data label changes.

[0032] Through the above technical solution, data tags can be configured for target users, so that the first public opinion data of the target users under the data tags can be obtained in a targeted manner, and then the corresponding public opinion analysis results can be generated according to the obtained first public opinion data. Therefore, events in the target field can be monitored in a targeted manner according to the data tags set by the user, that is, the first public opinion data can be obtained in a targeted manner and targeted public opinion analysis can be performed. On the one hand, the processing volume of the first public opinion data can be reduced and the efficiency of public opinion analysis can be improved. On the other hand, since the first public opinion data is the public opinion data under the data tag, when the public opinion analysis results are generated based on the first public opinion data, irrelevant data can be excluded from the public opinion analysis results, thereby improving the accuracy of the public opinion analysis results and avoiding interference of irrelevant data to users.

[0033] To facilitate understanding of the GPT-based directional label public opinion analysis method provided by the present disclosure, the possible implementation methods of this solution are described below.

[0034] In a possible manner, the target user may include multiple, and accordingly, the public opinion topics and sentiment categories also include multiple. Accordingly, generating the public opinion analysis results according to the public opinion topics, sentiment categories and user information of the target user may include: Taking the public opinion topic as the classification dimension, the emotion category corresponding to the first public opinion data is divided into the public opinion topic corresponding to the first public opinion data to obtain the emotion category under each public opinion topic; for each public opinion topic, the emotion category under the public opinion topic is aggregated to obtain the emotion distribution information of the public opinion topic; according to the emotion distribution information of each public opinion topic, multiple public opinion topics and user information of multiple target users, the public opinion analysis results are generated.

[0035] For example, public opinion topics include topic 1, topic 2, and topic 3, and emotion categories include emotion 1, emotion 2, emotion 3, emotion 4, and emotion 5, wherein the emotion categories for topic 1 include emotion 1, emotion 4, and emotion 5, the emotion categories for topic 2 include emotion 2 and emotion 4, and the emotion categories for topic 3 include emotion 2, emotion 3, and emotion 5. Therefore, after the emotion categories are divided into corresponding public opinion topics, by aggregating and counting the number of each emotion category in each public opinion topic, the emotion distribution information under each public opinion topic can be obtained, that is, the number or proportion of each emotion category. For example, the emotion distribution information can be expressed as: Theme 1: Emotion 1 (30%), Emotion 4 (40%), Emotion 5 (30%); Theme 2: Emotion 2 (60%), Emotion 4 (40%); Theme 3: Emotion 2 (20%), Emotion 3 (50%), Emotion 5 (30%).

[0036] Afterwards, the sentiment distribution information of multiple public opinion topics and the user information of the target user can be combined to generate a comprehensive public opinion analysis result. For example, the generated public opinion analysis result may include: the overall sentiment tendency of each public opinion topic, the distribution of each sentiment category in different topics, and / or the prediction of public opinion hot spots and trends.

[0037] In this embodiment, by subject classification and sentiment analysis of public opinion data, specific topics or areas of concern to users can be more accurately located, thereby improving the public opinion analysis results. In addition, by aggregating sentiment categories and generating sentiment distribution information, it is helpful to deeply understand the public's emotional attitudes towards specific topics and enrich the content of public opinion analysis results.

[0038] It should be understood that the content in the above public opinion analysis results is only for illustrative purposes and does not constitute a limitation on the solution. In a possible manner, it is also possible to obtain implicit content and / or opinions for the first public opinion data based on GPT and the first public opinion data, so as to generate public opinion analysis results, and it is also possible to combine implicit content and / or opinions to generate more accurate public opinion analysis results.

[0039] Among the possible ways, the target users can be obtained through: Through the application program interface of the target network platform, the second public opinion data of the users in the target network platform for the preset data tags are obtained, wherein the preset data tags are data tags in the target field; based on the second public opinion data, the public opinion influence value of the user corresponding to the second public opinion data in the target field is determined; based on the public opinion influence value of the user corresponding to the second public opinion data in the target field, the target user is determined.

[0040] In this embodiment, the target network platform can be a social networking platform, a news platform, a video sharing platform, etc., and of course it can also be other network platforms, and the present disclosed embodiment does not impose any restrictions on this. Among them, the target network platform can be one or more. When there is one target network platform, the second public opinion data corresponding to each user in the target network platform can be obtained through the API of the target network platform, and then the second public opinion data corresponding to each user is analyzed to obtain the public opinion influence value of each user. When there are multiple target network platforms, users who have accounts in multiple target network platforms can be screened out first, and then for each screened user, the second public opinion data of the user in the target network platform can be obtained through the application program interface API of the target network platform, and finally the second public opinion data of the user in each target network platform can be analyzed to obtain the public opinion influence value of the user.

[0041] In this embodiment, according to the public opinion influence value of the user corresponding to the second public opinion data in the target field, the target user can be determined by screening the target user according to the size of the public opinion influence value of each user. For example, the user with the largest public opinion influence value is used as the target user, or the first N users with the largest public opinion influence value, where N is a positive integer. It is also possible to pre-set a public opinion influence threshold and use the user corresponding to the public opinion influence value greater than or equal to the public opinion influence threshold as the target user.

[0042] In this embodiment, the second public opinion data can be determined according to actual conditions, and the present disclosure embodiment does not impose any restrictions on this. For example, the second public opinion data may include content posted by users on the target network platform, comments by users on the target network platform, content liked by users on the target network platform, and user information of users on the target network platform.

[0043] In a possible manner, the second public opinion data may include content data published by the user in the target network platform, user data of the user in the target network platform, and interaction data of the user in the target network platform. Accordingly, according to the second public opinion data, determining the public opinion influence value of the user corresponding to the second public opinion data in the target field may include: Based on user data, determine the first influence value of the user on other users in the target network platform, and determine the first influence value as the user influence value; based on content data and interaction data, determine the second influence value of the content posted by the user in the target network platform in the target field, and determine the second influence value as the field influence value; based on the user influence value and the field influence value, determine the public opinion influence value of the user corresponding to the second public opinion data in the target field.

[0044] In this embodiment, content data, user data, and interactive data can all be determined according to actual conditions, and the embodiments of the present disclosure do not impose any restrictions on this. For example, content data may include content published by users on the target network platform, the time, frequency, and quantity of content published by users on the target network platform, etc. User data may include the number of fans of the user, user status (e.g., unauthenticated user, authenticated user, active user, official user), and the number of followers, etc. Interactive data may include like data, forwarding data, sharing data, or access data, etc.

[0045] In this embodiment, the first influence value of a user on other users in the target network platform can be determined based on the user's credibility, network centrality, and number of followers. Among them, the user's credibility can be determined according to the user's status. For example, the user's credibility of an unauthenticated user is lower than that of an active user, and the credibility of an active user is lower than that of an official user. Therefore, the user's credibility can be determined according to the user's status. For example, the credibility of an unauthenticated user is set to 0.8, the credibility of an active user is set to 0.9, and the credibility of an official user is set to 1.0. Therefore, after obtaining the user data, the user's credibility can be obtained according to the user's status.

[0046] Network centrality is an indicator for judging the importance / influence of nodes in a network. It can include degree centrality, closeness centrality, betweenness centrality and eigenvector centrality. Among them, degree centrality refers to the number of connections of a node in the network, that is, the degree of the node. The higher the degree centrality, the more connections the node has in the network and the greater its influence on the network. Closeness centrality refers to the distance between a node and other nodes, that is, the reciprocal of the shortest path length to other nodes. The higher the closeness centrality, the closer the connection between the node and other nodes and the faster the information transmission speed. Betweenness centrality refers to the degree of mediation of a node in the network, that is, the number of times a node appears on the shortest path between other nodes. A high betweenness centrality indicates that the node plays an important role in the information transmission and connection of the network. Eigenvector centrality refers to the importance of a node in the network, which is related to the importance of the node's neighboring nodes. The higher the eigenvector centrality, the more important the nodes around the node are and the greater the influence of the node. Thus, the network centrality can be obtained by constructing a social network through the user's follower information and fan information, and then by analyzing the social network to obtain degree centrality, closeness centrality, betweenness centrality and eigenvector centrality. For example, a weight can be assigned to degree centrality, closeness centrality, betweenness centrality and eigenvector centrality respectively, so that after obtaining degree centrality, closeness centrality, betweenness centrality and eigenvector centrality, the network centrality can be obtained based on degree centrality, closeness centrality, betweenness centrality, eigenvector centrality and the corresponding weights.

[0047] The number of followers is the most intuitive way to reflect user influence. However, since the number of followers varies greatly, we can obtain part of the user's influence by taking the logarithm of the followers and normalizing them.

[0048] That is to say, in a possible manner, the user data may include first user data, second user data, and third user data, the first user data is used to indicate the credibility of the content published by the user in the target network platform, the second user data includes the fan information of the user in the target network platform, and the third user data includes the follower information of the user in the target network platform. Accordingly, according to the user data, determining the first influence value of the user on other users in the target network platform may include: Based on the correspondence between the preset trust score and the credibility and the first user data, a third influence value is determined; based on the fan information and the follower information, a social network of the user in the target network platform is constructed, and based on the social network, the degree centrality, closeness centrality, betweenness centrality and eigenvector centrality of the user in the social network are determined, and based on the degree centrality, closeness centrality, betweenness centrality and eigenvector centrality, a fourth influence value is determined; using a target constant as a base and the number of followers in the follower information as an index, a target value is obtained, and the target value is normalized to obtain a fifth influence value; based on the third influence value, the fourth influence value and the fifth influence value, a first influence value of the user on other users in the target network platform is determined.

[0049] After obtaining the third influence value (i.e., user credibility), the fourth influence value (i.e., network centrality), and the fifth influence value (i.e., the number of followers) in the above manner, the first influence value of the user on other users in the target network platform is obtained by substituting the third influence value, the fourth influence value, and the fifth influence value into the following formula:

[0050] in, Indicates user The first impact value on other users in the target network platform, represents the third impact value, represents the fourth impact value, represents the fifth impact value, , and represents a coefficient, and the size of the coefficient can be determined according to actual conditions, and the embodiment of the present disclosure does not impose any restrictions on this. For example, .

[0051] It should be understood that in the relevant technologies, user influence is generally evaluated based on simple indicators such as the number of fans, the relationship between users, and interactions (such as likes, comments, and reposts). For example, user influence is evaluated based on FansRank, PageRank, and TwitterRank. Among them, FansRank evaluates influence based on the number of fans of users, believing that the more fans a user has, the greater the influence, but ignores the quality of interaction and the existence of fake fans, and easily overestimates the influence of users. PageRank evaluates the importance of users in the network by analyzing the strength of the connection between users and other users. It focuses on the structure of social networks, but ignores the quality of content and domain relevance. TwitterRank introduces topic relevance on the basis of PageRank to evaluate the influence of users in specific topics. It is more suitable for identifying key users in specific fields, but the analysis of the quality of user interaction is insufficient. Therefore, although the above algorithms have their own focuses, they simply use one or two indicators to measure the influence of users, lack comprehensive considerations, and result in insufficient comprehensive and accurate user influence. When evaluating user influence based on the method of this embodiment, not only the user's social network structure is taken into account, but also the interaction and field relevance of the content posted by the user are combined, so that a more comprehensive evaluation can be performed from multiple dimensions, thereby overcoming the limitations of single-dimensional analysis in related technologies and improving the accuracy of user influence evaluation.

[0052] In a possible manner, the interaction data may include first interaction data, second interaction data, third interaction data, and fourth interaction data, the first interaction data may include the number of likes, the second interaction data may include the number of comments, the third interaction data may include the number of reposts, and the fourth interaction data may include the number of replies. Accordingly, determining the second impact value of the content published by the user in the target network platform in the target field according to the content data and the interaction data may include: Obtain hot topics in the target field and target historical content data published by users in the target network platform, where the hot topics are topics with a network heat value greater than a preset network heat value, and the target historical content data are content data published by users in a preset time interval; determine the semantic similarity between the historical content data and the hot topics based on the semantic recognition model, the historical content data, and the hot topics; when the semantic similarity is greater than a preset semantic similarity threshold, calculate the sum of the likes, comments, reposts, and replies, and calculate the ratio of the sum to the corresponding duration of the preset time interval to obtain the content interaction value; determine the second influence value of the content published by the user in the target network platform in the target field based on the semantic similarity and the content interaction value.

[0053] In this embodiment, the preset network heat value, the preset time interval, the semantic recognition model and the semantic similarity threshold can all be determined according to actual conditions, and the embodiment of the present disclosure does not impose any restrictions on this.

[0054] For example, the preset network heat value can be set to 10,000, the preset time interval can be set to 15 days, the semantic recognition model can be set to the SBERT (Sentence-BERT) model, and the semantic similarity threshold can be set to 0.3. Therefore, when the network heat value of a topic in the target field is greater than 10,000, it can be used as a hot topic, and then all content data published by the target user in the first 15 days are obtained. For each content data, the content data and the hot topic are input into the SBERT model to obtain the semantic similarity between the content data and the hot topic. If the semantic similarity is less than 0.3, it means that the content data is not related to the target field, then the semantic similarity between the next content data and the hot network is calculated. If the semantic similarity is greater than or equal to 0.3, it means that the content data is related to the target field, then the likes, comments, forwardings and replies of the content data are obtained, and the likes, comments, forwardings and replies are substituted into the following formula to obtain the content interaction value:

[0055] in, Indicates Content interaction value of historical content data, Indicates The number of likes for the historical content data, Indicates The number of comments on the historical content data, Indicates The number of forwardings of historical content data, Indicates The number of replies to historical content data. Indicates the preset time interval.

[0056] After obtaining the content interaction value and semantic similarity of each piece of historical content data, the second influence value is obtained by substituting the content interaction value and semantic similarity of each piece of historical content data into the following formula:

[0057] in, Indicates user The second impact value, Indicates The semantic similarity between the historical content data and the hot network, Indicates user Historical content data.

[0058] After obtaining the first influence value and the second influence value, the public opinion influence value of the user in the target field is obtained by substituting the first influence value and the second influence value into the following formula:

[0059] in, Indicates user Public opinion impact value, and represents a coefficient, and the size of the coefficient can be determined according to actual conditions, and the embodiment of the present disclosure does not impose any restrictions on this. For example, .

[0060] In summary, through the above technical solutions, on the one hand, this solution can conduct public opinion analysis based on the targeted tags set by users, monitor specific fields or events in a targeted manner, and improve the accuracy of data processing and analysis. On the other hand, it can combine the user's credibility, network centrality, number of fans, interactivity of user-posted content and its relevance to the field and other dimensions for analysis, so as to more comprehensively and accurately evaluate the user's public opinion influence under specific tags, and further improve the accuracy of public opinion analysis results. In addition, this solution also performs sentiment analysis through the THOR framework of GPT, so as to capture more subtle sentiment changes and further improve the accuracy of public opinion analysis results.

[0061] Based on the same concept, the embodiment of the present disclosure also provides a directional label public opinion analysis device based on GPT, such as Figure 6 As shown, the public opinion analysis device 600 may include: A first acquisition module 601 is used to obtain, in response to a data tag configured for a target user, first public opinion data of the target user for the data tag, wherein the target user is a domain expert in a target domain, the data tag is a data tag used to identify a data category in the target domain, and the target user is determined based on a user influence value and a domain influence value of the user in the target domain, the user influence value is used to characterize the degree of influence of the user on other users, and the domain influence value is used to characterize the degree of influence of the user on the target domain; A first processing module 602 is used to obtain a public opinion theme and a sentiment category for the first public opinion data based on the GPT and the first public opinion data, wherein the GPT is used to output a corresponding public opinion theme and sentiment category according to the input first public opinion data; The second processing module 603 is used to generate a public opinion analysis result according to the public opinion topic, the emotion category and the user information of the target user.

[0062] Through the above-mentioned public opinion analysis device 600, a data tag can be configured for the target user, so that the first public opinion data of the target user under the data tag can be obtained in a targeted manner, and then the corresponding public opinion analysis result can be generated according to the obtained first public opinion data. Therefore, according to the data tag set by the user, the events in the target field can be monitored in a targeted manner, that is, the first public opinion data can be obtained in a targeted manner and a targeted public opinion analysis can be performed. On the one hand, the processing amount of the first public opinion data can be reduced and the efficiency of the public opinion analysis can be improved. On the other hand, since the first public opinion data is the public opinion data under the data tag, when the public opinion analysis result is generated based on the first public opinion data, irrelevant data can be excluded from the public opinion analysis result, thereby improving the accuracy of the public opinion analysis result and avoiding interference of irrelevant data to the user.

[0063] In a possible manner, the target user includes multiple users, and accordingly, the public opinion topic and the emotion category include multiple users. Accordingly, the second processing module 603 may include: A first processing unit is configured to use the public opinion topic as a classification dimension, classify the emotion category corresponding to the first public opinion data into the public opinion topic corresponding to the first public opinion data, and obtain the emotion category under each public opinion topic; The second processing unit is used to aggregate the sentiment categories under each public opinion topic to obtain the sentiment distribution information of the public opinion topic; The third processing unit is used to generate a public opinion analysis result based on the sentiment distribution information of each public opinion topic, multiple public opinion topics and user information of multiple target users.

[0064] In a possible manner, the public opinion analysis device 600 may further include: A second acquisition module is used to acquire second public opinion data of users in the target network platform for preset data tags through an application program interface of the target network platform, wherein the preset data tags are data tags in the target field; A first determination module, configured to determine, based on the second public opinion data, a public opinion influence value of a user corresponding to the second public opinion data in the target field; The second determination module is used to determine the target user according to the public opinion influence value of the user corresponding to the second public opinion data in the target field.

[0065] In a possible manner, the second public opinion data includes content data published by the user in the target network platform, user data of the user in the target network platform, and interaction data of the user in the target network platform. Accordingly, the first determination module may include: A first determining unit, configured to determine, based on the user data, a first influence value of the user on other users in the target network platform, and determine the first influence value as the user influence value; A second determining unit is used to determine a second impact value of the content published by the user in the target network platform in the target field according to the content data and the interaction data, and determine the second impact value as the field impact value; The third determining unit is used to determine the public opinion influence value of the user corresponding to the second public opinion data in the target field according to the user influence value and the field influence value.

[0066] In a possible manner, the user data includes first user data, second user data, and third user data, the first user data is used to indicate the credibility of the content published by the user in the target network platform, the second user data includes the fan information of the user in the target network platform, and the third user data includes the follower information of the user in the target network platform. Accordingly, the first determination unit may include: a first determining subunit, configured to determine a third impact value based on a correspondence between a preset trust score and the credibility and the first user data; a second determining subunit, configured to construct a social network of the user in the target network platform based on the fan information and the follower information, and determine, according to the social network, the degree centrality, the closeness centrality, the betweenness centrality, and the eigenvector centrality of the user in the social network, and determine a fourth influence value according to the degree centrality, the closeness centrality, the betweenness centrality, and the eigenvector centrality; A third determining subunit is used to obtain a target value by taking the target constant as a base and the number of followers in the follower information as an index, and to perform normalization processing on the target value to obtain a fifth influence value; The fourth determining subunit is used to determine the first influence value of the user on other users in the target network platform according to the third influence value, the fourth influence value and the fifth influence value.

[0067] In a possible manner, the interaction data includes first interaction data, second interaction data, third interaction data and fourth interaction data, the first interaction data includes the number of likes, the second interaction data includes the number of comments, the third interaction data includes the number of reposts, and the fourth interaction data includes the number of replies. Accordingly, the second determination unit may include: An acquisition subunit is used to acquire hot topics in the target field and target historical content data published by the user in the target network platform, wherein the hot topics are topics with a network popularity value greater than a preset network popularity value, and the target historical content data are content data published by the user in a preset time interval; a fifth determining subunit, configured to determine the semantic similarity between the historical content data and the hot topic based on the semantic recognition model, the historical content data and the hot topic; a sixth determination subunit, configured to calculate the sum of the likes, the comments, the forwardings, and the replies when the semantic similarity is greater than a preset semantic similarity threshold, and calculate the ratio of the sum to the duration corresponding to the preset time interval to obtain a content interaction value; The seventh determination subunit is used to determine the second influence value of the content published by the user in the target network platform in the target field according to the semantic similarity and the content interaction value.

[0068] Regarding the public opinion analysis device 600 in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0069] Based on the same concept, an embodiment of the present disclosure also provides a computer-readable medium on which a computer program is stored. When the program is executed by a processing device, the steps of any of the above-mentioned GPT-based directional label public opinion analysis methods are implemented.

[0070] Based on the same concept, the present disclosure also provides an electronic device, which may include: a storage device having a computer program stored thereon; A processing device is used to execute a computer program in a storage device to implement any step of the above-mentioned GPT-based directional label public opinion analysis method.

[0071] Based on the same concept, an embodiment of the present disclosure also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of any of the above-mentioned GPT-based directional label public opinion analysis methods.

[0072] Figure 7 FIG. 7 is a block diagram of an electronic device 700 according to an exemplary embodiment. Figure 7 As shown, the electronic device 700 may include: a processor 701 , a memory 702 . The electronic device 700 may also include one or more of a multimedia component 703 , an input / output (I / O) interface 704 , and a communication component 705 .

[0073] The processor 701 is used to control the overall operation of the electronic device 700 to complete all or part of the steps in the above-mentioned GPT-based directional tag public opinion analysis method. The memory 702 is used to store various types of data to support the operation of the electronic device 700, and these data may include, for example, instructions for any application or method used to operate on the electronic device 700, and application-related data, such as contact data, sent and received messages, pictures, audio, video, etc. The memory 702 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read-Only Memory, referred to as EPROM), programmable read-only memory (Programmable Read-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic memory, flash memory, disk or optical disk. The multimedia component 703 may include a screen and an audio component. The screen may be, for example, a touch screen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone, which is used to receive external audio signals. The received audio signal may be further stored in the memory 702 or sent through the communication component 705. The audio component also includes at least one speaker for outputting audio signals. The I / O interface 704 provides an interface between the processor 701 and other interface modules, and the above-mentioned other interface modules may be a keyboard, a mouse, buttons, etc. These buttons may be virtual buttons or physical buttons. The communication component 705 is used for wired or wireless communication between the electronic device 700 and other devices. Wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G or 4G, or a combination of one or more of them, so the corresponding communication component 705 may include: Wi-Fi module, Bluetooth module, NFC module.

[0074] In an exemplary embodiment, the electronic device 700 can be implemented by one or more application specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPD), programmable logic devices (PLD), field programmable gate arrays (FPGA), controllers, microcontrollers, microprocessors or other electronic components to execute the above-mentioned GPT-based directional label public opinion analysis method.

[0075] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, and when the program instructions are executed by the processor, the steps of the above-mentioned GPT-based directional tag public opinion analysis method are implemented. For example, the computer-readable storage medium can be the above-mentioned memory 702 including program instructions, and the above-mentioned program instructions can be executed by the processor 701 of the electronic device 700 to complete the above-mentioned GPT-based directional tag public opinion analysis method.

[0076] In another exemplary embodiment, a computer program product is also provided, which includes a computer program that can be executed by a processor, and when the computer program is executed by the processor, the steps of the above-mentioned GPT-based directional tag public opinion analysis method are implemented.

[0077] Figure 8 8 is a block diagram of an electronic device 800 according to an exemplary embodiment. For example, the electronic device 800 may be provided as a server. Figure 8 The electronic device 800 includes a processor 822, which may be one or more, and a memory 832 for storing a computer program executable by the processor 822. The computer program stored in the memory 832 may include one or more modules, each corresponding to a set of instructions. In addition, the processor 822 may be configured to execute the computer program to perform the above-mentioned GPT-based directional tag public opinion analysis method.

[0078] In addition, the electronic device 800 may further include a power supply component 826 and a communication component 850, wherein the power supply component 826 may be configured to perform power management of the electronic device 800, and the communication component 850 may be configured to implement communication of the electronic device 800, for example, wired or wireless communication. In addition, the electronic device 800 may further include an input / output (I / O) interface 858. The electronic device 800 may operate based on an operating system stored in the memory 832, such as Windows ServerTM, Mac OSXTM, UnixTM, LinuxTM, etc.

[0079] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, and when the program instructions are executed by the processor, the steps of the above-mentioned GPT-based directional tag public opinion analysis method are implemented. For example, the computer-readable storage medium can be the above-mentioned memory 832 including program instructions, and the above-mentioned program instructions can be executed by the processor 822 of the electronic device 800 to complete the above-mentioned GPT-based directional tag public opinion analysis method.

[0080] In another exemplary embodiment, a computer program product is also provided, which includes a computer program that can be executed by a processor, and when the computer program is executed by the processor, the steps of the above-mentioned GPT-based directional tag public opinion analysis method are implemented.

[0081] The preferred embodiments of the present disclosure are described in detail above in conjunction with the accompanying drawings; however, the present disclosure is not limited to the specific details in the above embodiments. Within the technical concept of the present disclosure, a variety of simple modifications can be made to the technical solution of the present disclosure, and these simple modifications all fall within the protection scope of the present disclosure.

[0082] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present disclosure will not further describe various possible combinations.

[0083] In addition, various embodiments of the present disclosure may be arbitrarily combined, and as long as they do not violate the concept of the present disclosure, they should also be regarded as the contents disclosed by the present disclosure.

Claims

1. A directional label public opinion analysis method based on GPT, characterized in that: The method comprises: In response to a data tag configured for a target user, first public opinion data of the target user for the data tag is obtained, wherein the target user is a domain expert in a target domain, the data tag is a data tag used to identify a data category in the target domain, and the target user is determined based on a user influence value and a domain influence value of the user in the target domain, the user influence value is used to characterize the degree of influence of the user on other users, and the domain influence value is used to characterize the degree of influence of the user on the target domain; Based on the GPT and the first public opinion data, obtaining a public opinion theme and a sentiment category for the first public opinion data, wherein the GPT is used to output a corresponding public opinion theme and sentiment category according to the input first public opinion data; Generate a public opinion analysis result based on the public opinion topic, the emotion category and the user information of the target user.

2. The method for directional label public opinion analysis based on GPT according to claim 1 is characterized in that: The target user includes a plurality of users, and correspondingly, the public opinion topics and the emotion categories include a plurality of users, and generating the public opinion analysis result according to the public opinion topics, the emotion categories and the user information of the target user includes: Taking the public opinion topic as a classification dimension, classifying the sentiment category corresponding to the first public opinion data into the public opinion topic corresponding to the first public opinion data, and obtaining the sentiment category under each public opinion topic; For each of the public opinion topics, the sentiment categories under the public opinion topic are aggregated to obtain the sentiment distribution information of the public opinion topic; A public opinion analysis result is generated based on the sentiment distribution information of each public opinion topic, multiple public opinion topics, and user information of multiple target users.

3. The method for directional label public opinion analysis based on GPT according to claim 1 or 2, characterized in that: The target user is obtained in the following manner: Obtaining second public opinion data of users in the target network platform for preset data tags through an application program interface of the target network platform, wherein the preset data tags are data tags in the target field; Determining, according to the second public opinion data, a public opinion influence value of a user corresponding to the second public opinion data in the target field; The target user is determined according to the public opinion influence value of the user corresponding to the second public opinion data in the target field.

4. The method for directional label public opinion analysis based on GPT according to claim 3 is characterized in that: The second public opinion data includes content data published by the user in the target network platform, user data of the user in the target network platform, and interaction data of the user in the target network platform. Determining the public opinion influence value of the user corresponding to the second public opinion data in the target field according to the second public opinion data includes: Determine, according to the user data, a first influence value of the user on other users in the target network platform, and determine the first influence value as the user influence value; Determine, based on the content data and the interaction data, a second impact value of the content published by the user in the target network platform in the target domain, and determine the second impact value as the domain impact value; The public opinion influence value of the user corresponding to the second public opinion data in the target field is determined according to the user influence value and the field influence value.

5. The method for directional label public opinion analysis based on GPT according to claim 4 is characterized in that: The user data includes first user data, second user data, and third user data, the first user data is used to indicate the credibility of the content published by the user in the target network platform, the second user data includes the fan information of the user in the target network platform, and the third user data includes the follower information of the user in the target network platform. Determining the first influence value of the user on other users in the target network platform according to the user data includes: Determining a third impact value based on a correspondence between a preset trust score and the credibility and the first user data; Based on the fan information and the follower information, construct a social network of the user in the target network platform, and determine the degree centrality, closeness centrality, betweenness centrality and eigenvector centrality of the user in the social network according to the social network, and determine a fourth influence value according to the degree centrality, the closeness centrality, the betweenness centrality and the eigenvector centrality; Taking the target constant as a base and the number of followers in the follower information as an index, a target value is obtained, and the target value is normalized to obtain a fifth influence value; A first influence value of the user on other users in the target network platform is determined according to the third influence value, the fourth influence value, and the fifth influence value.

6. The method for directional label public opinion analysis based on GPT according to claim 4 is characterized in that: The interactive data includes first interactive data, second interactive data, third interactive data and fourth interactive data, the first interactive data includes the number of likes, the second interactive data includes the number of comments, the third interactive data includes the number of reposts, and the fourth interactive data includes the number of replies, and determining the second impact value of the content published by the user in the target network platform in the target field according to the content data and the interactive data includes: Acquire hot topics in the target field and target historical content data published by the user in the target network platform, wherein the hot topics are topics with a network popularity value greater than a preset network popularity value, and the target historical content data are content data published by the user in a preset time interval; Determining the semantic similarity between the historical content data and the hot topic based on the semantic recognition model, the historical content data and the hot topic; When the semantic similarity is greater than a preset semantic similarity threshold, the sum of the likes, the comments, the forwardings, and the replies is calculated, and the ratio of the sum to the duration corresponding to the preset time interval is calculated to obtain a content interaction value; A second influence value of the content published by the user in the target network platform in the target field is determined according to the semantic similarity and the content interaction value.

7. A directional label public opinion analysis device based on GPT, characterized in that: The device comprises: A first acquisition module is used to obtain, in response to a data tag configured for a target user, first public opinion data of the target user for the data tag, wherein the target user is a domain expert in a target domain, the data tag is a data tag used to identify a data category in the target domain, and the target user is determined based on a user influence value and a domain influence value of the user in the target domain, the user influence value is used to characterize the degree of influence of the user on other users, and the domain influence value is used to characterize the degree of influence of the user on the target domain; A first processing module, configured to obtain a public opinion theme and a sentiment category for the first public opinion data based on the GPT and the first public opinion data, wherein the GPT is configured to output a corresponding public opinion theme and sentiment category according to the input first public opinion data; The second processing module is used to generate a public opinion analysis result according to the public opinion topic, the emotion category and the user information of the target user.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 6 are implemented.

9. An electronic device, characterized in that: include: a memory having a computer program stored thereon; A processor, configured to execute the computer program in the memory to implement the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images