An opinion intelligent monitoring method, system, device and medium

Abstract

The application discloses an opinion intelligent monitoring method, system, device and medium, comprising the following steps: continuously collecting original public opinion data containing user-generated content and associated user attribute data from multiple user interaction channels of online interactive applications; preprocessing the original public opinion data, uniquely identifying and associating the processed user-generated content with the corresponding user attribute data, and forming a structured public opinion data set; performing automatic semantic analysis on the structured public opinion data set to obtain a semantic analysis result, wherein the automatic semantic analysis comprises sentiment tendency quantization determination, semantic topic automatic identification and topic heat calculation; comprehensively displaying the semantic analysis result through a multi-dimensional graphical interface; presetting multi-dimensional monitoring rules through the multi-dimensional graphical interface, continuously monitoring the semantic analysis result based on the multi-dimensional monitoring rules, and when an early warning condition is triggered, automatically generating a hierarchical early warning information and starting an early warning response process.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, and in particular to a method, system, device and medium for intelligent monitoring of public opinion. Background Technology

[0002] With the rapid development of internet technology, online interactive applications such as online games, online social platforms, live streaming platforms, and online education platforms have become an important part of people's daily lives and entertainment. Within these applications, users generate massive amounts of real-time comments and interactive data through chatting, commenting, and providing feedback. This data constitutes the public opinion ecosystem within the applications. Timely and accurate monitoring and analysis of this public opinion data is crucial for application operators to understand user experience, identify product issues, evaluate campaign effectiveness, and mitigate negative risks.

[0003] Especially in the operation of online games, real-time feedback from players through in-game chat channels, mini-programs, and other channels is the most direct basis for understanding the game experience, locating system problems, and evaluating the effectiveness of activities. Currently, game operation teams mostly rely on passive methods such as manually checking chat logs and reviewing customer service order transfers, which has significant shortcomings: 1. Massive amounts of text rely on manual reading, which cannot be processed in real time and results in low processing efficiency; 2. It is difficult to quantify sentiment tendencies, and it cannot automatically discover hot topics and trends; the analysis is superficial. 3. The feedback content is disconnected from key attributes such as player spending, level, and server, making detailed analysis impossible; 4. It is impossible to set up automatic warnings for negative public opinion or specific issues (such as concentrated bug reports), and often the problem is only discovered after it has escalated. Summary of the Invention

[0004] In order to overcome the shortcomings of existing technologies, such as low efficiency due to manual processing and inability to perform detailed analysis.

[0005] Firstly, a method for intelligent monitoring of public opinion includes the following steps: Continuously collect raw public opinion data, including user-generated content and its associated user attribute data, from multiple user interaction channels of online interactive applications; preprocess the raw public opinion data, and uniquely identify and associate the processed user-generated content with the corresponding user attribute data to form a structured public opinion dataset. Automated semantic analysis is performed on the structured public opinion dataset to obtain semantic analysis results. The automated semantic analysis includes sentiment tendency quantification, automatic semantic topic identification, and topic popularity calculation. The semantic analysis results are comprehensively displayed through a multi-dimensional graphical interface. By pre-setting multi-dimensional monitoring rules through a multi-dimensional graphical interface, the semantic analysis results are continuously monitored based on the multi-dimensional monitoring rules. When the warning conditions are triggered, hierarchical warning information is automatically generated and the warning response process is initiated.

[0006] Optionally, the online interactive application includes online games; the user interaction channels include in-game chat systems and game mini-program feedback systems; and the user attribute data includes one or more of the following: unique user identifier, server, historical recharge amount, and player level.

[0007] Optionally, the specific steps for uniquely identifying and associating the processed user-generated content with the corresponding user attribute data to form a structured public opinion dataset are as follows: Construct a real-time data transmission pipeline to receive raw public opinion data streams from various user interaction channels, perform flow control and data buffering, and output the buffered raw public opinion data streams. The buffered raw public opinion data is cleaned and preprocessed, and the text is standardized and segmented to output standardized text. Based on the unique user identifier carried by each data point in the standardized text data, the corresponding full-dimensional user attribute data is obtained from the user data center and associated with it, and semi-structured data associated with user attributes is output. Based on semi-structured data associated with user attributes, it is stored in a database system that supports full-text search and distributed aggregation computing, and is partitioned and indexed by time and topic dimensions to finally form a structured public opinion dataset.

[0008] Optionally, the sentiment tendency quantification determination specifically involves: using a pre-trained language model adjusted with target domain corpus to perform multi-dimensional sentiment quantification scoring on a single user-generated content, while simultaneously statistically analyzing the overall sentiment distribution and sentiment change trend within a specified time window.

[0009] Optionally, the automatic semantic topic identification specifically includes: Extract high-frequency words and key phrases within a specified time period to generate a dynamic hot word map; Based on the generated dynamic hot word map, a word co-occurrence matrix and semantic similarity matrix are constructed among words. A clustering algorithm is used to cluster words with high semantic relevance, and the topic clusters with semantic relevance are automatically identified. Automatic summarization and topic naming are performed for each topic cluster, and statistics on the number of public opinions, sentiment distribution, characteristics of participating users, and time distribution under each topic are collected.

[0010] Optionally, the topic popularity calculation specifically involves calculating the popularity value of a single public opinion or corresponding topic within a preset time window based on the dissemination speed, interaction frequency, and emotional intensity of the public opinion content.

[0011] Optionally, the specific steps for generating tiered early warning information and initiating the early warning response process are as follows: Read all enabled multi-dimensional monitoring rules; Read structured public opinion data in real time and combine it with multi-dimensional analysis results to match all enabled multi-dimensional monitoring rules in parallel; push early warning information to designated receiving groups or individuals; The warning information includes at least the warning level, trigger time, trigger rules, summary of hit content, impact range assessment, and sentiment distribution statistics.

[0012] Secondly, this invention provides an intelligent public opinion monitoring system, comprising: The multi-source data acquisition module is used to continuously collect raw public opinion data, including user-generated content and its associated user attribute data, from multiple user interaction channels of online interactive applications. The data processing and storage module is used to preprocess the original public opinion data and uniquely identify and associate the processed user-generated content with the corresponding user attribute data to form a structured public opinion dataset. The multi-dimensional analysis module is used to perform automated semantic analysis on the structured public opinion dataset to obtain semantic analysis results. The automated semantic analysis includes sentiment tendency quantitative determination, automatic semantic topic identification, and topic popularity calculation. The visualization and interaction module is used to comprehensively display the semantic analysis results through a multi-dimensional graphical interface; The configurable early warning module is used to preset multi-dimensional monitoring rules through a multi-dimensional graphical interface, continuously monitor the semantic analysis results based on the multi-dimensional monitoring rules, and automatically generate hierarchical early warning information and start the early warning response process when the early warning conditions are triggered.

[0013] Thirdly, a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the intelligent public opinion monitoring method described in the first aspect.

[0014] Fourthly, an electronic device includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus. Memory, used to store computer programs; When a processor executes a program stored in memory, it implements the steps of the intelligent public opinion monitoring method described in the first aspect.

[0015] The beneficial effects of this invention are as follows: It automates the entire process of in-game public opinion monitoring, eliminating the need for game operators to manually review massive amounts of player chat logs and feedback, significantly improving the efficiency and real-time performance of game public opinion monitoring. Through multi-dimensional intelligent analysis, it can delve deeper into the real game experience and potential problems behind player comments, rather than simply relying on superficial keyword statistics. Visualized displays allow game operators to quickly grasp the overall game public opinion situation, while the configurable early warning mechanism enables a shift from passive response to proactive prevention, allowing for timely detection and handling of negative public opinion in its early stages, such as game bugs and version issues, effectively reducing the impact of public opinion risks on the game's reputation and player retention. Attached Figure Description

[0016] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0017] Figure 1 These are flowcharts from some of the embodiments; Figure 2 These are flowcharts for automatic semantic subject identification in some embodiments; Figure 3 These are system block diagrams from some embodiments. Detailed Implementation

[0018] The following will clearly and completely describe the concept, specific structure, and technical effects of the present invention in conjunction with embodiments and accompanying drawings, so as to fully understand the purpose, features, and effects of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of them. Other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are all within the scope of protection of the present invention. Furthermore, all connections / linkages involved in the patent do not simply refer to direct contact between components, but rather to the ability to form a better connection structure by adding or reducing connecting accessories according to specific implementation conditions. The various technical features in this invention can be combined interactively without contradicting each other.

[0019] This invention provides an intelligent public opinion monitoring method, which is executed by an intelligent public opinion monitoring system. The method, illustrated in this application embodiment primarily based on online games, includes the following steps: S1. Continuously collect raw public opinion data, including user-generated content and its associated user attribute data, from multiple user interaction channels of online interactive applications; preprocess the raw public opinion data, and uniquely identify and associate the processed user-generated content with the corresponding user attribute data to form a structured public opinion dataset. S2. Perform automated semantic analysis on the structured public opinion dataset to obtain semantic analysis results. The automated semantic analysis includes sentiment tendency quantification, automatic semantic topic identification, and topic popularity calculation. The semantic analysis results are comprehensively displayed through a multi-dimensional graphical interface. S3. Through a multi-dimensional graphical interface, preset multi-dimensional monitoring rules are used to continuously monitor the semantic analysis results based on the multi-dimensional monitoring rules. When the warning conditions are triggered, hierarchical warning information is automatically generated and the warning response process is initiated.

[0020] Specifically, the process begins by continuously collecting raw public opinion data from multiple user interaction channels of online interactive applications. Here, "online interactive applications" are primarily represented by online games, and "user interaction channels" are the specific entry points for game players to express their opinions and provide feedback. Core channels include the game server log interface (used to obtain real-time chat logs from all public channels, team channels, guild channels, and private chat channels across the entire server) and the game mini-program backend interface (used to obtain player-submitted feedback, bug reports, suggestions, and complaints). The collected "raw public opinion data" not only includes text content generated by players but also acquires associated user attribute data, namely, the identity and game behavior information related to the speaking player.

[0021] Next, preprocessing is performed on the collected raw data, which involves filtering and organizing the game chat and feedback data to remove useless information. Then, an operation is performed to associate the data with the unique identifier of the user attribute data, that is, by using the unique player character ID carried in each data entry, the player's comments are bound to the corresponding player's game attribute information. Finally, a structured public opinion dataset is formed, which is to transform the originally messy raw game data into a unified, standardized, and computer-processable data format.

[0022] Subsequently, automated semantic analysis is performed on the structured dataset, which uses artificial intelligence algorithms to automatically understand and analyze the meaning of the game text content. Specifically, this includes four dimensions: quantitative determination of sentiment tendency (judging whether the emotions expressed by players are positive or negative), automatic identification of semantic topics (discovering hot topics in the game discussed by players), and "topic popularity calculation" which measures the degree of attention a topic receives.

[0023] After the analysis is completed, the semantic analysis results are displayed in a comprehensive manner through a multi-dimensional graphical interface. This "multi-dimensional graphical interface" is a dedicated visualization interface for game operations built on the AMIS low-code development framework and the ECharts chart library. It can transform complex analysis results into intuitive public opinion trend charts, hot word clouds, sentiment distribution maps, and real-time public opinion lists for game operations personnel to quickly view and understand.

[0024] Finally, the system supports game operators in setting up multi-dimensional monitoring rules through a multi-dimensional graphical interface. This means setting up various monitoring conditions and triggering standards for game-related public opinion on the visual interface. The system will continuously monitor the semantic analysis results based on the multi-dimensional monitoring rules and compare the newly generated analysis results with the preset rules in real time. When the warning conditions are triggered, that is, when the analysis results meet the preset monitoring standards, the system will automatically generate graded warning information. That is, different levels of alarms will be generated according to the severity of the game problem, and the warning response process will be initiated. The warning information will be pushed to relevant personnel such as game maintenance, operation and customer service through communication tool robots (such as Lark robots).

[0025] This invention automates the entire process of in-game public opinion monitoring, eliminating the need for game operators to manually review massive amounts of player chat logs and feedback, significantly improving the efficiency and real-time performance of game public opinion monitoring. Through multi-dimensional intelligent analysis, it can delve deeper into the real game experience and potential problems behind player comments, going beyond mere keyword statistics. Visualized displays allow game operators to quickly grasp the overall game public opinion situation, while configurable early warning mechanisms enable a shift from passive response to proactive prevention, allowing for timely detection and handling of negative public opinion in its early stages, such as game bugs and version issues, effectively reducing the impact of public opinion risks on the game's reputation and player retention.

[0026] In some embodiments, the online interactive application includes online games; the user interaction channels include in-game chat systems and game mini-program feedback systems; the user attribute data includes one or more of the following: user unique identifier, server, historical recharge amount, and player level.

[0027] Specifically, online games are the primary application scenario for this method, characterized by a large user base, high real-time requirements, and rapid dissemination of public opinion. "User interaction channels include the in-game chat system and the game mini-program feedback system." These two channels are the main venues for game players to express their opinions and submit feedback. The in-game chat system is used for real-time communication between players, while the game mini-program feedback system is used for players to submit questions, suggestions, and complaints. "User attribute data includes one or more of the following: unique user identifier, server / region, historical recharge amount, and player level." The "unique user identifier" is used to distinguish different players; the "server / region" refers to the game server the player is on; the "historical recharge amount" reflects the player's spending power; and the "player level" reflects the player's level of engagement in the game.

[0028] In some embodiments, the specific steps for uniquely associating the processed user-generated content with the corresponding user attribute data to form a structured public opinion dataset are as follows: S101. Construct a real-time data transmission pipeline to receive raw public opinion data streams from various user interaction channels, perform flow control and data buffering, and output the buffered raw public opinion data streams. S102. Clean and preprocess the buffered original public opinion data, and standardize and segment the text to output standardized text. S103. Based on the unique user identifier carried by each data point in the standardized text data, obtain and associate the corresponding full-dimensional user attribute data from the user data center, and output semi-structured data associated with user attributes. S104. Based on the semi-structured data associated with user attributes, it is stored in a database system that supports full-text search and distributed aggregation computing, and partitioned and indexed according to the time dimension and the topic dimension to finally form a structured public opinion dataset.

[0029] Specifically, the first step involves building a real-time data transmission pipeline to receive raw public opinion data streams from various user interaction channels. This pipeline undergoes flow control and data buffering before outputting the buffered raw public opinion data stream. This is the entry point for the entire data processing flow. The system uses a Kafka message queue to construct this real-time data transmission pipeline, specifically designed to handle the massive amounts of real-time data pouring in from core channels such as the game server log interface (retrieving real-time chat logs from all chat channels across the server) and the game mini-program backend interface (retrieving player-submitted feedback and bug reports). When game version updates or large-scale events cause a sudden surge in data volume, the pipeline automatically performs flow control, temporarily caching data exceeding the system's processing capacity to prevent system crashes due to overload and ensure that all raw game data is received completely and without loss. The output of this step is the raw game public opinion data stream, which has undergone preliminary aggregation and ordered buffering, serving as the sole input for the next step of data cleaning and preprocessing.

[0030] Next, the buffered raw public opinion data undergoes cleaning and preprocessing, including text standardization and word segmentation, outputting standardized text. This step uses the buffered raw public opinion data stream from the previous step as the sole input to comprehensively clean up the chaotic game chat and feedback data: first, it removes invalid content (such as special symbols and garbled characters), duplicate data (such as player spam), noisy data (such as advertising information), and non-human language content (such as messages automatically sent by bots); then, it standardizes the encoding format and capitalization of all text, and uses the Jieba word segmentation tool to accurately segment and tag the Chinese game text, breaking down continuous natural language sentences into independent word units that the computer can understand and process. The output of this step is cleaned and standardized plain text data, which serves as the sole input for the next step of user attribute association.

[0031] Next, the system performs an operation based on the unique user identifier carried in each piece of standardized text data. It retrieves and associates the corresponding full-dimensional user attribute data from the user data center, outputting semi-structured data linked to user attributes. This step uses the standardized text data output from the previous step as the sole input. Each piece of standardized text data naturally carries the unique character ID of the speaking player. Based on this unique character ID, the system calls the standardized service interface of the game user center to obtain the player's full-dimensional game attribute information in real time, including their region / server, historical cumulative recharge amount, character level, registration duration, activity level, and device model. This attribute information is then bound one-to-one with the corresponding player's comments, achieving a precise "player-comment" correspondence. The output of this step is semi-structured data linked with complete player attribute information, serving as the sole input for the next step of storage and indexing.

[0032] Finally, the semi-structured data associated with user attributes is stored in a database system that supports full-text search and distributed aggregation computing. It is then partitioned and indexed by time and topic dimensions, ultimately forming a structured public opinion dataset. This step uses the semi-structured data associated with user attributes from the previous step as the sole input, storing it in the Elasticsearch distributed full-text search database system. This database is specifically optimized for fast querying and complex statistical analysis of massive amounts of text data. Simultaneously, the system partitions the data by time dimension (hourly / daily) and discussion topic dimension, establishing multi-level composite indexes, significantly improving the response speed of subsequent public opinion queries and analysis. The final output of this step is a complete, standardized, and rapidly searchable structured game public opinion dataset, providing a solid data foundation for subsequent multi-dimensional semantic analysis and early warning monitoring.

[0033] In some embodiments, the sentiment quantification determination specifically involves: using a pre-trained language model adjusted with target domain corpus to perform multi-dimensional sentiment quantification scoring on a single user-generated content, while simultaneously statistically analyzing the overall sentiment distribution and sentiment change trend within a specified time window.

[0034] Specifically, when performing sentiment quantification, a pre-trained language model adjusted with target domain corpus is first used. Here, "pre-trained language model" refers to an artificial intelligence model that has been trained on a large amount of general text. "Target domain corpus adjustment" refers to further training (also known as "fine-tuning") the model using a large amount of player chat records and feedback text in the game domain, so that it can better understand the professional terminology, internet slang, and expression habits of the game domain. This system uses a StructBERT pre-trained language model that has been fine-tuned with game domain-specific corpus, which has higher accuracy in sentiment analysis of game text.

[0035] This model performs multi-dimensional emotional quantification scoring on individual user-generated content. In other words, it scores each player's comment from multiple different emotional perspectives, rather than simply classifying it into "positive" and "negative" categories. The multi-dimensional emotions include positive, negative, neutral, angry, joyful, disappointed, and surprised. Each dimension will output a quantitative score between 0 and 1. The higher the score, the stronger the emotion. For example, "This new hero is too weak" will be judged as negative and disappointed with a high score.

[0036] At the same time, the system will statistically analyze the overall emotional distribution and emotional change trend within a specified time window. Here, "time window" refers to a continuous period of time (such as 1 hour or 1 day). The system will statistically analyze the emotional scores of all players' statements within this time period, calculate the proportion of each emotion (i.e., "emotional distribution"), and observe how the proportion of emotions changes over time (i.e., "emotional change trend").

[0037] In some embodiments, automatic semantic topic identification specifically includes: S201. Extract high-frequency words and key phrases within a specified time period and generate a dynamic hot word map; S202. Based on the generated dynamic hot word map, construct the word co-occurrence matrix and semantic similarity matrix among words, and use a clustering algorithm to cluster words with high semantic relevance, and automatically identify topic clusters with semantic relevance. S203. Perform automatic summarization and topic naming for each topic cluster, and count the number of public opinions, sentiment distribution, characteristics of participating users, and time distribution under each topic.

[0038] Specifically, the first step involves extracting high-frequency words and key phrases within a specified time period to generate a dynamic heatmap. "High-frequency words and key phrases" refer to words that appear significantly more frequently than other words in player chat and feedback, typically representing the issues players are currently most concerned about. The "dynamic heatmap" is a real-time updated set of words, where the size of each word directly reflects its frequency and importance. The system periodically scans the game's public opinion data in the database, counting the number of times each word appears within a specified time window (e.g., 1 hour, 1 day), extracting the top-ranking high-frequency words and key phrases, and generating the dynamic heatmap. For example, after a game version update, words like "crash," "lag," and "bug" quickly become high-frequency words and appear in the heatmap. The output of this step is a dynamic heatmap containing information on the words and their frequency, which serves as input for the next step of topic clustering.

[0039] Next, based on the generated dynamic hot word map, the system constructs a word co-occurrence matrix and a semantic similarity matrix. A clustering algorithm is then used to cluster words with high semantic relevance, automatically identifying semantically related topic clusters. The "word co-occurrence matrix" counts the frequency of two words appearing together in the same player's comment, reflecting the degree of association between words. The "semantic similarity matrix" calculates the semantic similarity of two words in the game context. The "clustering algorithm" is a computer algorithm that automatically groups similar data. Based on the hot word map generated in the previous step, the system constructs these two matrices to quantify the association between words. Then, a clustering algorithm is used to aggregate words that frequently appear together and are semantically similar into a group. Each group represents a hot topic discussed by players. For example, words like "crash," "black screen," "can't enter the game," and "loading failure" will be aggregated into the topic cluster "game login and startup issues." The output of this step is multiple topic clusters with clear semantics, which serve as input for the next step of topic statistics.

[0040] Finally, the system automatically summarizes and names each topic cluster, and analyzes the number of public opinions, sentiment distribution, participating user characteristics, and time distribution under each topic. Based on the core keywords within each topic cluster, the system automatically generates a concise topic name and summary, allowing operations personnel to quickly understand the core content of the topic. Simultaneously, the system analyzes the total number of public opinions under each topic, the proportion of various sentiments, the spending amount and level distribution of participating players, and the trend of the topic over time. The final output of this step is structured topic analysis results, which are used by the visualization module and the early warning module.

[0041] In some embodiments, topic popularity calculation specifically involves calculating the popularity value of a single public opinion or corresponding topic within a preset time window based on the dissemination speed, interaction frequency, and emotional intensity of public opinion content.

[0042] Specifically, this system calculates the popularity value of a single public opinion event or corresponding topic within a preset time window based on the speed of dissemination, frequency of interaction, and emotional intensity of the content. "Dissemination speed" refers to the number of times the content is forwarded and discussed per unit of time; "interaction frequency" refers to the number of user interactions such as comments and likes; "emotional intensity" refers to the strength of the emotions expressed in the content; "preset time window" refers to the pre-defined time period used to calculate popularity; and "popularity value" is a comprehensive numerical value reflecting the degree of attention and influence of the public opinion event or topic. This system achieves a quantitative assessment of the importance of public opinion, enabling objective and accurate measurement of the influence of different public opinion events and topics. By comprehensively considering the three dimensions of dissemination speed, interaction frequency, and emotional intensity, it avoids the one-sidedness of a single indicator. The calculation of popularity value helps operators distinguish the priority of different public opinion events, prioritizing those with high popularity and significant impact, thus improving the efficiency and targeting of public opinion management.

[0043] In some embodiments, the specific steps for generating tiered early warning information and initiating the early warning response process are as follows: S301. Read all enabled multi-dimensional monitoring rules; S302. Read structured public opinion data in real time and combine it with multi-dimensional analysis results to match all enabled multi-dimensional monitoring rules in parallel; push the warning information to the designated receiving groups or individuals; The warning information includes at least the warning level, trigger time, trigger rules, summary of hit content, impact range assessment, and sentiment distribution statistics.

[0044] Specifically, the first step is to read all enabled multi-dimensional monitoring rules. These "multi-dimensional monitoring rules" are various public opinion monitoring conditions pre-configured by game operators through a visual interface, including keyword rules, sentiment threshold rules, and popularity rules. Each rule contains trigger conditions, warning levels, and recipient information. These rules are stored in the system's database, and the system will periodically read all enabled rules from the database to ensure that the latest configurations take effect immediately.

[0045] Next, the system performs real-time reading of structured public opinion data and combines it with multi-dimensional analysis results, while simultaneously matching all enabled multi-dimensional monitoring rules. "Real-time reading of structured public opinion data" refers to the system continuously acquiring cleaned, standardized game chat and feedback data linked to player attributes (such as recharge amount, level, and server) from the data processing stage. "Multi-dimensional analysis results" refers to the results obtained after the system performs sentiment analysis, topic identification, and influence assessment on this data. "Parallel matching" means that the system simultaneously compares each new data point with all enabled monitoring rules, rather than matching them one by one. This method significantly improves matching speed and ensures the real-time nature of alerts.

[0046] When a piece of data meets the trigger condition of any rule, an alert will be pushed to the designated receiving group or individual. Based on the preset alert level in the rule, the system will push the alert information to the corresponding game maintenance, operations, or customer service team via instant messaging tools such as Lark robots and WeChat Work robots.

[0047] The warning information includes at least the warning level, trigger time, trigger rule, summary of the affected content, impact assessment, and sentiment distribution statistics. "Warning Level" distinguishes the severity of the problem, typically categorized into three levels: general, important, and urgent. "Trigger Time" precisely records the moment the warning was triggered. "Trigger Rule" explains which monitoring rule was triggered. "Summary of Affected Content" displays fragments of player comments that triggered the warning. "Impact Assessment" tallies the number of affected players and their server distribution. "Sentiment Distribution Statistics" displays the overall sentiment trend of the relevant public opinion.

[0048] This invention provides an intelligent public opinion monitoring system, comprising: The multi-source data acquisition module is used to continuously collect raw public opinion data, including user-generated content and its associated user attribute data, from multiple user interaction channels of online interactive applications. The data processing and storage module is used to preprocess the original public opinion data and uniquely identify and associate the processed user-generated content with the corresponding user attribute data to form a structured public opinion dataset. The multi-dimensional analysis module is used to perform automated semantic analysis on the structured public opinion dataset to obtain semantic analysis results. The automated semantic analysis includes sentiment tendency quantitative determination, automatic semantic topic identification, and topic popularity calculation. The visualization and interaction module is used to comprehensively display the semantic analysis results through a multi-dimensional graphical interface; The configurable early warning module is used to preset multi-dimensional monitoring rules through a multi-dimensional graphical interface, continuously monitor the semantic analysis results based on the multi-dimensional monitoring rules, and automatically generate hierarchical early warning information and start the early warning response process when the early warning conditions are triggered.

[0049] The present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for intelligent monitoring of public opinion.

[0050] This invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; The processor, when executing programs stored in memory, implements the steps of an intelligent public opinion monitoring method.

[0051] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A method for intelligent monitoring of public opinion, characterized in that, Includes the following steps: Continuously collect raw public opinion data, including user-generated content and its associated user attribute data, from multiple user interaction channels of online interactive applications; preprocess the raw public opinion data, and uniquely identify and associate the processed user-generated content with the corresponding user attribute data to form a structured public opinion dataset. Automated semantic analysis is performed on the structured public opinion dataset to obtain semantic analysis results. The automated semantic analysis includes sentiment tendency quantification, automatic semantic topic identification, and topic popularity calculation. The semantic analysis results are comprehensively displayed through a multi-dimensional graphical interface. By pre-setting multi-dimensional monitoring rules through a multi-dimensional graphical interface, the semantic analysis results are continuously monitored based on the multi-dimensional monitoring rules. When the warning conditions are triggered, hierarchical warning information is automatically generated and the warning response process is initiated.

2. The intelligent public opinion monitoring method according to claim 1, characterized in that, The online interactive applications include online games; the user interaction channels include in-game chat systems and game mini-program feedback systems; the user attribute data includes one or more of the following: unique user identifier, server, historical recharge amount, and player level.

3. The intelligent public opinion monitoring method according to claim 1, characterized in that, The specific steps for uniquely identifying and associating the processed user-generated content with the corresponding user attribute data to form a structured public opinion dataset are as follows: Construct a real-time data transmission pipeline to receive raw public opinion data streams from various user interaction channels, perform flow control and data buffering, and output the buffered raw public opinion data streams. The buffered raw public opinion data is cleaned and preprocessed, and the text is standardized and segmented to output standardized text. Based on the unique user identifier carried by each data point in the standardized text data, the corresponding full-dimensional user attribute data is obtained from the user data center and associated with it, and semi-structured data associated with user attributes is output. Based on semi-structured data associated with user attributes, it is stored in a database system that supports full-text search and distributed aggregation computing, and is partitioned and indexed by time and topic dimensions to finally form a structured public opinion dataset.

4. The intelligent public opinion monitoring method according to claim 1, characterized in that, The sentiment quantification judgment specifically involves: using a pre-trained language model adjusted with target domain corpus to perform multi-dimensional sentiment quantification scoring on a single user-generated content, while simultaneously statistically analyzing the overall sentiment distribution and sentiment change trend within a specified time window.

5. The intelligent public opinion monitoring method according to claim 1, characterized in that, The automatic semantic topic identification specifically includes: Extract high-frequency words and key phrases within a specified time period to generate a dynamic hot word map; Based on the generated dynamic hot word map, a word co-occurrence matrix and semantic similarity matrix are constructed among words. A clustering algorithm is used to cluster words with high semantic relevance, and the topic clusters with semantic relevance are automatically identified. Automatic summarization and topic naming are performed for each topic cluster, and statistics on the number of public opinions, sentiment distribution, characteristics of participating users, and time distribution under each topic are collected.

6. The intelligent public opinion monitoring method according to claim 1, characterized in that, The topic popularity calculation specifically involves calculating the popularity value of a single public opinion post or corresponding topic within a preset time window based on the dissemination speed, interaction frequency, and emotional intensity of public opinion content.

7. The intelligent public opinion monitoring method according to claim 1, characterized in that, The specific steps for generating tiered early warning information and initiating the early warning response process are as follows: Read all enabled multi-dimensional monitoring rules; Read structured public opinion data in real time and combine it with multi-dimensional analysis results to match all enabled multi-dimensional monitoring rules in parallel; push early warning information to designated receiving groups or individuals; The warning information includes at least the warning level, trigger time, trigger rules, summary of hit content, impact range assessment, and sentiment distribution statistics.

8. A public opinion intelligent monitoring system, characterized in that, include: The multi-source data acquisition module is used to continuously collect raw public opinion data, including user-generated content and its associated user attribute data, from multiple user interaction channels of online interactive applications. The data processing and storage module is used to preprocess the original public opinion data and uniquely identify and associate the processed user-generated content with the corresponding user attribute data to form a structured public opinion dataset. The multi-dimensional analysis module is used to perform automated semantic analysis on the structured public opinion dataset to obtain semantic analysis results. The automated semantic analysis includes sentiment tendency quantitative determination, automatic semantic topic identification, and topic popularity calculation. The visualization and interaction module is used to comprehensively display the semantic analysis results through a multi-dimensional graphical interface; The configurable early warning module is used to preset multi-dimensional monitoring rules through a multi-dimensional graphical interface, continuously monitor the semantic analysis results based on the multi-dimensional monitoring rules, and automatically generate hierarchical early warning information and start the early warning response process when the early warning conditions are triggered.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-7.

10. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, when executing a program stored in memory, implements the steps of the method according to any one of claims 1-7.

Images