Public opinion situation awareness method and device, electronic equipment and storage medium

Abstract

The invention relates to the technical field of artificial intelligence, and provides a public opinion situation awareness method and device, electronic equipment and a storage medium, and the method comprises the steps: generating a demand parameter based on an analysis demand of a user; based on the demand parameters, performing data retrieval on a preset multi-source database to obtain a target public opinion data set, each public opinion data in the target public opinion data set carrying a data tag, and the data tags representing entity attributes and semantic features of the corresponding public opinion data; and based on the demand parameters and the data labels, selecting a target agent from a plurality of preset agents, and calling the target agent to perform public opinion situation analysis on the target public opinion data set to obtain a public opinion situation perception report aiming at the analysis demand, thereby changing the current situation that traditional public opinion monitoring is only available for trees but not forests, and improving the public opinion monitoring efficiency. According to the method, crossing from single information monitoring to overall situation awareness is achieved, complex and changeable analysis requirements can be flexibly met, and the depth and accuracy of public opinion situation analysis are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method, apparatus, electronic device, and storage medium for public opinion situation perception. Background Technology

[0002] With the rapid development of the Internet and social media, online public opinion has become an important carrier for reflecting the opinions, emotions and attitudes of the public. Public opinion situation awareness provides data support for decision-making by collecting and analyzing online information, and has significant application value in fields such as corporate brand management, public safety, and crisis early warning.

[0003] Currently, public opinion situation awareness solutions mainly consist of traditional public opinion monitoring systems and machine learning-based public opinion analysis systems. Traditional systems employ a rule-based and keyword matching architecture to output statistical reports; machine learning-based systems analyze data through processes such as data collection, preprocessing, and model inference before finally outputting statistical reports. However, both solutions have significant drawbacks: firstly, they often remain at the level of single text or single indicator, allowing users to obtain only fragmented information; secondly, their system architectures are mostly pieced together from single capabilities, requiring manual reconfiguration to handle complex needs, resulting in insufficient flexibility and making it difficult to achieve truly intelligent situation awareness. Summary of the Invention

[0004] This invention provides a method, device, electronic device, and storage medium for perceiving public opinion trends, in order to solve the problems in the prior art that it is difficult to perceive public opinion trends as a whole, lacks integrated design and intelligent scheduling capabilities, and has poor flexibility.

[0005] This invention provides a method for perceiving public opinion trends, comprising: Determine the user's analytical needs, and generate requirement parameters based on those needs; Based on the aforementioned requirements parameters, data retrieval is performed on a preset multi-source database to obtain a target public opinion dataset. Each public opinion data in the target public opinion dataset carries a data tag, which represents the entity attributes and semantic features of the corresponding public opinion data. Based on the required parameters and the data tags, a target intelligent agent is selected from a set of preset intelligent agents, and the target intelligent agent is invoked to perform public opinion situation analysis on the target public opinion dataset to obtain a public opinion situation awareness report for the analysis requirements.

[0006] According to a public opinion situation awareness method provided by the present invention, the method involves selecting a target intelligent agent from a set of preset intelligent agents based on the requirement parameters and the data tags, and calling the target intelligent agent to perform public opinion situation analysis on the target public opinion dataset to obtain a public opinion situation awareness report based on the analysis requirements, including: The required parameters and the data tags are matched with a pre-set capability library of multiple intelligent agents, and a target intelligent agent combination is determined based on the matched target intelligent agents; the capability library includes the functional description information and trigger condition information of the corresponding intelligent agents; Based on the data dependencies of each target agent in the target agent combination, a task execution graph is constructed, which contains the serial or parallel execution paths of each target agent. According to the task execution map, each target agent in the target agent combination is controlled to perform public opinion situation analysis on the target public opinion dataset to obtain a public opinion situation perception report based on the analysis requirements.

[0007] According to the present invention, a method for perceiving public opinion trends includes, based on the required parameters, performing data retrieval on a preset multi-source database to obtain a target public opinion dataset, comprising: Based on the aforementioned demand parameters, data retrieval is performed on the multi-source database to obtain the various public opinion data. The aforementioned public opinion data are subjected to entity aggregation processing to obtain static tags that characterize the entity attributes of the aforementioned public opinion data; Based on the aforementioned requirement parameters, inference prompt words are constructed, and based on the inference prompt words, a large language model is driven to perform semantic inference on each public opinion data to obtain dynamic tags that characterize the semantic features of each public opinion data. Based on the aforementioned public opinion data and their corresponding static and dynamic tags, the target public opinion dataset is determined; The dynamic tags include at least one of the following: event aggregation tags for marking the same topic, information sentiment tags for characterizing emotional tendencies, risk level tags for assessing the degree of threat, and opinion conflict tags for identifying opposing viewpoints.

[0008] According to a public opinion situation awareness method provided by the present invention, the step of retrieving data from the multi-source database based on the demand parameters to obtain the various public opinion data includes: Based on the situation type and demand tag in the demand parameters, the multi-source database is filtered to obtain the target database; Based on the situation type, demand tag and industry map in the demand parameters, the data in the target database is semantically filtered to obtain the target data range, and the data in the target database within the target data range are determined as the various public opinion data. The aforementioned public opinion data are data from the target database that are associated with the industry map and conform to the situation type and the demand tag.

[0009] According to a public opinion situation perception method provided by the present invention, the target intelligent agent includes at least one of a public opinion funnel intelligent agent, a risk identification intelligent agent, a key event identification intelligent agent, and an industry map visualization intelligent agent; The step of invoking the target intelligent agent to perform public opinion situation analysis on the target public opinion dataset and obtaining a public opinion situation awareness report based on the analysis requirements includes: Based on the aforementioned public opinion funnel intelligent agent, the data retention ratio of each public opinion data in the target public opinion dataset is determined, and a data loss result is generated based on the data retention ratio. Based on the risk identification intelligent agent, the risk level label in the data tags of the target public opinion dataset is identified as high-risk public opinion data, and a risk evolution trend is generated based on the high-risk public opinion data; Based on the key event identification intelligent agent, the event aggregation tag in the data tag of the target public opinion dataset represents scattered public opinion data belonging to the same topic, and the scattered public opinion data is merged to generate the development process of key events; Based on the aforementioned industry map visualization intelligent agent, the entity interaction relationship in the static and dynamic tags of the data tags of each public opinion data is identified, and an industry chain impact map is constructed based on the entity interaction relationship. Based on at least one of the data loss results, the risk evolution trend, the development process of key events, and the industrial chain impact map, a public opinion situation awareness report is generated to meet the analysis needs.

[0010] According to a public opinion situation awareness method provided by the present invention, the step of generating a public opinion situation awareness report based on at least one of the data loss results, the risk evolution trend, the development process of key events, and the industrial chain impact map, includes: Extract unstructured conclusion text and structured statistical data from at least one of the data loss results, the risk evolution trend, the development process of key events, and the industrial chain impact map; The unstructured conclusion texts output by each target agent are fused and summarized to generate an overall situational conclusion. The structured statistical data output by each target agent is converted into visual charts, and a public opinion situation awareness report is generated based on the visual charts and the overall situation conclusions to meet the analysis requirements.

[0011] According to a public opinion situation perception method provided by the present invention, the step of generating demand parameters based on the analysis requirements includes: The analysis requirements are analyzed to obtain the user's core focus and situational awareness intent; Based on the core focus subject and the situational awareness intent, the situation type, demand tags, and industry map are determined; when the core focus subject is an enterprise, the industry map includes the enterprise's industry, upstream enterprises, downstream enterprises, and competitors; The demand parameters are generated based on the situation type, the demand tag, and the industry map.

[0012] The present invention also provides a public opinion situation awareness device, comprising: The parameter determination unit is used to determine the user's analysis requirements and generate requirement parameters based on the analysis requirements; The data retrieval unit is used to retrieve data from a preset multi-source database based on the required parameters to obtain a target public opinion dataset. Each public opinion data in the target public opinion dataset carries a data tag, and the data tag represents the entity attributes and semantic features of the corresponding public opinion data. The situation analysis unit is used to select a target intelligent agent from a set of preset intelligent agents based on the requirement parameters and the data tags, and call the target intelligent agent to perform public opinion situation analysis on the target public opinion dataset to obtain a public opinion situation perception report for the analysis requirements.

[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the public opinion situation perception method as described above.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the public opinion situation perception method as described above.

[0015] The public opinion situation perception method, device, electronic device, and storage medium provided by this invention transform user analysis needs into structured demand parameters through a large model. It utilizes multi-source heterogeneous data carrying entity attributes and semantic feature tags, and dynamically schedules dedicated intelligent agents to perform public opinion situation analysis based on demand parameters and data tags. This changes the traditional public opinion monitoring situation of only seeing the trees but not the forest, and realizes the leap from monitoring single information to overall situation perception. It can flexibly adapt to complex and ever-changing analysis needs and significantly improve the depth and accuracy of public opinion situation analysis. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the public opinion situation perception method provided by the present invention; Figure 2 This is a flowchart illustrating the semantic filtering process provided by the present invention; Figure 3 This is a flowchart illustrating the requirement parameter generation process provided by the present invention; Figure 4 This is an overall architecture diagram of the public opinion situation awareness system provided by the present invention; Figure 5 This is a schematic diagram of the public opinion situation awareness device provided by the present invention; Figure 6 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0019] With the rapid development of the internet and social media, online public opinion has become an important vehicle for reflecting public opinions, emotions, and attitudes. Public opinion situation awareness refers to the real-time monitoring of public attention, sentiment, event aggregation, conflict detection, and development trends regarding specific events, topics, or entities by collecting, analyzing, and understanding various types of information online, thus providing data support for decision-making. Public opinion situation awareness has significant application value in areas such as corporate brand management, public safety, and crisis early warning.

[0020] Current public opinion situation awareness solutions mainly include traditional public opinion monitoring systems and machine learning-based public opinion analysis systems. Traditional public opinion monitoring systems employ a rule-based and keyword matching architecture, outputting statistical reports through data collection, keyword matching, and other processes. Machine learning-based public opinion analysis systems, on the other hand, analyze data through data collection, preprocessing, and machine learning model inference, ultimately outputting statistical reports. This method has stronger semantic understanding capabilities compared to traditional public opinion monitoring systems.

[0021] In recent years, with breakthroughs in large language models in the field of natural language processing, the industry has begun to explore their application in public opinion analysis, attempting to build analysis workflows based on large models. However, despite the continuous evolution of existing technologies, significant shortcomings still exist when facing the complex needs of public opinion situation analysis, namely: On the one hand, existing systems mostly remain at the level of single text or single indicator, making it difficult to achieve a holistic understanding of public opinion trends. Whether based on rules or traditional machine learning models, they can only judge the positive or negative nature of a single piece of data, lacking the ability to model the linkage between events, topics, and subjects. As a result, users can only see fragmented information and cannot gain insight into the evolution of events, the conflict patterns, and the overall development trend.

[0022] On the other hand, the architecture of existing systems is mostly a patchwork of single-point capabilities, lacking integrated design and intelligent scheduling capabilities. When faced with complex and diverse analytical needs, it is often necessary to manually reconfigure complex rules or processes. There is a lack of a "brain" that can deeply understand user needs and automatically drive and call different analytical skills according to data characteristics. This rigid architecture leads to insufficient flexibility in the system when processing multi-source heterogeneous data, making it difficult to achieve true intelligent situational awareness.

[0023] In response, this invention provides a public opinion situation awareness method. It aims to use a large model as the core driver, transforming users' fuzzy analysis needs into precise demand parameters. Based on the entity attributes and semantic features carried by the data, it selects and calls target intelligent agents from multiple preset intelligent agents for collaborative analysis. This achieves a leap from single-point information monitoring to overall situation awareness, solving the current problems of insufficient intelligent situation awareness and scheduling capabilities, as well as the inability to achieve overall situation awareness. It significantly improves the depth and intelligence level of public opinion situation analysis.

[0024] Figure 1 This is a flowchart illustrating the public opinion situation awareness method provided by the present invention, as shown below. Figure 1 As shown, this method can be applied to public opinion situation awareness systems. The method includes: Step 110: Determine the user's analysis requirements and generate requirement parameters based on those requirements; Step 120: Based on the requirement parameters, perform data retrieval on the preset multi-source database to obtain the target public opinion dataset. Each public opinion data in the target public opinion dataset carries a data tag, which represents the entity attributes and semantic features of the corresponding public opinion data. Step 130: Based on the requirement parameters and data labels, select the target intelligent agent from multiple preset intelligent agents, and call the target intelligent agent to perform public opinion situation analysis on the target public opinion dataset to obtain a public opinion situation perception report based on the analysis requirements.

[0025] Specifically, in practical applications, when conducting public opinion situation awareness, the system first needs to accurately understand the user's intent. Unlike traditional solutions that rely on users manually configuring complex keywords, in this embodiment of the invention, the system first determines the user's analytical needs and then understands the user's intent based on these needs. These analytical needs can be unstructured descriptions input by the user through natural language, for example, "Please analyze the recent competitive situation of a certain new energy vehicle company in the European market."

[0026] Upon receiving the analysis request, the system does not directly perform simple keyword matching. Instead, it utilizes its built-in Large Language Model (LLM) as the core of its understanding to identify the user's intent and structure parameters, thereby generating requirement parameters. These requirement parameters are machine language or structured instructions, such as JSON (JavaScript Object Notation) formatted data, that the system can recognize and process.

[0027] To ensure comprehensiveness of the analysis, in this embodiment of the invention, the requirement parameters can specifically cover multiple dimensions: including the core focus subject directly extracted from the user's analysis requirements (such as a certain car company), time range, and geographical range; as well as the industry map generated using a large model (such as the industry, automatically expanded upstream companies, downstream companies, competitors, etc.), the specific situation type identified (such as competitors, technology transformation, market sentiment, etc.), and requirement tags. Through this process, the system can transform the user's vague, colloquial requirements into structured parameters that precisely control subsequent processes.

[0028] After generating clear requirements parameters, the system will perform a comprehensive data retrieval from a pre-defined multi-source database based on these parameters, such as the core focus, industry landscape, time frame, and geographical scope. This multi-source database refers to the heterogeneous data sources accessed at the system's underlying layer. Depending on the specific application scenario, it can encompass standard information repositories (such as news media and announcements), personal knowledge bases (such as research reports and documents), and real-time online data (such as data scraped from social media and forums).

[0029] After retrieval and initial aggregation, a target public opinion dataset for subsequent in-depth analysis can be obtained. To achieve a deep understanding of the situation, each piece of public opinion data in this dataset carries a data tag, which is used to characterize the entity attributes and semantic features of the corresponding public opinion data. Among them, entity attributes refer to relatively static objective information, such as the names of people, companies, cities, geographical locations, industries (such as new energy vehicles), and the attributes of the publishing media contained in the data; semantic features refer to dynamic information that needs to be derived through in-depth understanding, such as the sentiment tendency (positive / negative / neutral) expressed by the public opinion data, the specific event topic to which it belongs, whether there is a conflict of opinion, and the potential risk level.

[0030] Here, the data tags for public opinion data can be generated either through preprocessing when the data is entered into the database, or through real-time inference using a large model when retrieving public opinion data. These tags provide a structured feature index for subsequent public opinion trend analysis.

[0031] After obtaining the target public opinion dataset, the system will enter the core public opinion situation analysis stage. That is, to cope with complex analysis scenarios, this embodiment of the invention no longer relies on a single general model, but pre-constructs multiple intelligent agents. Each intelligent agent can be regarded as an independent expert unit encapsulating specific analytical skills, prompt word templates, and logical processes. For example, a public opinion funnel intelligent agent specifically responsible for identifying false information can be pre-constructed, a risk identification intelligent agent specifically responsible for identifying risks, an industry map visualization intelligent agent specifically responsible for drawing industry chains and relationships, and a competitor comparison intelligent agent specifically responsible for comparing competing products, etc.

[0032] When the system executes, it first performs route selection based on demand parameters and data labels to obtain the target intelligent agent. For example, when the demand parameters point to competitors and the data labels show a large amount of price war semantics, the system will automatically select risk identification intelligent agents, competitor comparison intelligent agents, etc., from multiple pre-built intelligent agents as target intelligent agents.

[0033] Furthermore, these selected target agents can be invoked to perform public opinion trend analysis on the target public opinion dataset, thereby obtaining a public opinion trend awareness report tailored to the user's analytical needs. Specifically, the target agents utilize the reasoning capabilities of large models to perform in-depth processing on the public opinion data in the target public opinion dataset, including event aggregation, causal inference, and sentiment evolution analysis, ultimately generating a public opinion trend awareness report. This public opinion trend awareness report is not a simple list display, but a comprehensive result that includes conclusive insights tailored to the user's analytical needs, visual charts, risk warnings, and the evolution path of the trend.

[0034] The public opinion situation perception method provided by this invention transforms users' analytical needs into structured demand parameters through a large model. It utilizes multi-source heterogeneous data carrying entity attributes and semantic feature tags, and dynamically schedules dedicated intelligent agents to perform public opinion situation analysis based on demand parameters and data tags. This changes the traditional public opinion monitoring situation of only seeing the trees but not the forest, and realizes the leap from monitoring single information to overall situation perception. It can flexibly adapt to complex and ever-changing analytical needs, and significantly improve the depth and accuracy of public opinion situation analysis.

[0035] Based on the above embodiments, step 130 includes: The requirement parameters and data tags are matched with the capability libraries of multiple pre-set intelligent agents, and the target intelligent agent combination is determined based on the matched target intelligent agents; the capability library includes the functional description information and trigger condition information of the corresponding intelligent agents; Based on the data dependencies of each target agent in the target agent combination, a task execution graph is constructed, which contains the serial or parallel execution paths of each target agent. According to the task execution map, each target agent in the control target agent group performs public opinion situation analysis on the target public opinion dataset and obtains a public opinion situation perception report based on the analysis needs.

[0036] Specifically, the process of selecting and invoking the target intelligent agent based on the requirement parameters and data labels can include: To achieve automated scheduling of agents, the system pre-maintains a capability library, which can be understood as a registry or skill menu for agents. It records in detail the functional description information of each pre-built agent (i.e., what the agent is good at, such as being good at extracting event context from long texts or being good at calculating emotion values), as well as the trigger condition information (i.e., under what circumstances the agent should be activated).

[0037] During runtime, the system takes upstream-generated demand parameters (representing user intent) and data tags (representing data features) as input conditions and performs multi-dimensional semantic matching with various information in the capability library. For example, when the demand parameters explicitly indicate the need for competitor analysis, and the data tags show that the target public opinion dataset contains a large number of semantic features related to competitor pricing, the system will accurately match agents with corresponding analytical capabilities, such as competitor comparison agents, based on trigger conditions in the capability library, while filtering out irrelevant agents. Through this mechanism, the system can select a set of agents that best suit the current task from numerous candidates, i.e., the target agent combination.

[0038] Furthermore, after determining the combination of target agents, it is also necessary to determine the order in which these target agents are invoked. This is because different target agents may have data dependencies in their processing logic; for example, the input of one target agent may depend on the output of another. Therefore, the system cannot simply invoke these target agents in an unordered manner. Instead, it must dynamically construct a task execution graph based on these dependencies. This graph is essentially a directed acyclic graph, which clearly defines the collaborative flow logic between the target agents.

[0039] Under the planning of the task execution graph, the system can identify which target agents do not interfere with each other, thereby planning parallel execution paths. For example, the data emotion statistics agent and the distribution map agent can process the same batch of data at the same time to improve efficiency. At the same time, it can also identify which target agents must be executed in sequence, thereby planning serial execution paths. For example, the risk identification agent and the key event identification agent are usually linked and executed serially.

[0040] Following this, the system strictly follows the logical flow of the task execution graph, systematically scheduling and controlling the various target agents within the target agent group to perform their tasks. This ensures that each target agent, along its specific execution path, analyzes the target public opinion dataset to understand the public opinion situation, and the intermediate results are circulated among the target agents. Finally, the system aggregates and integrates the fragmented analysis results output by each target agent (such as risk evolution trends, the development process of key events, and the impact map of the industrial chain) to generate a logically rigorous and detailed public opinion situation awareness report.

[0041] The capability library described above is shown in Table 1 below: Table 1: Relationship between Functional Descriptions and Triggering Conditions of Each Agent In this embodiment of the invention, a capability library matching mechanism and a task execution graph orchestration mechanism are used to achieve personalized scheduling of intelligent agents. This not only enables the flexible formation of expert teams based on user needs and data characteristics, but also significantly improves analysis efficiency through a combination of parallel and serial execution paths, ensuring the accuracy and timeliness of complex public opinion situation analysis tasks.

[0042] Based on the above embodiments, step 120 includes: Based on the demand parameters, data retrieval is performed on multi-source databases to obtain various public opinion data; Entity aggregation processing is performed on each public opinion data to obtain static labels that represent the entity attributes of each public opinion data. Based on the demand parameters, inference prompt words are constructed, and the large language model is driven by the inference prompt words to perform semantic inference on each public opinion data to obtain dynamic tags that represent the semantic features of each public opinion data. The dynamic tags include at least one of the following: event aggregation tags for marking the same topic, information sentiment tags for representing emotional tendencies, risk level tags for assessing the degree of threat, and opinion conflict tags for identifying opposing viewpoints. Based on various public opinion data and their corresponding static and dynamic tags, the target public opinion dataset is determined.

[0043] Specifically, the process of using demand parameters to retrieve data from multiple source databases to obtain public opinion data carrying data tags can include: First, the system uses demand parameters, such as core stakeholders, industry landscape, time range, and geographical scope, as search criteria to conduct a broad search across multiple databases, including information repositories, personal knowledge bases, and online data, to obtain massive amounts of public opinion data. While this public opinion data is relevant, it remains unstructured text and lacks the deep attributes that machines can directly understand.

[0044] Therefore, in this embodiment of the invention, after retrieving each piece of public opinion data, it is also necessary to perform tagging processing to obtain data tags for each piece of public opinion data. That is, entity aggregation processing and semantic reasoning can be performed on each piece of public opinion data to obtain static tags representing the entity attributes of each piece of public opinion data and dynamic tags representing the semantic features.

[0045] Specifically, in order to quickly sort out the basic structure of public opinion data, the system can perform entity aggregation processing on the retrieved public opinion data to gather the scattered information into specific objective objects. That is, by using named entity recognition, dictionary matching and other technologies, the system can identify the names of people, companies, cities, geographical locations, industries and other information involved in each piece of public opinion data.

[0046] Through this process, the system assigns static tags to each piece of public opinion data. These tags are called static because they represent entity attributes—objectively existing information that doesn't change with the analytical perspective. For example, if an article mentions product A, this entity attribute remains constant regardless of what the user wants to analyze.

[0047] Simultaneously, the system can construct inference prompts based on demand parameters and drive a large language model to perform semantic inference on various public opinion data, thereby obtaining dynamic tags representing the semantic features of each public opinion data. Specifically, the system also needs to acquire in-depth information that reflects changes in the situation. To this end, the system will dynamically construct inference prompts based on demand parameters, such as whether the user is concerned about the risk of a price war. These prompts are equivalent to specific instructions to the large model, such as asking whether the text contains price competition behavior, and if so, determining its risk level.

[0048] The system utilizes a large model driven by inference prompts as its inference engine to perform deep semantic reasoning on various public opinion data, thereby obtaining dynamic tags for each data point. These dynamic tags represent the semantic features of the public opinion data. They are called dynamic because these tags are generated in real-time based on the inference prompts and are highly customizable.

[0049] In this embodiment of the invention, the dynamic tag may include at least one of the following: Event aggregation tags are used to identify whether multiple seemingly different reports are discussing the same core event, thereby tagging the same topic and avoiding information redundancy. Information sentiment tags: not only judge positive or negative, but also use them to represent the subtlety of emotional tendencies (such as anger, panic, expectation), and even support users to define emotional boundaries; Risk level label: used to assess the level of threat, such as classifying risks into three levels: low, medium, and high, in order to trigger subsequent early warning mechanisms; Opinion Conflict Tag: Used to identify opposing viewpoints, such as identifying two camps in the market that support and oppose a certain regulation.

[0050] After the above processing, the retrieved public opinion data has been fully structured. At this point, the system can encapsulate the original public opinion data with its accompanying static and dynamic tags to finally obtain the target public opinion dataset.

[0051] In this embodiment of the invention, a tagging system combining static and dynamic elements is used. Static tags ensure the efficiency of basic processing, while dynamic tags driven by a large model enable a deep understanding of data semantics and on-demand customization. This allows the system to accurately extract high-value information such as events, emotions, risks, and opinions from massive amounts of data, laying a solid data foundation for subsequent public opinion situation analysis of the target intelligent agent.

[0052] Based on the above embodiments, and based on the demand parameters, data retrieval is performed on multi-source databases to obtain various public opinion data, including: Based on the situation type and demand label in the demand parameters, database filtering is performed on the multi-source database to obtain the target database; Based on the situation type, demand tag and industry map in the demand parameters, semantic filtering is performed on the data in the target database to obtain the target data range, and the data within the target data range in the target database are identified as each public opinion data. Each piece of public opinion data is data from the target database that is associated with the industry map and conforms to the situation type and demand tags.

[0053] Specifically, the above implementation of data retrieval from multi-source databases does not involve a simple full database scan, but rather employs a coarse-to-fine funnel-style filtering strategy to ensure the accuracy and relevance of the data sources. This process specifically includes: First, the system performs an initial selection of data sources, i.e., database filtering. The system reads the requirement parameters and, based on the core focus, situation type (e.g., whether the user wants to analyze external market competition or internal management risks), requirement tags, time range, and geographical scope, determines which databases are relevant to the user's analytical needs from information repositories, personal knowledge bases, and online data. For example, if the situation type is external market trends, the system may only select information repositories and online data, while filtering out personal knowledge bases. Through this source control, the system identifies the target databases for subsequent retrieval, thus avoiding wasting computing power on irrelevant data sources.

[0054] Next, after identifying the data source, the system further utilizes the semantic understanding capabilities of the large model to perform semantic filtering to determine the data scope for acquiring public opinion data, i.e., the target data scope. In this process, the system introduces the industry map, which is extremely crucial in the requirement parameters. This industry map depicts the ecological network of the core stakeholders (including upstream enterprises, downstream enterprises, competitors, etc.). Figure 2 This is a flowchart illustrating the semantic filtering process provided by the present invention, such as... Figure 2 As shown, the system uses industry map, situation type, and demand tag as joint filtering conditions to scan the data in the target database to determine the target data range for obtaining public opinion data. Data within this range that is associated with the industry map and meets the situation type and demand tag are identified as public opinion data.

[0055] Unlike traditional keyword matching, semantic filtering here determines whether the data content substantially falls within the scope of analysis. For example, the system will determine whether a news story truly involves upstream raw material suppliers in the industry map and whether its content conforms to the market trend type. Only data that the system determines is related to the industry map and meets the trend type and demand tag will be retained and identified as the final required public opinion data.

[0056] In this embodiment of the invention, a two-level filtering mechanism of database screening and semantic screening is used to achieve high-precision range locking during the data acquisition stage by utilizing industry maps and situation types. This ensures that every piece of public opinion data that subsequently enters the analysis process is high-value data that is closely related to the industrial chain ecosystem that users are concerned about and meets specific situation characteristics, effectively solving the signal-to-noise ratio problem in massive data.

[0057] Based on the above embodiments, the target intelligent agent includes at least one of a public opinion funnel intelligent agent, a risk identification intelligent agent, a key event identification intelligent agent, and an industry map visualization intelligent agent; in step 130, the target intelligent agent is invoked to perform public opinion situation analysis on the target public opinion dataset to obtain a public opinion situation awareness report tailored to the analysis needs, including: Based on the public opinion funnel intelligent agent, the data retention ratio of each public opinion data in the target public opinion dataset is determined, and the data loss result is generated based on the data retention ratio; Based on the risk identification intelligent agent, identify public opinion data with high risk level labels in the data labels of the target public opinion dataset, and generate risk evolution trends based on the high-risk public opinion data; Based on the key event identification intelligent agent, the event aggregation label in the data label of the target public opinion dataset represents scattered public opinion data belonging to the same topic. The scattered public opinion data is merged to generate the development process of key events. Based on the industry map visualization intelligent agent, the static labels and dynamic labels in the data tags of each public opinion data are identified to identify the entity interaction relationship, and the industry chain impact map is constructed based on the entity interaction relationship. Based on at least one of the following: data loss results, risk evolution trends, development process of key events, and industrial chain impact map, generate a public opinion situation awareness report tailored to the analysis needs.

[0058] Specifically, the process of calling the target intelligent agent to perform public opinion situation analysis on the target public opinion dataset and obtaining a public opinion situation awareness report tailored to the analysis needs may include: In detail, embodiments of the present invention provide a variety of intelligent agents with different functions. According to actual analysis needs, target intelligent agents can be flexibly selected or combined, specifically including at least one of the following: a public opinion funnel intelligent agent responsible for data quantification and monitoring, a risk identification intelligent agent responsible for security early warning, a key event identification intelligent agent responsible for context sorting, and an industry graph visualization intelligent agent responsible for relationship mining.

[0059] The process by which the system invokes these intelligent agents to perform in-depth analysis and processing of the target public opinion dataset specifically covers the following dimensions: In public opinion analysis, users are often concerned with the filtering process that transforms massive amounts of data into the final high-value information. To address this, the system utilizes a public opinion funnel agent. This agent does not directly analyze the content but monitors the data flow process. It statistically calculates the data retention rate for each piece of public opinion data, analyzing how the data volume decreases at each stage—from initial data collection to deduplication, relevance filtering, and final selection. Based on this, the agent generates data loss results. These results are typically presented as a funnel chart, visually illustrating the information filtering process and loss, helping users assess the breadth and accuracy of their analysis.

[0060] To address the core risk avoidance need in public opinion monitoring, the system invokes a risk identification AI. This AI uses data tags within public opinion data for precise scanning, particularly identifying data with high-risk risk level tags, such as those displaying "Legal Risk: High" or "Public Opinion Crisis: Severe." After identifying these key data points, the AI ​​further analyzes the density changes of risk signals over time, generating a risk evolution trend. This trend can be a curve reflecting the rise or fall of risk intensity over time, helping users determine whether a crisis is escalating or has subsided.

[0061] To address the cognitive burden caused by fragmented information, the system utilizes a key event identification agent. This agent, employing event aggregation tags, astutely identifies scattered public opinion data that, while seemingly different in title, actually belong to the same topic. For example, multiple reports might discuss the postponement of a car company's product launch, an executive's explanation of the postponement, and shareholders' reactions to the delay. The agent logically merges these scattered public opinion data, connecting them into a complete event chain and outlining its causes, process, and results, thereby generating a key event development process. This allows users to examine public opinion on an event-by-event basis rather than on a single news article basis.

[0062] To present the macro-level industry impact, the system invokes an industry graph visualization agent. This agent delves into the implicit entity interaction relationships within static labels (such as enterprise entities) and dynamic labels (such as conflicting viewpoints). For example, it can identify that a price increase by company A (upstream) led to a decrease in profits for company B (downstream). Based on these interactions, the agent generates a dynamic industry chain impact graph, which visually demonstrates how public opinion events are transmitted and spread between upstream and downstream sectors of the industry chain.

[0063] Finally, the system will select one or more results from the outputs of the aforementioned target intelligent agents based on the user's analytical needs, and generate the final public opinion situation awareness report.

[0064] In this embodiment of the invention, through the professional division of labor among multiple agents, public opinion data is deeply deconstructed from four dimensions: data quality monitoring, risk warning, event context sorting, and industrial chain relationship mining. This not only achieves comprehensive coverage of public opinion points (single high-risk data), lines (event development process), and surfaces (industrial chain map), but also presents the complex situation changes to users intuitively through visualization.

[0065] Based on the above embodiments, and based on at least one of the following: data loss results, risk evolution trends, key event development processes, and industry chain impact maps, a public opinion situation awareness report tailored to analytical needs is generated, including: Extract unstructured conclusion text and structured statistical data from at least one of the following: data loss results, risk evolution trends, development process of key events, and industrial chain impact map; The unstructured conclusion texts output by each target agent are fused and summarized to generate an overall situational conclusion. The structured statistical data output by each target agent is converted into visual charts, and based on the visual charts and overall situation conclusions, a public opinion situation awareness report is generated to meet the analysis needs.

[0066] Specifically, the process of generating the public opinion situation awareness report includes: First, the system deconstructs and extracts the results output by each target agent. Specifically, the output of a target agent typically contains two distinct types of information: one is unstructured conclusion text generated by the target agent based on large model reasoning, such as qualitative analysis descriptions like "Due to a recent surge in quality complaints, brand trust has reached an inflection point" output by a risk identification agent; the other is structured statistical data generated during the calculation process, such as specific numerical sequences of risk indices, specific timestamps of event nodes, and coordinate data of industry chain nodes. The system separates these two types of information in this step, thus preparing for subsequent classification processing.

[0067] Next, to avoid the report content appearing fragmented and pieced together, the system needs to deeply integrate the extracted unstructured conclusion texts. This process will again call the large model to understand the logical connections between the conclusions output by different target agents. For example, by combining the unstructured conclusion texts in the development process of key events with the unstructured conclusion texts in the risk evolution trend, duplicate information can be removed and the core viewpoints can be extracted.

[0068] Through this process, the system generates an overall situational conclusion from a macro perspective. This conclusion is no longer a list of single-point analyses, but a summary or abstract of the current public opinion situation, which can directly answer users' core questions of "what happened, what are the impacts, and what should be done."

[0069] Meanwhile, the system maps and converts the extracted structured statistical data into visual charts, turning the statistical data into intuitive visual elements. For example, it converts the data loss ratio in the data loss results into a funnel chart, the statistical data in the risk evolution trend into a dynamic line chart, and the relationships in the industrial chain impact map into a knowledge graph network.

[0070] Finally, the system combines the highly summarized overall situation conclusions (as a summary or interpretation of the report) with intuitive visualization charts (as data support for the report) to generate the final public opinion situation awareness report.

[0071] In this embodiment of the invention, by combining text summaries with data visualization, the pain points of traditional public opinion reports—either being difficult to understand due to excessive data or having empty viewpoints without data—are solved. The generated public opinion situation awareness report has both the deep logical explanation capabilities provided by the large model and the objective intuitiveness supported by data, thereby greatly improving the efficiency of users' perception of public opinion situation and the quality of their decision-making.

[0072] Based on the above embodiments, requirement parameters are generated based on the analyzed requirements, including: The analysis requirements are analyzed to obtain the user's core focus and situational awareness intent; Based on the core focus and situational awareness intent, determine the situation type, demand tags, and industry map; when the core focus is an enterprise, the industry map includes the enterprise's industry, upstream enterprises, downstream enterprises, and competitors; Demand parameters are generated based on situation type, demand tags, and industry map.

[0073] Specifically, the process of generating requirement parameters based on the analyzed requirements can include: Figure 3 This is a flowchart illustrating the requirement parameter generation process provided by the present invention, such as... Figure 3 As shown, when a user inputs an unstructured description via natural language, such as "Please analyze the recent competitive landscape of a certain new energy vehicle company in the European market" as an analysis request, the system first uses a large model to perform semantic parsing. In this process, the system mainly extracts two key elements: one is the user's core focus, i.e., the core object the user wants to analyze, such as a certain new energy vehicle company; the other is the situational awareness intent, i.e., the direction or purpose the user wants to analyze, such as the competitive landscape in the European market. This process filters out meaningless interjections and accurately targets the core of the analysis.

[0074] Once the target is identified, the system does not limit itself to the words mentioned by the user, but rather conducts in-depth knowledge expansion and classification based on the core focus and situational awareness intent. That is, the system first determines the situation type based on the situational awareness intent, such as mapping the competitive situation in the European market to the system's preset situation types such as competitors and competitive advantages, and generates demand tags to assist in filtering, such as tags for exports, tariffs, and anti-subsidies.

[0075] More importantly, the system utilizes the knowledge base of a large model to construct an industry map. For example, when the core focus is on a company, such as a new energy vehicle manufacturer, the industry map constructed by the system using the large model not only includes the company itself but also automatically associates and expands to the company's related industries (e.g., new energy vehicles), upstream companies (e.g., battery suppliers, lithium mining companies), downstream companies (e.g., dealer groups), and competitors. This automatic association mechanism is extremely crucial, as it instantly broadens the analytical perspective from a single company to the entire industry chain ecosystem, thereby ensuring the comprehensiveness of subsequent situational analysis.

[0076] Finally, the system standardizes and encapsulates the identified situation types, generated demand tags, and constructed industry maps to generate demand parameters. These demand parameters act like a detailed battle map, instructing the subsequent data retrieval process on what kind of public opinion data (involving upstream and downstream of the industry chain) to look for, which agents to invoke, and what these agents should focus on analyzing (involving specific situation types and tags).

[0077] In this embodiment of the invention, the semantic parsing and knowledge association capabilities of the large model enable an unexpected understanding of user needs. The system not only understands the user's stated requirements, but also automatically completes the industry chain background information that the user did not express but which is crucial for the analysis of public opinion trends. This lays a key logical foundation for generating high-quality public opinion trend reports.

[0078] Figure 4 This is an overall architecture diagram of the public opinion situation awareness system provided by the present invention, as shown below. Figure 4 As shown, the system constructs a four-layer collaborative closed-loop architecture driven by a large language model, realizing full-process automation from user fuzzy commands to professional situation reports.

[0079] Specifically, the entire workflow is initiated by the demand extraction layer, which serves as the system's understanding area. It uses a large model to deeply analyze the user's natural language input, automatically associates and constructs an industry map, and transforms unstructured descriptions into precise structured parameters. Next, it flows to the data scope layer, which serves as the filtering area. Based on the demand parameters, it accurately defines the target data scope in a multi-source database and uses a dynamic and static tagging system to complete the semantic cleaning and feature labeling of public opinion data.

[0080] Following this, the public opinion data carrying data tags enters the core intelligent agent processing layer. As the processing area, this layer no longer relies on a single model, but dynamically schedules multiple specialized intelligent agents, such as public opinion funnel, risk identification, and event aggregation, to perform serial or parallel collaborative operations based on the characteristics of the current task, thereby completing a deep deconstruction of the public opinion situation. Finally, the results produced by all target intelligent agents converge to the release preview layer. This layer transforms the complex and multi-dimensional results into intuitive visualization maps and interpretable overall situation conclusions, and generates the final public opinion situation awareness report based on these two, thus completing the systematic reconstruction of online public opinion from single-point monitoring to global situation awareness.

[0081] The public opinion situation perception device provided by the present invention is described below. The public opinion situation perception device described below and the public opinion situation perception method described above can be referred to in correspondence.

[0082] Figure 5 This is a schematic diagram of the public opinion situation awareness device provided by the present invention, as shown below. Figure 5 As shown, the device includes: The parameter determination unit 510 is used to determine the user's analysis requirements and generate requirement parameters based on the analysis requirements; The data retrieval unit 520 is used to retrieve data from a preset multi-source database based on the required parameters to obtain a target public opinion dataset. Each public opinion data in the target public opinion dataset carries a data tag, and the data tag represents the entity attributes and semantic features of the corresponding public opinion data. The situation analysis unit 530 is used to select a target intelligent agent from a plurality of preset intelligent agents based on the requirement parameters and the data tags, and call the target intelligent agent to perform public opinion situation analysis on the target public opinion dataset to obtain a public opinion situation perception report for the analysis requirements.

[0083] The public opinion situation awareness device provided by this invention transforms users' analysis needs into structured demand parameters through a large model. It utilizes multi-source heterogeneous data carrying entity attributes and semantic feature tags, and dynamically schedules dedicated intelligent agents to perform public opinion situation analysis based on demand parameters and data tags. This changes the traditional public opinion monitoring situation of only seeing the trees but not the forest, and realizes the leap from monitoring single information to overall situation awareness. It can flexibly adapt to complex and ever-changing analysis needs, and significantly improve the depth and accuracy of public opinion situation analysis.

[0084] Based on the above embodiments, the situation analysis unit 530 is used for: The required parameters and the data tags are matched with a pre-set capability library of multiple intelligent agents, and a target intelligent agent combination is determined based on the matched target intelligent agents; the capability library includes the functional description information and trigger condition information of the corresponding intelligent agents; Based on the data dependencies of each target agent in the target agent combination, a task execution graph is constructed, which contains the serial or parallel execution paths of each target agent. According to the task execution map, each target agent in the target agent combination is controlled to perform public opinion situation analysis on the target public opinion dataset to obtain a public opinion situation perception report based on the analysis requirements.

[0085] Based on the above embodiments, the data retrieval unit 520 is used for: Based on the aforementioned demand parameters, data retrieval is performed on the multi-source database to obtain the various public opinion data. The aforementioned public opinion data are subjected to entity aggregation processing to obtain static tags that characterize the entity attributes of the aforementioned public opinion data; Based on the aforementioned requirement parameters, inference prompt words are constructed, and based on the inference prompt words, a large language model is driven to perform semantic inference on each public opinion data to obtain dynamic tags that characterize the semantic features of each public opinion data. Based on the aforementioned public opinion data and their corresponding static and dynamic tags, the target public opinion dataset is determined; The dynamic tags include at least one of the following: event aggregation tags for marking the same topic, information sentiment tags for characterizing emotional tendencies, risk level tags for assessing the degree of threat, and opinion conflict tags for identifying opposing viewpoints.

[0086] Based on the above embodiments, the data retrieval unit 520 is used for: Based on the situation type and demand tag in the demand parameters, the multi-source database is filtered to obtain the target database; Based on the situation type, demand tag and industry map in the demand parameters, the data in the target database is semantically filtered to obtain the target data range, and the data in the target database within the target data range are determined as the various public opinion data. The aforementioned public opinion data are data from the target database that are associated with the industry map and conform to the situation type and the demand tag.

[0087] Based on the above embodiments, the target intelligent agent includes at least one of a public opinion funnel intelligent agent, a risk identification intelligent agent, a key event identification intelligent agent, and an industry map visualization intelligent agent; the situation analysis unit 530 is used for: Based on the aforementioned public opinion funnel intelligent agent, the data retention ratio of each public opinion data in the target public opinion dataset is determined, and a data loss result is generated based on the data retention ratio. Based on the risk identification intelligent agent, the risk level label in the data tags of the target public opinion dataset is identified as high-risk public opinion data, and a risk evolution trend is generated based on the high-risk public opinion data; Based on the key event identification intelligent agent, the event aggregation tag in the data tag of the target public opinion dataset represents scattered public opinion data belonging to the same topic, and the scattered public opinion data is merged to generate the development process of key events; Based on the aforementioned industry map visualization intelligent agent, the entity interaction relationship in the static and dynamic tags of the data tags of each public opinion data is identified, and an industry chain impact map is constructed based on the entity interaction relationship. Based on at least one of the data loss results, the risk evolution trend, the development process of key events, and the industrial chain impact map, a public opinion situation awareness report is generated to meet the analysis needs.

[0088] Based on the above embodiments, the situation analysis unit 530 is used for: Extract unstructured conclusion text and structured statistical data from at least one of the data loss results, the risk evolution trend, the development process of key events, and the industrial chain impact map; The unstructured conclusion texts output by each target agent are fused and summarized to generate an overall situational conclusion. The structured statistical data output by each target agent is converted into visual charts, and a public opinion situation awareness report is generated based on the visual charts and the overall situation conclusions to meet the analysis requirements.

[0089] Based on the above embodiments, the parameter determination unit 510 is used for: The analysis requirements are analyzed to obtain the user's core focus and situational awareness intent; Based on the core focus subject and the situational awareness intent, the situation type, demand tags, and industry map are determined; when the core focus subject is an enterprise, the industry map includes the enterprise's industry, upstream enterprises, downstream enterprises, and competitors; The demand parameters are generated based on the situation type, the demand tag, and the industry map.

[0090] Figure 6 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 6As shown, the electronic device may include: a processor 610, a communications interface 620, a memory 630, and a communication bus 640, wherein the processor 610, the communications interface 620, and the memory 630 communicate with each other through the communication bus 640. The processor 610 can call logical instructions in the memory 630 to execute a public opinion situation awareness method, which includes: determining the user's analysis needs and generating requirement parameters based on the analysis needs; based on the requirement parameters, performing data retrieval on a preset multi-source database to obtain a target public opinion dataset, wherein each public opinion data in the target public opinion dataset carries a data tag, and the data tag represents the entity attributes and semantic features of the corresponding public opinion data; based on the requirement parameters and the data tags, selecting a target intelligent agent from a preset plurality of intelligent agents, and calling the target intelligent agent to perform public opinion situation analysis on the target public opinion dataset to obtain a public opinion situation awareness report for the analysis needs.

[0091] Furthermore, the logical instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0092] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, wherein when the program instructions are executed by a computer, the computer is able to execute the public opinion situation perception method provided by the above methods, the method comprising: determining the user's analysis needs and generating requirement parameters based on the analysis needs; performing data retrieval on a preset multi-source database based on the requirement parameters to obtain a target public opinion dataset, wherein each public opinion data in the target public opinion dataset carries a data tag, the data tag representing the entity attributes and semantic features of the corresponding public opinion data; selecting a target intelligent agent from a preset plurality of intelligent agents based on the requirement parameters and the data tags, and calling the target intelligent agent to perform public opinion situation analysis on the target public opinion dataset to obtain a public opinion situation perception report for the analysis needs.

[0093] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the public opinion situation awareness method provided by the above-described methods. The method includes: determining the user's analysis needs and generating requirement parameters based on the analysis needs; performing data retrieval on a preset multi-source database based on the requirement parameters to obtain a target public opinion dataset, wherein each public opinion data in the target public opinion dataset carries a data tag, and the data tag represents the entity attributes and semantic features of the corresponding public opinion data; selecting a target intelligent agent from a preset plurality of intelligent agents based on the requirement parameters and the data tags, and calling the target intelligent agent to perform public opinion situation analysis on the target public opinion dataset to obtain a public opinion situation awareness report for the analysis needs.

[0094] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0095] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for perceiving public opinion trends, characterized in that, include: Determine the user's analytical needs, and generate requirement parameters based on those needs; Based on the aforementioned requirements parameters, data retrieval is performed on a preset multi-source database to obtain a target public opinion dataset. Each public opinion data in the target public opinion dataset carries a data tag, which represents the entity attributes and semantic features of the corresponding public opinion data. Based on the required parameters and the data tags, a target intelligent agent is selected from a set of preset intelligent agents, and the target intelligent agent is invoked to perform public opinion situation analysis on the target public opinion dataset to obtain a public opinion situation awareness report for the analysis requirements.

2. The public opinion situation perception method according to claim 1, characterized in that, Based on the requirement parameters and the data tags, a target intelligent agent is selected from a set of preset intelligent agents, and the target intelligent agent is invoked to perform public opinion situation analysis on the target public opinion dataset to obtain a public opinion situation awareness report for the analysis requirements, including: The required parameters and the data tags are matched with a pre-set capability library of multiple intelligent agents, and a target intelligent agent combination is determined based on the matched target intelligent agents; the capability library includes the functional description information and trigger condition information of the corresponding intelligent agents; Based on the data dependencies of each target agent in the target agent combination, a task execution graph is constructed, which contains the serial or parallel execution paths of each target agent. According to the task execution map, each target agent in the target agent combination is controlled to perform public opinion situation analysis on the target public opinion dataset to obtain a public opinion situation perception report based on the analysis requirements.

3. The public opinion situation perception method according to claim 1, characterized in that, The step of retrieving data from a preset multi-source database based on the aforementioned requirement parameters to obtain the target public opinion dataset includes: Based on the aforementioned demand parameters, data retrieval is performed on the multi-source database to obtain the various public opinion data. The aforementioned public opinion data are subjected to entity aggregation processing to obtain static tags that characterize the entity attributes of the aforementioned public opinion data; Based on the aforementioned requirement parameters, inference prompt words are constructed, and based on the inference prompt words, a large language model is driven to perform semantic inference on each public opinion data to obtain dynamic tags that characterize the semantic features of each public opinion data. Based on the aforementioned public opinion data and their corresponding static and dynamic tags, the target public opinion dataset is determined; The dynamic tags include at least one of the following: event aggregation tags for marking the same topic, information sentiment tags for characterizing emotional tendencies, risk level tags for assessing the degree of threat, and opinion conflict tags for identifying opposing viewpoints.

4. The public opinion situation perception method according to claim 3, characterized in that, The process of retrieving data from the multi-source database based on the aforementioned demand parameters to obtain the various public opinion data includes: Based on the situation type and demand tag in the demand parameters, the multi-source database is filtered to obtain the target database; Based on the situation type, demand tag and industry map in the demand parameters, the data in the target database is semantically filtered to obtain the target data range, and the data in the target database within the target data range are determined as the various public opinion data. The aforementioned public opinion data are data from the target database that are associated with the industry map and conform to the situation type and the demand tag.

5. The public opinion situation perception method according to any one of claims 1 to 4, characterized in that, The target intelligent agent includes at least one of the following: public opinion funnel intelligent agent, risk identification intelligent agent, key event identification intelligent agent, and industry map visualization intelligent agent; The step of invoking the target intelligent agent to perform public opinion situation analysis on the target public opinion dataset and obtaining a public opinion situation awareness report based on the analysis requirements includes: Based on the aforementioned public opinion funnel intelligent agent, the data retention ratio of each public opinion data in the target public opinion dataset is determined, and a data loss result is generated based on the data retention ratio. Based on the risk identification intelligent agent, the risk level label in the data tags of the target public opinion dataset is identified as high-risk public opinion data, and a risk evolution trend is generated based on the high-risk public opinion data; Based on the key event identification intelligent agent, the event aggregation tag in the data tag of the target public opinion dataset represents scattered public opinion data belonging to the same topic, and the scattered public opinion data is merged to generate the development process of key events; Based on the aforementioned industry map visualization intelligent agent, the entity interaction relationship in the static and dynamic tags of the data tags of each public opinion data is identified, and an industry chain impact map is constructed based on the entity interaction relationship. Based on at least one of the data loss results, the risk evolution trend, the development process of key events, and the industrial chain impact map, a public opinion situation awareness report is generated to meet the analysis needs.

6. The public opinion situation perception method according to claim 5, characterized in that, The process of generating a public opinion situation awareness report based on at least one of the data loss results, the risk evolution trend, the development process of key events, and the industrial chain impact map, including: Extract unstructured conclusion text and structured statistical data from at least one of the data loss results, the risk evolution trend, the development process of key events, and the industrial chain impact map; The unstructured conclusion texts output by each target agent are fused and summarized to generate an overall situational conclusion. The structured statistical data output by each target agent is converted into visual charts, and a public opinion situation awareness report is generated based on the visual charts and the overall situation conclusions to meet the analysis requirements.

7. The public opinion situation perception method according to any one of claims 1 to 4, characterized in that, The generation of requirement parameters based on the analysis requirements includes: The analysis requirements are analyzed to obtain the user's core focus and situational awareness intent; Based on the core focus subject and the situational awareness intent, the situation type, demand tags, and industry map are determined; when the core focus subject is an enterprise, the industry map includes the enterprise's industry, upstream enterprises, downstream enterprises, and competitors; The demand parameters are generated based on the situation type, the demand tag, and the industry map.

8. A public opinion situation awareness device, characterized in that, include: The parameter determination unit is used to determine the user's analysis requirements and generate requirement parameters based on the analysis requirements; The data retrieval unit is used to retrieve data from a preset multi-source database based on the required parameters to obtain a target public opinion dataset. Each public opinion data in the target public opinion dataset carries a data tag, and the data tag represents the entity attributes and semantic features of the corresponding public opinion data. The situation analysis unit is used to select a target intelligent agent from a set of preset intelligent agents based on the requirement parameters and the data tags, and call the target intelligent agent to perform public opinion situation analysis on the target public opinion dataset to obtain a public opinion situation perception report for the analysis requirements.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the public opinion situation perception method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the public opinion situation perception method as described in any one of claims 1 to 7.

Images