A method for multi-dimensional clustering and dynamic evaluation of crowd psychological risks for grassroots governance

Abstract

The application discloses a kind of group psychological risk multidimensional clustering and dynamic evaluation method for basic treatment, it is related to psychological risk clustering analysis technical field, including the following steps: S1, first, obtain group related text data, and set data sensitive interface, constantly collect data related to group, and then filter update;S2, text data is preprocessed, while extracting emotion features and semantic features therein;S3, the emotional elements and semantic elements are regarded as graph nodes, and the emotional correlation and semantic correlation between nodes are regarded as graph edges;S4, feature representation and similarity calculation are carried out on the emotion semantic graph, to generate feature vector for clustering analysis;Compared with the traditional method, the technical scheme of the application is not only more accurate in data processing, but also has strong adaptability, can automatically adjust the evaluation model according to the change of group emotion, to ensure efficient application in different situations.

Description

Technical Field

[0001] This invention relates to the field of psychological risk clustering analysis technology, specifically a multidimensional clustering and dynamic assessment method for group psychological risks used in grassroots governance. Background Technology

[0002] With the rapid development of information technology and the Internet, the popularization of social networks and digital platforms has made the monitoring and analysis of group psychological states increasingly important. Especially in the fields of mental health management, social public opinion monitoring, and public safety, the identification and intervention of group psychological risks has become an urgent technical problem to be solved. At present, many existing group psychological risk analysis methods mainly rely on human experience or simple emotion classification techniques, which have certain limitations in practical applications. With the continuous advancement of the modernization of rural grassroots governance systems and capabilities, the application of information technology and the internet in rural governance is becoming increasingly widespread. Grassroots governance is gradually showing a trend towards digitalization, networking, and platformization. Various social networks, government service platforms, and rural information systems have become important carriers reflecting the emotional demands and psychological states of grassroots communities. Effective identification and dynamic monitoring of group psychological risks are of significant practical importance for preventing and resolving grassroots social conflicts, maintaining social harmony and stability, and promoting safe and stable development. First, existing methods typically analyze emotional or semantic features independently, failing to effectively combine the relationship between the two. This results in low accuracy of emotion analysis and an inability to fully reflect the psychological state of a group. Second, existing emotion analysis methods often cannot effectively process data from multiple sources and multiple modalities, leading to inaccurate emotion assessment results and making it difficult to achieve comprehensive and timely identification of group psychological risks. Furthermore, traditional group psychological risk assessment methods usually rely on static data analysis and cannot capture the dynamic characteristics of group psychological changes in real time, limiting the timeliness and effectiveness of psychological intervention measures. Therefore, how to provide more effective technical support for group psychological risk analysis by combining emotional and semantic features in a more accurate and comprehensive way has become an urgent technical problem to be solved. Summary of the Invention

[0003] This invention provides a multidimensional clustering and dynamic assessment method for group psychological risk in grassroots governance, which can effectively solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a multidimensional clustering and dynamic assessment method for group psychological risk in grassroots governance, comprising the following steps: S1. First, obtain text data related to the group and set up a data-sensitive interface to continuously collect data related to the group, and then filter and update it. S2. Preprocess the text data and extract its emotional and semantic features. S3. Treat emotional and semantic elements as graph nodes and the emotional and semantic relationships between nodes as graph edges. S4. Perform feature representation and similarity calculation on the emotion semantic graph to generate feature vectors for cluster analysis; S5. Based on the preset clustering analysis algorithm, the feature vectors are clustered to obtain the psychological risk clustering results; S6 performs judgment processing based on the results of group psychological risk clustering to determine the corresponding group psychological risk status; S7. Process the data on the determined group psychological risk status to generate analysis results for group psychological risk monitoring or early warning.

[0005] According to the above technical solution, the emotional features in S2 include emotional category and emotional intensity; The semantic features include semantic elements and relationships in the text; In S2, during data preprocessing, the influence tendency of the data is divided into positive influence Ip, negative influence In, and undetermined influence Io. The process of extracting emotional features is constrained or modified based on the influence tendency. Positive influence Ip is used to enhance the weight of the corresponding emotional feature, negative influence In is used to weaken the weight of the corresponding emotional feature, and undetermined influence Io is used to retain or postpone the determination of emotional features. In step S2, the initial sentiment intensity values ​​extracted from the text data are corrected for influence bias constraints, and the corrected sentiment intensity is denoted as... The calculation method is as follows: in: E represents the initial emotional intensity value corresponding to the text; I∈{Ip, In, Io} represents the type of influence tendency of the text; This represents a modified weighting function based on the influence tendency; Corrected weight function Defined as: in: α>1 is used to enhance the emotional intensity corresponding to texts with positive influence; 0 < β < 1, used to weaken the emotional intensity corresponding to a negatively impacted text; When the text's influence tendency is undetermined influence At that time, the intensity of the emotion is not modified.

[0006] According to the above technical solution, step S1 involves extracting group-related text data and forming a raw dataset for subsequent processing, specifically including: Access data sources, which include one or more of the following: social media platform text, questionnaire hotline records, communication text, comment text, instant messaging group chat text, and work order feedback text; Data collection is performed according to a preset data collection strategy, including at least one of the following: data collection by time window, data collection by keyword, data collection by topic tag, data collection by group identifier, and data collection by geographic or organizational scope. To bind group identifiers to the collected text data, the group identifiers include at least one of the following: population category, organization, class, department, region, topic circle, or event set; Add metadata to each text record. The metadata shall include at least one of the following: timestamp, source identifier, group identifier, author anonymity identifier or session identifier, subject, topic identifier. The collected text and metadata are stored as a raw dataset and then output to the preprocessing module. In S1, the sensitive factors of the data sensitive interface are the group and time. By setting time nodes, the relevant data of the group before time node T is used as the search object and the data is continuously updated. After the data sensitive interface receives new data, the new data will execute steps S2-S7 again for analysis and optimization. The dataset with existing data D is denoted as D. t The data input to the data-sensitive interface is denoted as D. i , i represents the number of inputs, and offset sensitivity is introduced to assist in the judgment of data updates; For the i-th newly added data Di, its offset sensitivity relative to existing data is defined as: in, It is a distribution offset metric that measures the difference between two datasets and judges the degree of difference between new data and historical data. Here, JS divergence analysis is performed on the new data and historical data. The higher the correlation between the new data and historical data, the lower the distribution offset metric value. The value range of the distribution offset metric is [0,1]. It is the Sigmoid function; and These are the weighting coefficients; It is the offset trigger threshold; This represents the data input to the data-sensitive interface for the i-th time. This refers to the historical dataset collected before inputting data into the data-sensitive interface. This represents the entire dataset after the (i-1)th input to the data-sensitive interface.

[0007] According to the above technical solution, step S2 is used to clean and preprocess the text data, and extract emotional and semantic features, specifically including: Preprocessing unit: performs at least one of the following: deduplication, noise reduction, garbled character filtering, removal of meaningless characters, emoji normalization, linking, and image placeholder replacement; The text is processed by sentence segmentation, word segmentation, word form unification, or synonym merging. Remove stop words, colloquialisms, and meaningless repetitions; Masking or de-identifying sensitive information such as names, phone numbers, and addresses; The validity of the text is judged, including at least one of the following: minimum length threshold, repetition threshold, language category recognition, spam text recognition, and relevance to the target topic threshold. Texts judged as invalid are removed. The preprocessing unit also includes an emotion feature extraction unit and an emotion intensity calculation subunit; Calculate the emotion intensity score or interval level. The emotion intensity can be based on at least one of the following: emotion dictionary weighting, model output probability, emotion trigger word weighting, emphasis, negation, and degree adverb correction. Furthermore, group emotions are aggregated according to time windows to obtain window-level emotion distribution, mean intensity, peak intensity, and volatility indicators.

[0008] According to the above technical solution, step S3 transforms emotional features and semantic features into a graph structure, forming a relationship model of nodes and edges, specifically including: Node construction: Graph nodes are constructed based on emotion and semantic features. Emotion nodes include emotion category and intensity information, while semantic nodes include entity, event, and keyword elements. Edge construction: Graph edges are constructed based on the relationship between emotion and semantic elements. Edges represent the correlation between different elements, such as the co-occurrence of emotion and semantics, and dependency relationships. Graph Update: The sentiment semantic graph is dynamically updated based on new text data to ensure that the graph accurately reflects changes in group emotions and semantic relationships.

[0009] According to the above technical solution, step S4 converts the nodes and edges in the graph into feature vectors that can be used for computation, facilitating subsequent analysis and clustering. Specifically, this includes: Node vectorization: Emotion nodes and semantic nodes are vectorized. Emotion nodes can be converted into vectors through category encoding and intensity calculation, and semantic nodes can be generated into vectors through semantic embedding methods. Edge vectorization: Representing the edges in a graph using features such as relation type and similarity value; Feature fusion: The features of nodes and edges are merged into a unified feature vector and then standardized.

[0010] According to the above technical solution, step S5 involves clustering the feature vectors to identify different patterns of group psychological risk, specifically including: Feature input: The feature vector from the feature representation module is used as input to prepare for cluster analysis; Clustering algorithm selection: Use common clustering algorithms to classify feature vectors and identify different types of group psychological states; Clustering results output: Output the clustering label for each group, reflecting the distribution of group emotional characteristics and semantic relationships, and further judging the psychological risk level of the group.

[0011] According to the above technical solution, step S6 involves clustering the feature vectors to identify different patterns of group psychological risk, specifically including: Feature input: The feature vector from the feature representation module is used as input to prepare for cluster analysis.

[0012] Clustering algorithm selection: Use common clustering algorithms to classify feature vectors and identify different types of group psychological states.

[0013] Clustering results output: Output the clustering label for each group, reflecting the distribution of group emotional characteristics and semantic relationships, and further judging the psychological risk level of the group.

[0014] According to the above technical solution, step S7, which involves judging the psychological risk of a group based on the cluster analysis results, specifically includes: Threshold determination: The clustering results are compared with a preset risk threshold to determine whether the group is in a high-risk state.

[0015] Model assessment: The clustering results are input into a pre-trained psychological risk assessment model, and the psychological risk status of the group is assessed based on the model output.

[0016] Rule-based judgment: Determine the psychological risk level of the group based on predefined rules.

[0017] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention has a scientific and reasonable structure, is safe and convenient to use, and adopts a combination of emotional and semantic features. By constructing an emotional semantic graph, it improves the accuracy and efficiency of group psychological risk analysis. It can not only comprehensively extract the emotional information of the group, but also establish the relationship between emotional and semantic elements through the fusion of semantic features, thereby improving the accuracy and depth of emotional analysis. By constructing an emotional semantic graph, it can more accurately capture the dynamic changes of group emotions and avoid the limitations of single emotional analysis. In addition, it adopts multi-source data fusion technology, which can integrate information from multiple data sources such as social platforms, questionnaires, and comment texts, making emotional analysis more comprehensive and data richer, effectively solving the problem of incomplete emotional data. This method uses a clustering analysis algorithm to conduct real-time psychological risk assessment of a group, and judges and warns of the group's psychological risk status based on the clustering results, further improving the response speed and effectiveness of mental health monitoring. Compared with traditional methods, the technical solution of this invention is not only more accurate in data processing, but also has strong adaptability. It can automatically adjust the evaluation model according to changes in group emotions, ensuring efficient application in different situations. Ultimately, this invention can be widely applied in various fields such as group mental health monitoring, social sentiment analysis, and public safety early warning, especially in the areas of group crisis intervention and mental health management, demonstrating great social value and application prospects. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0019] In the attached diagram: Figure 1 This is a flowchart of the overall method of the present invention; Figure 2 This is a schematic diagram of the data-sensitive interface and distributed offset triggering mechanism of the present invention; Figure 3 This is a schematic diagram of the emotional influence tendency constraint and emotional intensity correction of the present invention; Figure 4 This is a schematic diagram of the emotion semantic graph and psychological risk clustering of the present invention. Detailed Implementation

[0020] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0021] Example: Figures 1-4As shown, the present invention provides a technical solution for multidimensional clustering and dynamic assessment of group psychological risk in grassroots governance, comprising the following steps: S1. First, obtain text data related to the group and set up a data-sensitive interface to continuously collect data related to the group, and then filter and update it. S2. Preprocess the text data and extract its emotional and semantic features. S3. Treat emotional and semantic elements as graph nodes and the emotional and semantic relationships between nodes as graph edges. S4. Perform feature representation and similarity calculation on the emotion semantic graph to generate feature vectors for cluster analysis; S5. Based on the preset clustering analysis algorithm, the feature vectors are clustered to obtain the psychological risk clustering results; S6 performs judgment processing based on the results of group psychological risk clustering to determine the corresponding group psychological risk status; S7. Process the data on the determined group psychological risk status to generate analysis results for group psychological risk monitoring or early warning.

[0022] According to the above technical solution, the emotional features in S2 include emotional category and emotional intensity; Semantic features include semantic elements and relationships within the text; In S2, during data preprocessing, the influence tendency of the data is divided into positive influence Ip, negative influence In, and undetermined influence Io; The process of extracting emotional features is constrained or modified based on the influence tendency. Positive influence Ip is used to enhance the weight of the corresponding emotional feature, negative influence In is used to weaken the weight of the corresponding emotional feature, and undetermined influence Io is used to retain or postpone the determination of emotional features. In step S2, the initial sentiment intensity values ​​extracted from the text data are corrected for influence bias constraints, and the corrected sentiment intensity is denoted as... The calculation method is as follows: in: E represents the initial emotional intensity value corresponding to the text; I∈{Ip, In, Io} represents the type of influence tendency of the text; This represents a modified weighting function based on the influence tendency; Corrected weight function Defined as: in: α>1 is used to enhance the emotional intensity corresponding to texts with positive influence; 0 < β < 1, used to weaken the emotional intensity corresponding to a negatively impacted text; When the text's influence tendency is undetermined influence At that time, the intensity of the emotion is not modified.

[0023] According to the above technical solution, S1 extracts group-related text data and forms a raw dataset for subsequent processing, specifically including: Access data sources, which include one or more of the following: social media platform text, questionnaire hotline records, communication text, comment text, instant messaging group chat text, and work order feedback text; Data collection is performed according to a preset data collection strategy, including at least one of the following: data collection by time window, data collection by keyword, data collection by topic tag, data collection by group identifier, and data collection by geographic or organizational scope. To bind group identifiers to the collected text data, the group identifiers include at least one of the following: population category, organization, class, department, region, topic circle, or event set; Add metadata to each text record. The metadata shall include at least one of the following: timestamp, source identifier, group identifier, author anonymity identifier or session identifier, subject, topic identifier. The collected text and metadata are stored as a raw dataset and then output to the preprocessing module. In S1, the sensitive factors of the data-sensitive interface are the group and time. By setting time nodes, the relevant data of the group before time node T is used as the search object and the data is continuously updated. After the data-sensitive interface receives new data, the new data will execute the steps of S2-S7 again for analysis and optimization. The dataset with existing data D is denoted as D. t The data input to the data-sensitive interface is denoted as D. i , i represents the number of inputs, and offset sensitivity is introduced to assist in the judgment of data updates; For the i-th newly added data Di, its offset sensitivity relative to existing data is defined as: in, It is a distribution offset metric that measures the difference between two datasets and judges the degree of difference between new data and historical data. Here, JS divergence analysis is performed on the new data and historical data. The higher the correlation between the new data and historical data, the lower the distribution offset metric value. The value range of the distribution offset metric is [0,1]. It is the Sigmoid function; and These are the weighting coefficients; It is the offset trigger threshold; This represents the data input to the data-sensitive interface for the i-th time. This refers to the historical dataset collected before inputting data into the data-sensitive interface. This represents the entire dataset after the (i-1)th input to the data-sensitive interface; Analysis of offset sensitivity reveals that data is not necessarily more important the more similar it is; rather, the greater the offset, the more worthwhile it is to re-analyze. This aligns with the psychological risk logic analysis of emotional semantics. Analyzing new data and data with low relevance to the original data—data that is easily overlooked—can provide new references for emotional semantics analysis, thus making the analysis more comprehensive and accurate.

[0024] According to the above technical solution, S2 is used to clean and preprocess the text data, and extract sentiment features and semantic features, specifically including: Preprocessing unit: performs at least one of the following: deduplication, noise reduction, garbled character filtering, removal of meaningless characters, emoji normalization, linking, and image placeholder replacement; The text is processed by sentence segmentation, word segmentation, word form unification, or synonym merging. Remove stop words, colloquialisms, and meaningless repetitions; Masking or de-identifying sensitive information such as names, phone numbers, and addresses; The validity of the text is judged, including at least one of the following: minimum length threshold, repetition threshold, language category recognition, spam text recognition, and relevance to the target topic threshold. Texts judged as invalid are removed. The preprocessing unit also includes an emotion feature extraction unit and an emotion intensity calculation subunit; The emotion category recognition subunit is used to output emotion category labels. Emotion categories include, but are not limited to, a multi-category system of positive, negative, neutral, angry, anxious, fearful, sad, disgusted, and joyful emotions. Calculate the emotion intensity score or interval level. The emotion intensity can be based on at least one of the following: emotion dictionary weighting, model output probability, emotion trigger word weighting, emphasis, negation, and degree adverb correction. Furthermore, the group's emotions are aggregated according to time windows to obtain indicators such as window-level emotion distribution, mean intensity, peak value, and volatility. The preprocessing unit also includes a semantic element extraction subunit: extracting semantic elements from the text, including at least one of entities, keywords, phrases, event elements, topic words, and topic tags; Relation extraction subunit: Extracts the relationships between semantic elements. Relationships include at least one of the following: co-occurrence relationship, dependency relationship, causal relationship, referential relationship, event chain relationship, and topic similarity relationship. Semantic representation subunit: Represents semantic elements as vectors or symbols, including at least one of word / sentence vectors, topic distribution vectors, and event structured representation.

[0025] According to the above technical solution, S3 transforms emotional and semantic features into a graph structure, forming a relationship model of nodes and edges, specifically including: Node construction: Graph nodes are constructed based on emotion and semantic features. Emotion nodes include emotion category and intensity information, while semantic nodes include entity, event, and keyword elements. Edge construction: Graph edges are constructed based on the relationship between emotion and semantic elements. Edges represent the correlation between different elements, such as the co-occurrence of emotion and semantics, and dependency relationships. Graph Update: The sentiment semantic graph is dynamically updated based on new text data to ensure that the graph accurately reflects changes in group emotions and semantic relationships.

[0026] According to the above technical solution, in S4, the nodes and edges in the graph are converted into feature vectors that can be used for computation, facilitating subsequent analysis and clustering. Specifically, this includes: Node vectorization: Emotion nodes and semantic nodes are vectorized. Emotion nodes can be converted into vectors through category encoding and intensity calculation, and semantic nodes can be generated into vectors through semantic embedding methods. Edge vectorization: Representing the edges in a graph using features such as relation type and similarity value; Feature fusion: The features of nodes and edges are merged into a unified feature vector and then standardized.

[0027] Based on the above technical solution, S5 performs cluster analysis on the feature vectors to identify different patterns of group psychological risk, specifically including: Feature input: The feature vector from the feature representation module is used as input to prepare for cluster analysis; Clustering algorithm selection: Use common clustering algorithms (such as K-means, DBSCAN, etc.) to classify feature vectors and identify different types of group psychological states; Clustering results output: Output the clustering label for each group, reflecting the distribution of group emotional characteristics and semantic relationships, and further judging the psychological risk level of the group.

[0028] Based on the above technical solution, S6 performs cluster analysis on the feature vectors to identify different patterns of group psychological risk, specifically including: Feature input: The feature vector from the feature representation module is used as input to prepare for cluster analysis.

[0029] Clustering algorithm selection: Use common clustering algorithms (such as K-means, DBSCAN, etc.) to classify feature vectors and identify different types of group psychological states.

[0030] Clustering results output: Output the clustering label for each group, reflecting the distribution of group emotional characteristics and semantic relationships, and further judging the psychological risk level of the group.

[0031] According to the above technical solution, S7 assesses the psychological risk of a group based on cluster analysis results, specifically including: Threshold determination: The clustering results are compared with a preset risk threshold to determine whether the group is in a high-risk state.

[0032] Model assessment: The clustering results are input into a pre-trained psychological risk assessment model, and the psychological risk status of the group is assessed based on the model output.

[0033] Rule-based judgment: Determine the psychological risk level of the group based on predefined rules.

[0034] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multidimensional clustering and dynamic assessment method for group psychological risk in grassroots governance, characterized in that: Includes the following steps: S1. First, obtain text data related to the group and set up a data-sensitive interface to continuously collect data related to the group, and then filter and update it. S2. Preprocess the text data and extract its emotional and semantic features. S3. Treat emotional and semantic elements as graph nodes and the emotional and semantic relationships between nodes as graph edges. S4. Perform feature representation and similarity calculation on the emotion semantic graph to generate feature vectors for cluster analysis; S5. Based on the preset clustering analysis algorithm, the feature vectors are clustered to obtain the psychological risk clustering results; S6 performs judgment processing based on the results of group psychological risk clustering to determine the corresponding group psychological risk status; S7. Process the data on the determined group psychological risk status to generate analysis results for group psychological risk monitoring or early warning.

2. The method for multidimensional clustering and dynamic assessment of group psychological risk for grassroots governance as described in claim 1, characterized in that, The emotional features in S2 include emotional category and emotional intensity; The semantic features include semantic elements and relationships in the text; In S2, during data preprocessing, the influence tendency of the data is divided into positive influence Ip, negative influence In, and undetermined influence Io. The process of extracting emotional features is constrained or modified based on the influence tendency. Positive influence Ip is used to enhance the weight of the corresponding emotional feature, negative influence In is used to weaken the weight of the corresponding emotional feature, and undetermined influence Io is used to retain or postpone the determination of emotional features. In step S2, the initial sentiment intensity values ​​extracted from the text data are corrected for influence bias constraints, and the corrected sentiment intensity is denoted as... The calculation method is as follows: in: E represents the initial emotional intensity value corresponding to the text; I∈{Ip, In, Io} represents the type of influence tendency of the text; This represents a modified weighting function based on the influence tendency; Corrected weight function Defined as: in: α>1 is used to enhance the emotional intensity corresponding to texts with positive influence; 0 < β < 1, used to weaken the emotional intensity corresponding to a negatively impacted text; When the text's influence tendency is undetermined influence At that time, the intensity of the emotion is not modified.

3. The method for multidimensional clustering and dynamic assessment of group psychological risk for grassroots governance as described in claim 1, characterized in that, The S1 step involves extracting group-related text data and forming a raw dataset for subsequent processing, specifically including: Access data sources, which include one or more of the following: social media platform text, questionnaire hotline records, communication text, comment text, instant messaging group chat text, and work order feedback text; Data collection is performed according to a preset data collection strategy, including at least one of the following: data collection by time window, data collection by keyword, data collection by topic tag, data collection by group identifier, and data collection by geographic or organizational scope. To bind group identifiers to the collected text data, the group identifiers include at least one of the following: population category, organization, class, department, region, topic circle, or event set; Add metadata to each text record. The metadata shall include at least one of the following: timestamp, source identifier, group identifier, author anonymity identifier or session identifier, subject, topic identifier. The collected text and metadata are stored as a raw dataset and then output to the preprocessing module. In S1, the sensitive factors of the data sensitive interface are the group and time. By setting time nodes, the relevant data of the group before time node T is used as the search object and the data is continuously updated. After the data sensitive interface receives new data, the new data will execute steps S2-S7 again for analysis and optimization. The dataset with existing data D is denoted as D. t The data input to the data-sensitive interface is denoted as D. i , i represents the number of inputs, and offset sensitivity is introduced to assist in the judgment of data updates; For the i-th newly added data Di, its offset sensitivity relative to existing data is defined as: in, It is a distribution offset metric that measures the difference between two datasets and judges the degree of difference between new data and historical data. Here, JS divergence analysis is performed on the new data and historical data. The higher the correlation between the new data and historical data, the lower the distribution offset metric value. The value range of the distribution offset metric is [0,1]. It is the Sigmoid function; and These are the weighting coefficients; It is the offset trigger threshold; This represents the data input to the data-sensitive interface for the i-th time. This refers to the historical dataset collected before inputting data-sensitive interfaces. This represents the entire dataset after the (i-1)th input to the data-sensitive interface.

4. The method for multidimensional clustering and dynamic assessment of group psychological risk for grassroots governance according to claim 1, characterized in that, S2 is used to clean and preprocess the text data, and extract emotional and semantic features, specifically including: Preprocessing unit: performs at least one of the following: deduplication, noise reduction, garbled character filtering, removal of meaningless characters, emoji normalization, linking, and image placeholder replacement; The text is processed by sentence segmentation, word segmentation, word form unification, or synonym merging. Remove stop words, colloquialisms, and meaningless repetitions; Masking or de-identifying sensitive information such as names, phone numbers, and addresses; The validity of the text is judged, including at least one of the following: minimum length threshold, repetition threshold, language category recognition, spam text recognition, and relevance to the target topic threshold. Texts judged as invalid are removed. The preprocessing unit also includes an emotion feature extraction unit and an emotion intensity calculation subunit; Calculate the emotion intensity score or interval level. The emotion intensity can be based on at least one of the following: emotion dictionary weighting, model output probability, emotion trigger word weighting, emphasis, negation, and degree adverb correction. Furthermore, group emotions are aggregated according to time windows to obtain window-level emotion distribution, mean intensity, peak intensity, and volatility indicators.

5. The method for multidimensional clustering and dynamic assessment of group psychological risk for grassroots governance according to claim 1, characterized in that, In S3, emotional and semantic features are transformed into a graph structure, forming a relationship model between nodes and edges, specifically including: Node construction: Graph nodes are constructed based on emotion and semantic features. Emotion nodes include emotion category and intensity information, while semantic nodes include entity, event, and keyword elements. Edge construction: Graph edges are constructed based on the relationship between emotion and semantic elements. Edges represent the correlation between different elements, such as the co-occurrence of emotion and semantics, and dependency relationships. Graph Update: The sentiment semantic graph is dynamically updated based on new text data to ensure that the graph accurately reflects changes in group emotions and semantic relationships.

6. The method for multidimensional clustering and dynamic assessment of group psychological risk for grassroots governance according to claim 1, characterized in that, In step S4, the nodes and edges in the graph are converted into feature vectors that can be used for computation, facilitating subsequent analysis and clustering. Specifically, this includes: Node vectorization: Emotion nodes and semantic nodes are vectorized. Emotion nodes can be converted into vectors through category encoding and intensity calculation, and semantic nodes can be generated into vectors through semantic embedding methods. Edge vectorization: Representing the edges in a graph using features such as relation type and similarity value; Feature fusion: The features of nodes and edges are merged into a unified feature vector and then standardized.

7. The method for multidimensional clustering and dynamic assessment of group psychological risk for grassroots governance according to claim 1, characterized in that, In step S5, cluster analysis is performed on the feature vectors to identify different patterns of group psychological risk. include: Feature input: The feature vector from the feature representation module is used as input to prepare for cluster analysis; Clustering algorithm selection: Use common clustering algorithms to classify feature vectors and identify different types of group psychological states; Clustering results output: Output the clustering label for each group, reflecting the distribution of group emotional characteristics and semantic relationships, and further judging the psychological risk level of the group.

8. The method for multidimensional clustering and dynamic assessment of group psychological risk for grassroots governance according to claim 1, characterized in that, In step S6, cluster analysis is performed on the feature vectors to identify different patterns of group psychological risk. include: Feature input: The feature vector from the feature representation module is used as input to prepare for cluster analysis. Clustering algorithm selection: Use common clustering algorithms to classify feature vectors and identify different types of group psychological states. Clustering results output: Output the clustering label for each group, reflecting the distribution of group emotional characteristics and semantic relationships, and further judging the psychological risk level of the group.

9. A multidimensional clustering and dynamic assessment method for group psychological risk in grassroots governance as described in claim 1, characterized in that, The S7 step involves assessing the psychological risk of a group based on cluster analysis results, specifically including: Threshold determination: The clustering results are compared with a preset risk threshold to determine whether the group is in a high-risk state. Model assessment: The clustering results are input into a pre-trained psychological risk assessment model, and the psychological risk status of the group is assessed based on the model output. Rule-based judgment: Determine the psychological risk level of the group based on predefined rules.

Images