Intelligent government hotline group appeal analysis and detection method

Abstract

The application provides a kind of intelligent government affair hotline mass appeal analysis, detection method, the method includes the following contents: pull daily work order data regularly;The work order content in work order data is vectorized coding;For vectorized work order data, using DBSCAN clustering algorithm, obtain classification cluster;Traversal classification cluster, utilize the address feature of classification cluster, complaint feature, with the data not included in any cluster, again using DBSCAN clustering algorithm expands classification cluster, for obtaining more accurate complaint quantity and complaint number information;Obtain the complaint summary of each classification cluster through large model;Data is written into data.The application can realize mass appeal detection considering semantics and spatial information, replace the redundancy, repetitive work of original manual statistics in an intelligent way, can effectively help to identify and integrate similar appeals, so as to improve the processing efficiency, optimize the service quality, quickly and efficiently provide daily mass event list.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of big data analysis, in particular to an intelligent government hotline group appeal analysis and detection method. BACKGROUND

[0002] According to the prior art such as patent document CN116805207A, a citizen public opinion early warning analysis method and device and storage medium are provided, which analyzes the public opinion event corresponding to each work order, constructs a public opinion early warning model corresponding to different event types, and triggers public opinion early warning for the public opinion event corresponding to each event type when the number of reports and complaints corresponding to each event type monitored based on the public opinion early warning model reaches a preset trigger threshold.

[0003] However, the above-mentioned prior art is based on different event types for public opinion early warning, that is, under the same event type, there is only one corresponding public opinion level warning. For example, property dispute problems, but the complaints are scattered in various communities in the city, and if group appeal is detected, there should be a certain spatial constraint.

[0004] On the other hand, in the existing scheme, the processing method of the hotline mainly depends on the call center operator to manually summarize and count the daily group appeal, however, this traditional manual statistical method not only consumes time and effort, but also easily causes delay in information processing, affecting the timely response to urgent problems. Therefore, the existing method is based on a pre-set complaint library or does not consider spatial information, which is too broad. The above-mentioned prior art is not conducive to law enforcement and problem solving. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application aims to provide an intelligent government hotline group appeal analysis and detection method to solve the problems raised in the background art. The present application can realize group appeal detection considering both semantic and spatial information, replace the redundant and repetitive work of manual statistics through intelligent means, effectively help identify and integrate similar appeals, thereby improving processing efficiency, optimizing service quality, and realizing the provision of a daily group event list quickly and efficiently.

[0006] In order to achieve the above-mentioned purpose, the present application is realized by the following technical scheme: an intelligent government hotline group appeal analysis and detection method, including work order information keyword extraction and group appeal detection, the work order information keyword extraction includes the following steps:

[0007] S1.1, periodically pull unanalyzed work order data;

[0008] S1.2, batch analyze work order data;

[0009] S1.3, model result analysis;

[0010] S1.4, result storage;

[0011] The mass appeal detection comprises the following steps:

[0012] S2.1, pull the work order data of the day in time, respectively pull the work order data from the business database and pull the analysis data associated with the work order from es;

[0013] S2.2, vector coding of the work order content in the work order data;

[0014] S2.3, work order data clustering;

[0015] S2.4, summarize the group complaint event, and count the number of complaints and the number of complainants;

[0016] S2.5, store the group complaint event in the database.

[0017] Further, in the process of extracting keywords of the work order information, the work order content and the work order detailed address of the unanalyzed work order are obtained by pulling the unanalyzed work order data; the process of batch analyzing the work order data comprises: inputting the work order content and the work order detailed address; constructing a large language model input; calling the interface deployed by glm4-9b-chat model in batches to obtain the result.

[0018] Further, the model result analysis process comprises parsing the keywords in the output json format text string, and storing the parsed result in Elasticsearch.

[0019] Further, in step S2.2, the open source text2vec-base-chinese model is used for vectorization processing, in which the content and address of each work order are first input into the model, the corresponding vector representation is obtained through the output layer of the BERT model, and the text feature vector is used for subsequent similarity calculation, classification or other tasks; through vectorization, the semantic information of the work order content and address is retained and converted into a numerical format that can be processed by a computer.

[0020] Further, the work order data clustering process in step S2.3 specifically comprises the following contents:

[0021] S2.3.1, calculate the cosine similarity between the work order content vectors;

[0022] S2.3.2, calculate the cosine similarity between the work order address vectors;

[0023] S2.3.3, the similarity between the work orders is calculated by weighting, the semantic similarity accounts for 0.5, and the spatial similarity accounts for 0.5, to obtain the work order similarity, and the value range is 0-1, and the greater the value is, the more similar it is;

[0024] S2.3.4, the work order similarity is converted to obtain a work order similarity distance, and the similarity distance ranges from 0 to 1, and the smaller the value is, the closer the distance is;

[0025] S2.3.5, the dbscan algorithm is run, and eps is set to 0.05 and num_samples is set to 5;

[0026] S2.3.6, the classification cluster result of vectorization clustering calculation is obtained;

[0027] S2.3.7, the data not marked is cleaned, and specific classification data is filtered out;

[0028] S2.3.8, keyword clustering, the keywords are divided into object class keywords and complaint reason keywords;

[0029] S2.3.9, the clustering results are fused, the clustering results of step 2.3.6 are fused with the result list of 2.3.8.3, to obtain a basic group complaint event list;

[0030] S2.3.10, the group complaint list data is expanded.

[0031] Further, the similarity of two vectors is judged by calculating the cosine value of the included angle, the value range is from-1 to 1, wherein 1 represents complete similarity, 0 represents no relation, and-1 represents complete opposite, and the calculation formula is:

[0032]

[0033] Further, the dbscan algorithm step includes: selecting an unvisited point from the data set, judging whether it is a core point; if it is a core point, all density-reachable points in its neighborhood are classified into the same cluster; for a non-core point, if it is adjacent to a certain core point, it becomes part of the cluster, otherwise it is marked as noise.

[0034] Further, in step S2.3.8, the keywords are divided, the keywords are vectorized, and then the object class keywords are clustered based on a large model, the number of traversed object class keywords satisfies a threshold, and finally it is judged by the large model whether the several most similar words point to the same entity.

[0035] Further, the obtained object class keyword classification cluster is traversed, the reason class keywords corresponding to the data in the cluster are clustered, the dbscan clustering method is used, the eps parameter is set to 0.3, the min_samples parameter is 5, the clustering result is traversed again, and the result of the keyword clustering step is obtained.

[0036] Further, in the expansion stage of the group lawsuit list data, first, the unclassified data is screened by using the classification cluster feature, and the semantic similarity of the unclassified data with the classification cluster feature is greater than 0.8; then, the classification cluster space feature is screened to obtain to-be-screened data that meets the semantic and spatial similarity threshold values; then, the to-be-screened data is mixed with the classified data, and the dbscan clustering is used to gradually increase the eps parameter and the min_samples parameter until the classified data is re-clustered and still belongs to the same category, at this time, the data of the same category as the classified data is regarded as expansion data.

[0037] The beneficial effects of the application are as follows:

[0038] 1. The intelligent government hotline group appeal analysis and detection method no longer needs to arrive at a fixed time every day to obtain the group event list of the day, and the group detection data and the actual data only have a delay of half an hour, so that the method has high near real-time performance.

[0039] 2. The intelligent government hotline group appeal analysis and detection method can replace the traditional manual statistical method, and adopts a more intelligent and automatic statistical method, and the method can not only statistically analyze the group lawsuit event information of the day, but also can be used to statistically analyze the group event information of the week, the month and the season. Different dimensions of group event analysis results are provided.

[0040] 3. The intelligent government hotline group appeal analysis and detection method is more comprehensive, can more effectively and accurately judge the group complaint event, and display the group complaint event in the group event list, and can obtain data from multiple channels. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 A work order information keyword extraction schematic diagram provided by the application is provided.

[0042] Figure 2 A work order keyword extraction case schematic diagram provided by the application is provided.

[0043] Figure 3 A group lawsuit detection total flow schematic diagram provided by the application is provided.

[0044] Figure 4 Work order data clustering provided by the application is provided.

[0045] Figure 5 A keyword clustering schematic diagram provided by the application is provided. DETAILED DESCRIPTION

[0046] In order to make the technical means, creative features, purposes and effects of the present application easy to understand, the present application will be further described below in conjunction with specific embodiments.

[0047] Please refer to Figures 1 to 5 The present application provides the following technical solutions: an intelligent government hotline group appeal analysis and detection method, which mainly includes the following contents:

[0048] 1. Pull daily work order data at regular intervals

[0049] 2. Vectorize and encode the work order content in the work order data

[0050] 3. Use the DBSCAN clustering algorithm for vectorized work order data to obtain classified clusters

[0051] 4. Traverse the classified clusters, use the address features and complaint features of the classified clusters, and use the DBSCAN clustering algorithm again to expand the classified clusters (the purpose is to obtain more accurate complaint volume and number of complainants information)

[0052] 5. Obtain the complaint summary of each classified cluster through a large model

[0053] 6. Write data to data.

[0054] Through the above method, it is no longer necessary to reach a fixed time every day to obtain the group event list of the day, and the group detection data and the actual data only have a delay of half an hour, which has high near real-time performance, can more effectively and accurately judge the group complaint event, and display it in the group event list, and can obtain data from multiple channels.

[0055] In this embodiment, the following technical process is also provided to explain the above method, which includes two parts of work order information keyword extraction and mass appeal detection, wherein:

[0056] The work order keyword extraction process schematic diagram is as shown in Figure 1 The work order keyword extraction case schematic diagram is as shown in Figure 2 The specific process of work order information keyword extraction is as follows:

[0057] Pull unanalyzed work order data at regular intervals

[0058] 1.1 Obtain the work order content of the unanalyzed work order, and the detailed address of the work order

[0059] 1.2. Batch analyze work order data

[0060] 1.2.1. Input the work order content, and the detailed address of the work order.

[0061] 1.2.2. Constructing large language model input

[0062] 1.2.3. Batch calling the interface deployed by glm4-9b-chat model (an open source large language model) to obtain results

[0063] 1.3. Model result analysis

[0064] According to the output json format text string, the keywords in it are parsed

[0065] 1.4. Result storage

[0066] The parsed results are stored in Elasticsearch (a database, a way to store and retrieve data)

[0067] The embodiment also provides a mass appeal detection method, wherein a mass appeal detection flowchart is as shown in Figure 3 , a work order data clustering flowchart is as shown in Figure 4 , and a keyword clustering flowchart is as shown in Figure 5 , and the specific detection process comprises the following steps:

[0068] 2.1 Pull the work order data of the day at regular intervals

[0069] Pull the work order data from the business database and the analysis data (keyword data) associated with the work order from es respectively

[0070] 2.2 Use the open source text2vec-base-chinese model to vectorize the work order content and work order address respectively

[0071] Vectorization is the process of converting text into fixed-length vectors, aiming to convert text data into numerical format that can be processed by models. In natural language processing (NLP), the goal of vectorization is to retain the semantic information in the text in the numerical space, facilitating subsequent calculation and analysis.

[0072] BERT (Bidirectional Encoder Representations from Transformers) is based on the Transformer architecture and learns contextual information from text through bidirectional training. BERT uses two pre-training tasks: Masked Language Model (MLM) and Next Sentence Prediction (NSP), enabling the model to understand the meaning of words within their context. Unlike traditional unidirectional language models (such as GPT), BERT captures richer semantic information by simultaneously considering both left and right contexts within the text.

[0073] `text2vec-base-chinese` is a BERT model specifically designed for Chinese text. It utilizes a large Chinese corpus during pre-training, enabling it to effectively handle the syntax and semantics of Chinese text. This model can be used to convert work order content and addresses into vectors. Specifically, the content and address of each work order are first input into the model, and the corresponding vector representations are obtained through the output layer of the BERT model. These vectors can then be used as feature vectors for subsequent similarity calculations, classification, or other tasks.

[0074] Vectorization preserves the semantic information of the work order content and address, converting it into a numerical format that computers can process, thus providing a foundation for further analysis and prediction. For example, the text: "Hello".

[0075] Vectorization is:

[0076] [-0.2139309197664261,0.3746878206729889,0.6576337814331055,.....,0.2119288593530655,-0.9286743402481079,-0.12038318812847137](Length 648)

[0077] 2.3 Work order data clustering

[0078] 2.3.1 Calculate the cosine similarity (semantic similarity) between work order content vectors.

[0079] Cosine similarity is a method for measuring the similarity between two vectors, often used in text analysis. It determines their similarity by calculating the cosine of the angle between the two vectors, with values ​​ranging from -1 to 1, where 1 indicates perfect similarity, 0 indicates no similarity, and -1 indicates complete opposites.

[0080] The calculation formula is:

[0081]

[0082] For example: Vector A [1,2,3] Vector B [4,5,6]

[0083] Calculate similarity:

[0084] A·B(dot product of vector A and B) = 1*4 + 2*5 + 3*6 = 32

[0085]

[0086] Similarity: 32 / 32.8 ≈ 0.975

[0087] Vector A and B have similar growth trends, and the overall similarity is calculated to be very similar.

[0088] 2.3.2 Calculate the cosine similarity between the address vectors of the work order (spatial similarity)

[0089] 2.3.3 Calculate the similarity between work orders, with semantic similarity accounting for 0.5 and spatial similarity accounting for 0.5, to get the work order similarity (value range 0-1, the larger the better)

[0090] For example: Semantic similarity is 0.9, spatial similarity is 0.8, then the final similarity between work orders is:

[0091] 0.9*0.5 + 0.8*0.5 = 0.85

[0092] 2.3.4 Use 1-work order similarity to convert to get work order similarity distance (0-1, the smaller the distance, the closer)

[0093] The following clustering algorithm is based on distance (the more similar, the closer the distance), and the current indicator is similarity (the more similar, the closer to 1), so a conversion is needed.

[0094] 2.3.5 Run the dbscan algorithm, set eps to 0.05 (within a certain distance, consider similar, same cluster), num_samples to 5 (how many similar together to consider a cluster).

[0095] DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is a density-based clustering algorithm. It forms clusters by connecting points of the same density, without specifying the number of clusters, and is suitable for processing data with noise and irregular shapes.

[0096] The main parameters of DBSCAN are:

[0097] - eps (maximum distance of a neighborhood): specifies a radius of a point's neighborhood. Two points are considered as in the same cluster if the distance between them is less than `eps`.

[0098] - min_samples (minimum number of points in a neighborhood to be considered as a core point): a point is considered as a core point if the number of points in its neighborhood is greater than or equal to `min_samples`.

[0099] Algorithm steps:

[0100] 1. Select an unvisited point from the dataset and determine whether it is a core point.

[0101] 2. If it is a core point, all density-reachable points in its neighborhood are assigned to the same cluster.

[0102] 3. For non-core points, if they are adjacent to a core point, they are part of the cluster, otherwise they are labeled as noise.

[0103] By setting eps = 0.05 and min_samples = 5, DBSCAN will cluster points that are close to each other into a cluster, and only sufficiently dense points can form a valid cluster, while sparse points will be considered as noise.

[0104] For example, if there are 1-10 work orders, using the dbscan clustering algorithm with parameters eps 0.15 and min_samples 3, the result is: [0, 0, 0, -1, -1, -1, 1, 1, 1, 1]

[0105] This means that work orders 1-3 are a cluster, 4-6 are unmarked work orders, and 7-10 are a cluster. The classification cluster here can be considered as the group event we expect to get, because the group event we expect to find is one that has consistent reasons for complaint and consistent objects of complaint, and we expect to aggregate them to represent the intensity of the public's demands. Therefore, they themselves have a high similarity value when calculating similarity between each other, while other work orders that do not have consistent reasons for complaint or objects of complaint will have a low similarity value. Therefore, the dbscan clustering algorithm can be used to efficiently find group events. In addition, the min_samples parameter can be considered as the minimum number of complaints that a group event must represent. The number of work orders in the classification cluster must be greater than or equal to the min_samples value, which acts as a threshold.

[0106] 2.3.6 Get the classification cluster result of vectorized clustering calculation (basic group event list);

[0107] 2.3.7 Clean up unmarked data and filter out specific classification data.

[0108] Such as: education training classification, consumer rights classification, etc.

[0109] Because of these special classification data, the complaint object may be consistent, but the complaint reason is not the same,

[0110] The final appeal may be consistent, resulting in that such events cannot be well detected based on semantics alone, and some company names are close, only one word apart, using vectorized calculation distance is not appropriate, which will cause some companies to be mixed together, which is not conducive to law enforcement departments to view and handle.

[0111] In the education and training classification, such situations often occur, such as the complaint company is the same, but because of different reasons, such as one is because of being scammed for applying for a social worker certificate, and the other is because of being scammed for reporting an upgrade of education level. The core appeal of both is to refund. In this complex situation, the appeal work order similarity clustering method is not applicable.

[0112] 2.3.8 Keyword clustering

[0113] Keywords are divided into two categories, one is object class keywords such as community, company name, and institution name, and the other is complaint reason class keywords.

[0114] 2.3.8.1 Divide keywords and keyword vectorization

[0115] 2.3.8.2 Object class keyword clustering based on large models

[0116] 2.3.8.2.1 Traverse the object class keywords that meet the threshold number (such as: object class keywords that appear more than 3 times)

[0117] 2.3.8.2.2 Use large models to determine whether the top 5 most similar words point to the same entity.

[0118] Example:

[0119] Construct the following large model input:

[0120] Determine whether the complaint companies in the following two events are the same:

[0121] Complaint A: In 2023, I enrolled in a social worker certificate course at Jiaoxue Education, paid 2700 yuan,

[0122] and promised to pass, but finally failed the exam and demanded a refund.

[0123] Complaint B: I previously enrolled in an adult college course at Wuhan Jiaoxue Education Technology Co., Ltd., paid 4500 yuan, and promised to have tutorials and pass, but no tutorials were provided, and demanded immediate refund.

[0124] Reason: xxx

[0125] Conclusion: 1 / 0

[0126] Where the red part is dynamically changed, and the large model can better judge the object class

[0127] Whether the keywords belong to the same entity, get more accurate clustering results.

[0128] Through the large model, it is determined that Jule Education and Wuhan Jule Education Technology Co., Ltd. are actually the same entity, and Jule Education is the abbreviation.

[0129] The reason for not using vector similarity here is that the effect is poor. Although the abbreviation and full name have the same four words of Jule Education, the Wuhan, Technology Co., Ltd. in the full name will be more similar to the full names of other similar companies, resulting in that multiple different companies are considered to be the same, and the error statistics occur.

[0130] 2.3.8.3 Traverse the object class keyword classification cluster obtained in the last step

[0131] The reason class keyword corresponding to the data in the clustering cluster is used. The dbscan clustering method is used to set the eps parameter to 0.3 and the min_samples parameter to 5. Then the clustering result is traversed to obtain the result of the keyword clustering step. In the case of the same complaint object, the reason can be appropriately relaxed.

[0132] 2.3.9 Fusion of clustering results

[0133] Fuse the clustering results of step 2.3.6 and the result list of 2.3.8.3 to obtain the basic group complaint event list.

[0134] 2.3.10 Data expansion of group complaint list

[0135] After obtaining the group complaint list, in order to obtain more real data of the number of complaint cases and the number of complainants, it is necessary to find out whether there is data that meets the characteristics of the classification cluster from the unclassified data according to the data characteristics of each classification cluster.

[0136] This embodiment first uses the classification cluster characteristics (calculate the mean value of the vectorization matrix of the data in the cluster to obtain the cluster vector) to screen the unclassified data, and the semantic similarity between the unclassified data and the classification cluster characteristics needs to be greater than 0.8. The spatial characteristics of the classification cluster are screened in the same way. Get the to-be-screened data that meets the semantic and spatial similarity threshold at the same time.

[0137] The data to be filtered is mixed with the already classified data, and clustering is performed using dbscan. Here, the eps and min_samples parameters are gradually increased. This continues until the already classified data are re-clustered and still belong to the same category. At this point, other data belonging to the same category as the already classified data are considered as augmented data.

[0138] In this embodiment, during the one-time clustering, since the threshold is just a numerical value and cannot quantify whether the two complaints are indeed about the same thing, the initial vectorized clustering usually uses a higher similarity threshold so that the results are reliable.

[0139] For example, complaints about high property management fees in Community A and Community B have a high semantic similarity, resulting in a combined similarity score of over 0.85.

[0140] The subsequent clustering after obtaining the basic clustering results is like following a map; it involves finding data from the existing data that matches the existing clusters. The restrictions can be appropriately relaxed to efficiently expand the existing cluster data.

[0141] 2.4 Summary of group litigation incidents, statistics on the number of complaints and the number of complainants

[0142] Summary was obtained by extracting two keywords from each group complaint service order. The keywords with the most frequent occurrences related to the object category were selected.

[0143] The combination of keywords related to the cause of action yields a summary of group litigation events.

[0144] For example: the most frequent occurrence of the object class is "xx residential community," and the most frequent reason for this is "excessively high property management fees."

[0145] The final conclusion of this group lawsuit was that the property management fees in XX residential community were too high.

[0146] 2.5 Group lawsuit cases are stored in the database.

[0147] The foregoing has shown and described the basic principles and main features of the present invention and its advantages. It will be apparent to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or basic features of the present invention.

[0148] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A smart government hotline mass appeal analysis and detection method, characterized in that, It includes two parts: keyword extraction of work order information and detection of public demands. The keyword extraction of work order information includes the following steps: S1.1, Periodically retrieve unanalyzed work order data; S1.2 Batch analysis of work order data; S1.3, Analysis of Model Results; S1.4 Result storage; The public demand assessment includes the following steps: S2.

1. Retrieve daily work order data on a regular schedule, retrieving work order data from the business database and analysis data associated with the work orders from Elasticsearch. S2.

2. Vectorize the work order content in the work order data; S2.3, Work order data clustering; S2.4 Summarize the group complaint incidents and compile statistics on the number of complaints and the number of complainants; S2.

5. Store the group lawsuit incidents in the database; The work order data clustering process in step S2.3 specifically includes the following: S2.3.1 Calculate the cosine similarity between the work order content vectors; S2.3.2 Calculate the cosine similarity between work order address vectors; S2.3.3 Calculate the similarity between work orders using weighted averages, with semantic similarity accounting for 0.5 and spatial similarity accounting for 0.5, to obtain the work order similarity score. The score ranges from 0 to 1, with larger values ​​indicating greater similarity. S2.3.

4. Using 1 - work order similarity, the work order similarity distance is obtained. The similarity distance ranges from 0 to 1, and the smaller the value, the closer the distance. S2.3.

5. Run the dbscan algorithm, set eps to 0.05 and num_samples to 5; S2.3.6 Obtain the classification cluster results from vectorized clustering calculation; S2.3.7 Clean the unlabeled data and filter out data of specific categories; S2.3.8 Keyword clustering: Keywords are divided into target keywords and complaint reason keywords; S2.3.

9. Merging the clustering results: Merge the clustering results from step 2.3.6 with the result list from 2.3.8.3 to obtain a basic list of group complaint events; S2.3.

10. Expand the list of group lawsuits; In the process of extracting keywords from work order information, the work order content and detailed address of unanalyzed work orders are obtained by pulling unanalyzed work order data. The process of batch analysis of work order data includes: inputting work order content and detailed address; constructing a large language model for input; batch calling the interface deployed by the glm4-9b-chat model to obtain results; the model result parsing process includes parsing the keywords in the output JSON format text string and storing the parsing results in Elasticsearch; in step S2.2, the open-source text2vec-base-chinese model is used for vectorization processing. In this process, the content and address of each work order are first input into the model, and the corresponding vector representation is obtained through the output layer of the BERT model and used as the feature vector of the text for subsequent similarity calculation, classification or other tasks; through vectorization, the semantic information of the work order content and address is preserved and converted into a numerical format that can be processed by a computer.

2. The intelligent method for analyzing and detecting group demands in government hotlines according to claim 1, characterized in that: The similarity between two vectors is determined by calculating the cosine of the angle between them, with values ​​ranging from -1 to 1, where 1 indicates perfect similarity, 0 indicates no similarity, and -1 indicates complete oppositeness. The formula is as follows: 。 3. The intelligent method for analyzing and detecting group demands in government hotlines according to claim 1, characterized in that, The steps of the dbscan algorithm include: selecting an unvisited point from the dataset and determining whether it is a core point; if it is a core point, grouping all density-reachable points in its neighborhood into the same cluster; for non-core points, if it is adjacent to a core point, it becomes part of that cluster, otherwise it is marked as noise.

4. The intelligent method for analyzing and detecting group demands in government hotlines according to claim 1, characterized in that: In step S2.3.8, keywords are divided and vectorized. Then, object-type keywords are clustered based on the large model. The number of object-type keywords that meet the threshold is traversed. Finally, the large model is used to determine whether the few words most similar to it point to the same entity.

5. The method according to claim 4, wherein the method further comprises: The keyword clustering steps are obtained by iterating through the obtained object class keyword clusters and the reason class keywords corresponding to the data in each cluster. The dbscan clustering method is used, with the eps parameter set to 0.3 and the min_samples parameter set to 5. The clustering results are then iterated through to obtain the results of the keyword clustering steps. ​ 6. The intelligent method for analyzing and detecting group demands in government hotlines according to claim 1, characterized in that: In the expansion phase of the group lawsuit list data, unclassified data is first filtered using classification cluster features, requiring a semantic similarity greater than 0.8 with the classification cluster features. Then, spatial features of the classification clusters are used for filtering to obtain data that simultaneously meets the semantic and spatial similarity thresholds. Next, the data to be filtered is mixed with the classified data and clustered using dbscan, gradually increasing the eps and min_samples parameters until the classified data are re-clustered and still belong to the same category. At this point, other data in the same category as the classified data are considered as expanded data.

Images