Work order real-time early warning method and model based on fuzzy address standardization and event multistage clustering of large model

Abstract

The invention provides a fuzzy address standardization and event multistage clustering work order early warning method based on a large model, and the method comprises the steps: firstly accessing multi-source channel work order data, extracting work order field contents, and then carrying out the calculation analysis of an algorithm model, and generating an early warning event association relation table; and constructing a service system synchronization event association relation table, and finally calling intelligent early warning scene multi-rule early warning judgment logic to finally generate six scene early warning events. Meanwhile, the invention further provides the early warning model, the address matching accuracy can be improved to 96% from 72% of a traditional method, the event clustering efficiency can be shortened to 15 minutes from 2 hours per 1000 work orders of a traditional system, and the risk early warning efficiency is shortened to be within 30 minutes from the average 4 hours.

Description

Technical Field

[0001] This invention belongs to the technical field of intelligent urban safety governance, and further relates to large-scale models, spatiotemporal clustering analysis, risk warning algorithms, and service hotline work order systems of related institutions, especially to a real-time early warning method and model for work orders based on fuzzy address standardization and multi-level clustering of events using large-scale models. Background Technology

[0002] With the development of the internet, organizations have established service hotlines for their respective services. Users submit their requests via work orders, connecting online with the organization to receive feedback or solutions, thus completing their requests. This has become a crucial channel for problem communication. Currently, from a technical perspective, processing these work orders typically involves initial event screening and classification. Using pre-defined keywords, the system analyzes information such as the event subject, address, and event attributes to promptly identify potential group or emergency events, automatically issuing warnings. Manual review and decision support are then provided, with key events identified by the intelligent system undergoing manual verification to ensure accuracy. As a vital communication bridge between organizations and users, the quality of service provided by the hotline directly impacts the organization's image and credibility, and profoundly influences the effectiveness of intelligent governance and user satisfaction.

[0003] With the rapid development of technologies such as large-scale modeling and the accelerated construction of digital institutions, upgrading intelligent applications has become a key measure to enhance the "soft power" of related service hotlines. AI technology is gradually being applied in service hotlines, but the integration of technology with real-world scenarios still faces challenges. For example, existing system technologies struggle to address the following issues: Address resolution is inefficient: When users report problems through the hotline, the address descriptions are diverse and ambiguous (such as "intersection of ** Road and ** Road"), making it impossible for the system to accurately identify risky locations. More than 30% of work orders with such ambiguous addresses require manual review, and the accuracy rate of traditional geocoding is less than 70%; the resolution recall rate is less than 65%, which can lead to work orders being misassigned or returned, wasting manpower.

[0004] Event dispersion: The same risk event (such as missing manhole cover or road collapse) may be recorded as multiple independent work orders due to differences in user descriptions or time intervals, which delays the efficiency of handling.

[0005] Risk warning lag: Traditional systems rely on manual sampling or single-dimensional (such as time, keyword) analysis, which makes it difficult to capture potential group risk events involving multiple people filing lawsuits in real time; the identification accuracy rate is less than 65%, and manual merging is inefficient, affecting user satisfaction. Summary of the Invention

[0006] Technical Solution: To address the shortcomings of traditional address encoding (which only supports standard addresses and cannot handle ambiguous descriptions such as "northwest corner of an intersection"), rule engines (which rely on threshold settings and cannot adapt to dynamic urban environments, resulting in lagging identification of emerging events), and keyword extraction (which relies solely on keywords to extract the topic and responsible party of a work order, ignoring technical issues such as address ambiguity and semantic association), this invention provides a real-time work order early warning method based on large-scale model-based fuzzy address standardization and multi-level event clustering. The method includes the following steps: S01: Access multi-source work order data and parse and extract data through text structuring; S02: After extracting the content of the work order fields, the algorithm model multidimensional event association analysis algorithm is used to generate a table of early warning event association relationships; S03: Establish a synchronous event association table, call the multi-rule early warning judgment logic for each scenario of intelligent early warning, and finally generate early warning events for six major scenarios; if the early warning threshold is exceeded, an early warning message will be issued, an early warning push will be sent, an intelligent routing strategy will be implemented, an automated handling will be carried out, and early warning suppression and optimization will be performed.

[0007] As an improvement, the data parsed in S01 includes work order number, request content, request type, work order creation time, request purpose, first-level jurisdiction, second-level jurisdiction, and third-level jurisdiction; the data parsing algorithm is not a necessary condition, which can adapt to the access of work order data from multiple sources.

[0008] As an improvement, the algorithm model operation and parsing in S02 includes basic information structuring, subject / address result extraction, and event tag result generation; After the work order information is structured and the data is extracted, different parameter weights are set to match and process the warning events, and the final warning event set result is generated based on the event window.

[0009] As an improvement, the basic information structure is parsed to include the type of claim, the area involved, the claimant, and the amount involved. The real-time calculations for processing include: (1) Through training with massive historical work order data, it can automatically identify the whole entity without relying on a dictionary: Specifically, it uses a large model to optimize prompt words to construct the whole entity. In terms of technical logic: upload the work order, the large model performs intelligent parsing, optimizes prompt words to extract the required entity, and returns structured work order information; (2) Utilize context to dynamically adjust word segmentation strategy: By using a BPE pre-trained word segmenter, input text, and set self-attention to dynamically adjust word boundaries. The attention mechanism calculates the similarity between query Q and key K by taking the dot product of query Q and key K and then dividing by the key dimension. d kThe square root of the similarity is then used, and the softmax function is used to convert this similarity into a set of weights, all of which sum to 1. Finally, these weights are multiplied by the value. V Add the results together to get the final output; (3) Directly process the original text, calculate weights through multi-head attention, and pass high-weight information to the downstream layer to automatically focus on key information. The multi-head attention mechanism formula is as follows: MultiHead(Q,K,V)=Concat(head1,…,headh,…,headn)WO The query Q, key K, and value V are preferably divided into n parts. Each part is then subjected to an attention calculation using the single-head attention method described earlier, resulting in n different output results. head1 arrive headn Right now head1,…, headh,…,headn , h , n The result is an integer. All results are concatenated (Concat) and then multiplied by a weight matrix. WO Perform a linear transformation to obtain the final output; (4) Utilize the capabilities of large models to capture semantic relationships between texts; employ bidirectional context encoding to output semantic relationships after text input; (5) Event structure analysis: Through entity recognition NER and relation extraction RE, the big model automatically extracts the type of complaint, the field involved, the complainant, and the amount involved, which are used as the data for early warning events and can be called to build a complete event profile; Preferably, the specific process of event structure parsing can be as follows: first, input the work order; then, extract it through generative extraction from a large model; and finally, output a structured event profile. (6) Cross-element association reasoning: Combining semantic logic, identifying subject differences, and determining whether they are the same event; (7) Use large model prompts to quickly adapt to new scenarios: Design new scenario prompts, optimize soft prompt vectors, and then use them directly for new scenario prediction.

[0010] As an improvement, the specific steps for extracting the subject / address results are as follows: For work order data, the subject / address data involved is extracted using an algorithm. The extracted data is then matched with the subject / address database. If the match is successful, the standard subject / address data from the subject / address database is returned. If the match fails, the extracted data will be called to perform standard data parsing using the Gaode API. After parsing, the data will be matched with the subject / address database. If the match is successful, the standard subject / address data from the subject / address database will be returned. If the data parsed by Gaode Maps fails to match the subject / address database, the parsed address will be compared and verified with the extracted data according to the subject / verification logic, and the subject / address data will be output. The data extracted by the algorithm and parsed by Gaode Maps will be synchronously added to the event subject / address database for subsequent data matching.

[0011] As an improvement, the specific steps for generating event tag results are as follows: For work order data, work order summary is extracted using an algorithm. The extracted summary data is matched with the fourth-level tags in the event tag library. If the match is successful, standard event tag data is output, which includes tags from level 1 to level 5, and the fifth-level tag library is updated. If the fourth-level tag matching fails, the algorithm will automatically generate new fourth-level tags based on the third-level classification data of the work order, output standard event tag data, including tags of levels 1-5, and supplement the initial tag library with the result data; combined with the tag system, work order summary event set and fourth-level tag event set data will be generated, and then different early warning events will be generated according to the aggregation threshold of the early warning rules.

[0012] As an improvement, the warning events for the six scenarios in S03 are processed according to the warning algorithm rules, which include... (1) Multiple people complaining: Based on the event information generated by the big model, when users have the same demands on the same thing, when different people complain about the same thing and the aggregation threshold is reached, multiple people complaining warning can be issued. (2) Multiple complaints by one person: Based on the event information generated by the large model, when the same or different demands are made by the same person multiple times within a certain time range and the aggregation threshold is reached, a warning can be issued for multiple complaints by one person, and the group of people making multiple demands can be linked. (3) Emergencies: Based on the event information generated by the large model, if the current work order contains emergency event keywords pre-configured by the system, and the number of occurrences of the emergency event keyword within a certain time range exceeds the aggregation threshold, the system will issue an emergency event warning; (4) Emerging events: Event information generated based on large models that grows rapidly in a short period of time and has the potential to become urban hot events. The system monitors the trend of the number of each event within a certain time range. If the number exceeds the aggregation threshold, the system will trigger an early warning for emerging events. (5) Initiating Organizations: Build a large model to analyze the organizational information involved in the event work order in real time, monitor whether the organization has an emerging growth trend within a certain time range, and if it exceeds the aggregation threshold, the system will trigger an early warning for the initiating organization. (6) Early warning location: Build a large-scale real-time analysis of the address information involved in the event work order, monitor whether there is an early warning trend in the address location within a certain time range, and if the aggregation threshold is exceeded, the system will trigger an early warning location warning.

[0013] As an improvement, early warning systems also include profiling and thematic early warnings. The profiling analysis is achieved by establishing a four-level linkage profiling system of "individual-group-region-institution". The system automatically extracts multi-dimensional data such as individual appeal behavior trajectory, group aggregation characteristics, and regional risk distribution to generate profiles of appealers, appeal groups, areas with concentrated appeals, and event handling institutions. The aforementioned thematic early warning involves establishing a thematic tagging system, conducting specialized training on thematic data, and generating thematic early warning results. Furthermore, it can include thematic early warnings for consumer disputes and insufficient business licenses for merchants.

[0014] Meanwhile, a real-time work order early warning model based on fuzzy address standardization and multi-level event clustering based on a large model is provided. The model structure includes: an input layer, a feature extraction layer, a relational reasoning layer, an output layer, and an intelligent return layer. The input layer is the original work order data, the feature extraction layer is a module for extracting structured data, the relational reasoning layer is a module for processing according to the rules of multiple parties filing complaints, the output layer outputs early warning questions that exceed the threshold, and the intelligent return layer selects an appropriate processing agency for direct return.

[0015] Beneficial effects: The early warning method proposed in this invention is a real-time early warning method for work orders based on large-scale model fuzzy address standardization and multi-level event clustering. It can realize early warning, analysis and closed-loop processing of events in multiple scenarios, and can improve the address matching accuracy from 72% of the traditional method to 96%. Event clustering efficiency: the screening time of the traditional system is reduced from 2 hours / 1000 work orders to 15 minutes. Risk early warning timeliness: the average response time is reduced from 4 hours to within 30 minutes. Attached Figure Description

[0016] Figure 1 This is the overall algorithm flow of the work order early warning method based on fuzzy address standardization and multi-level event clustering of the present invention.

[0017] Figure 2 This is a schematic diagram of the structure of the present invention during basic information structuring.

[0018] Figure 3 This is a flowchart illustrating the algorithm for refining and standardizing the subject / address results in this invention.

[0019] Figure 4 This is a flowchart illustrating the event tag result generation algorithm of the present invention. Detailed Implementation

[0020] The specific embodiments of the present invention will be described in further detail below with reference to the examples. These examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0021] The work order early warning method based on large model fuzzy address standardization and multi-level clustering of events of the present invention includes (1) early warning events: based on real-time calculation of large model, accurately extracting event elements, structuring basic work order information, combining subject / address result accurate standardization processing algorithm and event tag result generation algorithm, and automatically generating six major early warning information according to the six major event early warning rules, realizing multiple people filing lawsuits, one person filing multiple lawsuits, sudden events, incipient events, incipient institutions, and incipient addresses, supporting early warning data query, export, and viewing, and can export early warning related work order data; providing a daily early warning summary automatic generation function, and can export the summary.

[0022] (2) Early warning analysis: It realizes the visualization analysis of 7 major topics, including monitoring and early warning overview, multiple people filing lawsuits, one person filing multiple lawsuits, sudden events, emerging events, emerging institutions, and emerging addresses, and analyzes from the dimensions of high-incidence events, changes in early warning trends, hot words of events, and event handling analysis.

[0023] See Figure 1 The diagram shows the overall algorithm flow of the work order early warning method based on fuzzy address standardization and multi-level event clustering of this invention. Specifically, it includes: accessing work order data from multiple sources, extracting the content of work order fields, and then performing calculations and parsing by the algorithm model to generate an early warning event association table; synchronizing the event association table with the business system, calling the multi-rule early warning judgment logic for each scenario of the intelligent early warning system, and finally generating early warning events for six major scenarios.

[0024] Furthermore, the algorithm parses the work order data as work order number, request content, request type, work order creation time, request purpose, first-level jurisdiction, second-level jurisdiction, and third-level jurisdiction. In this invention, the data that the algorithm needs to parse should be set as a non-essential condition to adapt to the access of work order data from multiple sources.

[0025] The internal logic of the algorithm model is divided into three parts: basic information structuring, subject / address result extraction, and event tag result generation. After extracting data from the work order information, different parameter weights need to be set to match and process the warning events, and the final warning event set result is generated based on the event window.

[0026] See Figure 2 As shown, the basic information is structured, including multiple components such as the type of request, the area involved, the complainant, and the amount involved. The specific processing steps include the following: (1) The large model was trained with massive historical work order data in the early stage. For example, it has mastered the complete semantics of combined words such as "meta-universe device" and can be automatically identified as a whole entity without relying on a dictionary.

[0027] (2)Dynamically adjust the word segmentation strategy according to the context. For example, in the scenario of "mobile phone repair", "charging port loose" is preferably recognized as "charging port - loose" rather than being split into "charging - port - loose" to avoid fragmentation of key information.

[0028] (3)The large model directly processes the original text without removing stop words, and automatically focuses on key information through the attention mechanism. For example, "unable to use" is the core, and words like "of" and "already" do not affect semantic judgment.

[0029] Utilize the capabilities of the large model to capture semantic associations before and after the text, and recognize that both "Wi-Fi unable to access the Internet" and "WeChat anomaly under data traffic" belong to the theme of "mobile phone network connection failure".

[0030] Event structure analysis: Through entity recognition (NER) and relation extraction (RE), the large model can automatically extract the type of appeal, the involved field, the appellant, and the involved amount, which are used as data components for early warning events for calling, and construct a complete event portrait.

[0031] Cross-element association reasoning: Combining semantic logic (for example, the object of complaint for "broadband network interruption" is usually the operator, not the property management), the large model can identify the subject differences between "community broadband network interruption" and "property management inaction" to avoid misjudging them as the same event.

[0032] At the same time, use prompt learning of the large model to quickly adapt to new scenarios, such as migrating from "mobile phone failure" to "smart home failure", solving the limitation problem of traditional system maintenance keywords, with strong later extensibility and low maintenance cost.

[0033] See Figure 3 As shown, it is a flowchart of the algorithm for precise and standardized processing of the subject / address results of the present invention. For work order data, the involved subject / address data is extracted through the algorithm, and the extracted data is matched with the data in the subject / address database. If the match is successful, the standard subject / address data in the subject / address database is returned; if the match is unsuccessful, the extracted data is called to the Gaode API for standard data parsing, and after the parsed data is matched with the data in the subject / address database again, if the match is successful, the standard subject / address data in the subject / address database is returned; If the data parsed by Gaode is not successfully matched with the subject / address database, the parsed address will be compared and verified with the extracted data according to the subject / review logic, and then the subject / address data will be output, and the data extracted by the algorithm and the data parsed by Gaode will be synchronously supplemented to the event subject / address database for subsequent data matching.

[0034] See Figure 4The diagram shows a flowchart of the event tag result generation algorithm of the present invention. For work order data, the algorithm extracts work order summaries and matches the extracted summary data with the fourth-level tags in the event tag library. If the match is successful, standard event tag data (containing first- to fifth-level tags) is output and the fifth-level tag library is updated. If the fourth-level tag matching fails, the algorithm will automatically generate new fourth-level tags based on the third-level classification data of the work order, output standard event tag data (including tags of levels 1-5) and supplement the initial tag library with the result data; combined with the tag system, work order summary event set and fourth-level tag event set data will be generated, and then different early warning events will be generated according to the aggregation threshold of the early warning rules.

[0035] As a specific implementation of the present invention, the present invention proposes six important event early warning algorithm rules, including: (1) Multiple people complaining: Based on the event information generated by the big model, when users have the same demands on the same thing, when different people complain about the same thing and the aggregation threshold is reached, multiple people complaining warning can be issued. (2) Multiple complaints by one person: Based on the event information generated by the large model, when the same or different demands are made by the same person multiple times within a certain time range and the aggregation threshold is reached, a warning can be issued for multiple complaints by one person, and the group of people making multiple demands can be linked. (3) Emergencies: Based on the event information generated by the large model, if the current work order contains emergency event keywords pre-configured by the system, and the number of occurrences of the emergency event keyword within a certain time range exceeds the aggregation threshold, the system will issue an emergency event warning; (4) Emerging events: Event information generated based on large models that grows rapidly in a short period of time and has the potential to become urban hot events. The system monitors the trend of the number of each event within a certain time range. If the number exceeds the aggregation threshold, the system will trigger an early warning for emerging events. (5) Initiating Organizations: Build a large model to analyze the organizational information involved in the event work order in real time, monitor whether the organization has an emerging growth trend within a certain time range, and if it exceeds the aggregation threshold, the system will trigger an early warning for the initiating organization. (6) Early warning location: Build a large-scale real-time analysis of the address information involved in the event work order, monitor whether there is an early warning trend in the address location within a certain time range, and if the aggregation threshold is exceeded, the system will trigger an early warning location warning.

[0036] As a specific embodiment of the present invention, a real-time early warning model for work orders based on fuzzy address standardization and multi-level event clustering based on a large model is also provided. The model structure includes: an input layer, a feature extraction layer, a relational reasoning layer, an output layer, and an intelligent return layer. The input layer is the original work order data, the feature extraction layer is a module for extracting structured data, the relational reasoning layer is a module for processing according to the rules of multiple parties filing complaints, the output layer outputs early warning questions that exceed a threshold, and the intelligent return layer selects an appropriate processing agency for direct return.

[0037] Preferably, the data training method includes: training set composition, feature selection, annotation method, and loss function optimization. Feature selection includes type, subject, address, topic, name, contact information, ID number, label type, and event description. Label studio is used for annotation, and cross-entropy loss is used for loss function optimization. Compared with traditional algorithms, the prediction model in this invention can improve detection accuracy, reduce false positives, and improve real-time performance, as shown in Table 1.

[0038] Table 1. Performance Comparison between Traditional Algorithm and the Improved Technology of This Invention Example

[0039] The method for early warning according to the present invention will be described in detail below through specific embodiments. First, address matching is performed using the method of the present invention, transforming the original basic algorithm for integrating structured addresses into a method using Gaode Maps. API Accuracy was improved from 72% to 96% by using road coding and standard addresses from Gaode Maps search. Then, the efficiency of event clustering, event clustering, and the timeliness of risk warnings were assessed. Event clustering efficiency refers to the speed improvement achieved by using multi-threaded parallel processing, incremental processing, and full computation, using the following formula: MinHash + LSH h(S) = min(h(x) | x∈S) See the formula above. h(x) A hash function that sets elements... x Mapped to an integer; S A set is the basic unit for similarity comparison; x :gather S An element in a feature is the smallest unit that constitutes a characteristic. MinHash The set is represented by the smallest hash value. LSH A bucketing strategy is used to quickly find similar sets, and the combination of these two approaches enables efficient approximate similarity calculation. Furthermore, the timeliness of risk warning refers to real-time data access... NLPReal-time analysis, multimodal fusion, and automatic threshold-based early warning are combined to comprehensively optimize the slow response and time-consuming issues of traditional algorithms. Table 2 below illustrates this process.

[0040] Table 2. Examples of event clustering efficiency and risk warning timeliness (using 1000 work order data as an example)

[0041] As shown in Table 2, the event clustering efficiency has been reduced from 2 hours / 1000 work orders to 15 minutes compared to the traditional system screening time, and the risk warning timeliness has been reduced from an average of 4 hours to within 30 minutes.

[0042] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A work order early warning method based on fuzzy address standardization and multi-level event clustering using a large model, characterized by: The steps of the method include S01: Access work order data from multiple sources and parse and extract data through text structuring; S02: After extracting the content of the work order fields, the algorithm model multidimensional event association analysis algorithm is used to generate a table of early warning event association relationships; S03: Establish a synchronous event association table, call the multi-rule early warning judgment logic for each scenario of intelligent early warning, and finally generate early warning events for six major scenarios; If the warning threshold is exceeded, a warning message will be issued, a warning push will be sent, an intelligent routing strategy will be implemented, automated handling will be carried out, and warning suppression and optimization will be implemented.

2. The work order early warning method based on large-scale model fuzzy address standardization and multi-level event clustering according to claim 1, characterized in that: The data parsed in S01 includes work order number, request content, request type, work order creation time, request purpose, primary jurisdiction, secondary jurisdiction, and tertiary jurisdiction; The algorithm parses data bits as a non-essential condition, and can adapt to the access of work order data from multiple sources.

3. The work order early warning method based on large-scale model fuzzy address standardization and multi-level event clustering according to claim 1, characterized in that: The algorithm model operation and parsing in S02 includes basic information structuring, subject / address result extraction, and event tag result generation; After the work order information is structured and the data is extracted, different parameter weights are set to match and process the warning events, and the final warning event set result is generated based on the event window.

4. The work order early warning method based on large model fuzzy address standardization and multi-level event clustering according to claim 3, characterized in that: The structured parsing of the basic information includes the type of complaint, the area involved, the complainant, and the amount involved. Real-time calculations for processing include: (1) Through training with massive historical work order data, it can automatically identify the whole entity without relying on a dictionary: Specifically, it uses a large model to optimize prompt words to construct the whole entity. In terms of technical logic: upload the work order, the large model performs intelligent parsing, optimizes prompt words to extract the required entity, and returns structured work order information; (2) Dynamically adjust word segmentation strategy using context: through BPE A pre-trained word segmenter is used as input text, and self-attention is set to dynamically adjust word boundaries. ； Attention mechanisms compute queries Q The similarity to key K is the query Q s and keys K Perform a dot product, then divide by the key dimension. d k The square root, then use softmax The function converts this similarity into a set of weights, all of which are added together to equal 1. Finally, these weights are multiplied by the value V, and the results are summed to obtain the final output. (3) Directly process the original text, calculate weights through multi-head attention, and pass high-weight information to the downstream layer to automatically focus on key information. The multi-head attention mechanism formula is as follows: MultiHead(Q,K,V)=Concat(head1,…,headh,…,headn)WO; Query Q ,key K ,value V Divided into n Each portion is processed individually, and attention is calculated separately to obtain... n Different output results head1,…,headh,…,headn , h , n The results are integers, then all the results are concatenated, and finally multiplied by a weight matrix. WO Perform a linear transformation to obtain the final output; (4) Utilize the capabilities of large models to capture semantic relationships between texts; employ bidirectional context encoding to output semantic relationships after text input; (5) Event structure analysis: Through entity recognition NER and relation extraction RE, the big model automatically extracts the type of complaint, the field involved, the complainant, and the amount involved, which are used as the data for early warning events and can be called to build a complete event profile; (6) Cross-element association reasoning: Combining semantic logic, identifying subject differences, and determining whether they are the same event; (7) Use large model prompts to quickly adapt to new scenarios: Design new scenario prompts, optimize soft prompt vectors, and then use them directly for new scenario prediction.

5. The work order early warning method based on large model fuzzy address standardization and multi-level event clustering according to claim 4, characterized in that: The specific method in (6) is to extract event entities, calculate semantic similarity values, perform entity alignment, set thresholds, and determine whether they are the same event by comparing them with the actual values.

6. The work order early warning method based on large model fuzzy address standardization and multi-level event clustering according to claim 3, characterized in that: The specific steps for extracting the subject / address results are as follows: For work order data, the subject / address data involved is extracted using an algorithm. The extracted data is then matched with the subject / address database. If the match is successful, the standard subject / address data from the subject / address database is returned. If the match fails, the extracted data will be called to perform standard data parsing using the Gaode API. After parsing, the data will be matched with the subject / address database. If the match is successful, the standard subject / address data from the subject / address database will be returned. If the data parsed by Gaode Maps fails to match the subject / address database, the parsed address will be compared and verified with the extracted data according to the subject / verification logic, and the subject / address data will be output. The data extracted by the algorithm and parsed by Gaode Maps will be synchronously added to the event subject / address database for subsequent data matching.

7. The work order early warning method based on large model fuzzy address standardization and multi-level event clustering according to claim 3, characterized in that: The specific steps for generating event tag results are as follows: For work order data, work order summary is extracted using an algorithm. The extracted summary data is matched with the fourth-level tags in the event tag library. If the match is successful, standard event tag data is output, including tags from level 1 to level 5, and the fifth-level tag library is updated. If the fourth-level tag matching fails, the algorithm will automatically generate new fourth-level tags based on the third-level classification data of the work order, output standard event tag data, including tags of levels 1-5, and supplement the initial tag library with the result data; combined with the tag system, work order summary event set and fourth-level tag event set data will be generated, and then different early warning events will be generated according to the aggregation threshold of the early warning rules.

8. The work order early warning method based on large-scale model fuzzy address standardization and multi-level event clustering according to claim 1, characterized in that: In S03, the six scenario-based early warning events are processed according to early warning algorithm rules, which include: Multiple complaints: Based on event information generated by a large model, when users have the same demands on the same matter, and when different people complain about the same matter and the aggregation threshold is reached, a warning for multiple complaints can be issued. Multiple complaints by one person: Based on event information generated by a large model, when the same person makes the same or different requests multiple times within a certain time range and the aggregation threshold is reached, a warning for multiple complaints by one person can be issued, and the group of people making multiple requests can be linked. Emergency Events: Based on event information generated from a large model, if the current work order contains emergency event keywords pre-configured by the system, and the number of occurrences of such emergency event keywords exceeds the aggregation threshold within a certain time range, the system will issue an emergency event warning; Emerging events: Event information generated based on a large model that grows rapidly in a short period of time and has the potential to become a hot topic in the city. The system monitors the trend of the number of each event within a certain time range. If the number exceeds the aggregation threshold, the system will trigger an early warning for emerging events. Early warning system: Build a large model to analyze the organizational information involved in event work orders in real time, monitor whether the organization has an early growth trend within a certain time range, and if it exceeds the aggregation threshold, the system will trigger an early warning system alert. Early warning location: The system builds a large-scale real-time analysis module to analyze the address information involved in the event work order and monitors whether there is an early warning trend at that address location within a certain time range. If the trend exceeds the aggregation threshold, the system will trigger an early warning location alert.

9. The work order early warning method based on large-scale model fuzzy address standardization and multi-level event clustering according to claim 1, characterized in that: Early warning methods also include profiling and thematic early warnings; The profiling analysis is achieved by establishing a four-level linkage profiling system of "individual-group-region-institution". The system automatically extracts multi-dimensional data such as individual appeal behavior trajectory, group aggregation characteristics, and regional risk distribution to generate profiles of appealers, appeal groups, areas with concentrated appeals, and event handling institutions. The aforementioned thematic early warning system involves establishing a thematic tagging system, conducting specialized training on thematic data, and generating thematic early warning results.

10. A real-time work order early warning model based on fuzzy address standardization and multi-level event clustering using a large-scale model, characterized in that... The model's structure includes: an input layer, a feature extraction layer, a relational reasoning layer, an output layer, and an intelligent return layer. The input layer contains the original work order data, the feature extraction layer is a module for extracting structured data, the relational reasoning layer is a module that processes data based on rules for multiple parties filing complaints, the output layer outputs warnings for issues exceeding a threshold, and the intelligent return layer selects an appropriate processing agency for direct return.

Images