Property voting method and system based on retrieval enhancement generation
By deploying large language models and text embedding models in the property voting system and building a knowledge base and database, the problem of understanding difficulty and professional knowledge acquisition during the voting process in the property voting system is solved, and the transparency and objectivity of the voting results are achieved.
Patent Information
- Application Number
- CN202411548692.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-01
- Publication Date
- 2024-12-03
AI Technical Summary
The existing property voting system has difficulty understanding the voting topics and content when dealing with a wide audience, and lacks expertise and data support, resulting in a lack of objectivity and transparency in voting results.
Design a property voting system based on search enhancement generation. By deploying large language models and text embedding models, building a knowledge base and database, realizing the answers to voting topics and content and visual display of data, providing detailed background information and professional knowledge. explain.
It improves the transparency and objectivity of the voting process, helps voters better understand voting issues and content, avoid blindness, and realizes the intuitive display of professional knowledge and data.
Smart Images

Figure CN119066051A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of smart property management, and specifically relates to a property voting method and system based on retrieval enhanced generation. Background Art
[0002] With the rapid development of information technology, the ways of social governance and residents' self-governance are also constantly innovating. As an important tool for the modernization of community management, the property voting system has continuously innovated to inject new impetus into community governance, and has made important contributions to improving the efficiency and transparency of decision-making and enhancing the enthusiasm of owners to participate in community affairs.
[0003] The current property voting system is mainly implemented through WeChat public accounts or mini-programs. Although it has brought great convenience to the property voting process, the general degree of intelligence is not high. In some property voting scenarios, the current property voting system still has many shortcomings: on the one hand, in the property voting process, the audience is wide, but due to individual differences, some voters may find it difficult to understand the theme and specific content of the vote, resulting in blindness in the voting results; on the other hand, for the professional knowledge and data involved in the voting process, such as policies and regulations, operational data, etc., for the existing property voting systems, it is difficult to obtain these professional knowledge and data, or they cannot be obtained at all, resulting in opaque and non-intuitive property data in the voting process, which in turn leads to a lack of objectivity in the voting results. Summary of the invention
[0004] In order to solve at least one of the problems mentioned in the above background technology, the present invention proposes a property voting method and system based on retrieval enhanced generation.
[0005] Step S1, build the property voting system according to the designed technical framework.
[0006] Specifically, the technical framework divides the voting end into three modules: government end, property end and owner end; divides voting matters into three modules: voting on government matters, voting on property matters and voting matters; and divides voting options into three modules: single choice, multiple choice and voting.
[0007] In addition, the technical framework divides the voting content into four modules: voting topic, content description, upload attachments and editing options.
[0008] Step S2: Building an analysis and retrieval system based on the application requirements of the property voting system, specifically including the following steps: Step S201, deploying a large language model and a text embedding model in an analysis and retrieval system; Step S202: deploying the knowledge base of the property voting system in the analysis and retrieval system. The knowledge base construction process specifically includes: Step A1, collect property-related documents; Step A2, converting the collected document data into text vectors; Specifically, the document data is converted into text vectors using a trained text embedding model. The text embedding model can learn how to map text vectors to index structures through training, so that similar vectors can be found quickly during retrieval. Step A3, storing the text vector into a vector database, so that the vector database after storing the text vector serves as a knowledge base.
[0009] By converting documents related to property management and operations into text vectors and storing them in a vector database, it can provide support for subsequent professional knowledge retrieval.
[0010] Step S203: deploying the database of the property system in the analysis and retrieval system. The database construction process specifically includes: Step B1, design the table structure of the database according to the data of the property system; Step B2, select a database management system and create a database; Step B3, converting the indicator parameters in the database table structure into text vectors, storing the text vectors in a vector database, and using the vector database after storing the indicator parameter text vectors as a retrieval library; Step B4, inserting specific data into the database table according to the designed database table structure.
[0011] By storing various property management and operation data into the database and storing indicator parameters into the search library, it can not only provide support for subsequent knowledge base retrieval, but also quickly retrieve data related to the questionable indicator parameters.
[0012] Step S204, using the retrieval enhancement generation framework to connect the large language model, the text embedding model, the knowledge base and the database, and connecting the API of the search engine to the large language model.
[0013] Step S3, establish a communication connection between the property voting system and the analysis and retrieval system, and publish the property voting program in the property voting system.
[0014] Step S4, the property voting system and the analysis and retrieval system provide services to the property voting program to complete the property voting process.
[0015] Specifically, the services provided include answering voting questions and analyzing voting content.
[0016] Among them, answering voting questions is achieved by performing online search through the analysis and retrieval system. The specific process is as follows: Step C1, the property voting system obtains the property voting topic in the property voting program, and sends the obtained voting topic to the analysis and retrieval system; Step C2, the large language model in the analysis and retrieval system generates keywords related to the property voting topic based on the property voting topic; Step C3, sending the keywords generated by the large language model to a search engine, and the search engine performs an online search for the keywords; In step C4, the content retrieved by the search engine is summarized by the large language model and sent to the property voting system, and then sent to the property voting program by the property voting system.
[0017] In view of the problem that voters may find it difficult to understand the voting topics and specific contents during the property voting process, which is difficult to be provided by the existing property voting system, the present invention proposes to obtain the property voting topics in the property voting program through the constructed property voting system, and send the obtained voting topics to the analysis and retrieval system, and then the large language model in the analysis and retrieval system performs an online search for the property voting topics, and summarizes the results of the online search, so as to provide detailed background information and explanations according to the property voting theme, so as to help voters better understand the voting topics and specific contents and avoid blind voting.
[0018] The analysis of voting content is achieved by performing knowledge base retrieval and database query through the analytical retrieval system. The knowledge base retrieval is to query and interpret the professional knowledge involved in the voting process; the database query is to query and visualize the various data involved; Specifically, the knowledge base retrieval process is as follows: Step D1, in the property voting procedure, the voting personnel intercept the doubtful professional knowledge in the property voting content and send the intercepted professional knowledge to the property voting system; Step D2, the property voting system receives the professional knowledge sent by the property voting program, and sends the received professional knowledge to the analysis and retrieval system; In step D3, the text embedding model in the analysis and retrieval system converts the professional knowledge into a text vector, matches the converted text vector in the knowledge base, and returns the matching result.
[0019] Specifically, the text vector of professional knowledge is used as the query vector, and the text vector stored in the knowledge base is used as the storage vector. The matching method is to calculate the similarity between the query vector and the storage vector, and sort the documents according to the similarity score, and finally return the first few documents with the highest similarity score as the matching result.
[0020] The similarity calculation formula is:
[0021] Where α is the query vector, is the storage vector, i is the number of the storage vector, and sim is the similarity.
[0022] In step D4, the matching results are summarized and integrated by the large language model in the analysis and retrieval system and sent to the property voting system, which then sends them to the property voting program.
[0023] The database query process is as follows: Step E1, in the property voting procedure, the voting personnel intercept the index parameters in question in the property voting content, and send the intercepted index parameters to the property voting system; Step E2, the property voting system receives the index parameters sent by the property voting program, and sends the received index parameters to the analysis and retrieval system; Step E3, the text embedding model in the analysis and retrieval system converts the index parameters into text vectors, matches the converted text vectors in the retrieval library, and returns the matching results.
[0024] Specifically, the text vector of the indicator parameter is used as the query vector, and the text vector stored in the retrieval library is used as the storage vector. The matching method is to calculate the similarity between the query vector and the storage vector, and sort the documents according to the similarity score, and finally return the first few indicator parameters with the highest similarity score as the matching result.
[0025] Step E4, the large language model in the analysis and retrieval system generates an SQL statement according to the matching index parameters, sends the generated SQL statement to the database to execute data query, and returns the result of the data query to the large language model; Specifically, the SQL generation rules can be used as prompt words for the large language model, so that the large language model can generate SQL statements according to the SQL generation rules, thereby having the function of generating SQL statements according to indicator parameters.
[0026] Step E5, the large language model visualizes the data query results and sends the visualization results to the property voting system, and the property voting system sends the visualization results to the property voting program.
[0027] Among them, the visualization of data query results is achieved by calling the data-to-image service. Specifically, the data-to-image service is provided by the constructed http server. The http service has a built-in data processing function, which can parse the read data and draw the parsed data into a chart. Furthermore, the drawn chart is returned to the caller.
[0028] In view of the professional knowledge and data involved in the property voting process, such as policies and regulations, operational data, etc., it is difficult or impossible to obtain these professional knowledge and data for the existing property voting system. Voters will intercept the questionable professional knowledge and indicator parameters in the property voting content, and send the intercepted professional knowledge and indicator parameters to the property voting system, and then the analysis and retrieval system will perform knowledge base retrieval and database query on the professional knowledge and indicator parameters. Furthermore, the retrieved professional knowledge will be integrated and summarized, and the queried data will be visualized, thus realizing the transparency and intuitiveness of the professional knowledge and data involved in the voting process, making the voting process more rational and objective.
[0029] The present invention has the following beneficial effects compared with the prior art: The present invention proposes to obtain property voting topics in the property voting program through the constructed property voting system, and send the obtained voting topics to the analysis and retrieval system, and then the large language model in the analysis and retrieval system performs an online search for the property voting topics, and summarizes the results of the online search, so as to provide detailed background information and explanations according to the property voting topics, help voters better understand the voting topics and specific contents, and avoid blind voting.
[0030] The present invention proposes that voters intercept questionable professional knowledge and indicator parameters in the property voting content, and send the intercepted professional knowledge and indicator parameters to the property voting system, and then the analysis and retrieval system performs knowledge base retrieval and database query on the professional knowledge and indicator parameters, and further integrates and summarizes the retrieved professional knowledge, and visualizes the queried data, thereby realizing the transparency and intuitiveness of the professional knowledge and data involved in the voting process, making the voting process more rational and objective. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] Figure 1 It is a system flow chart of the present invention.
[0032] Figure 2 It is a schematic diagram of the architecture of the property voting system based on retrieval enhancement generation of the present invention. DETAILED DESCRIPTION
[0033] In order to make the purpose and features of the present invention more obvious and easy to understand, the present technical solution is described in detail below through embodiments and in conjunction with the accompanying drawings.
[0034] like Figure 2As shown, a property voting system based on retrieval enhanced generation includes a property voting system and an analysis retrieval system, wherein the property voting system is used to publish the property voting program; the analysis retrieval system is used to provide the property voting system with answers to voting topics, and perform knowledge base retrieval and database query services.
[0035] like Figure 1 As shown, a property voting method and system based on retrieval enhancement generation includes the following steps: Step S1, build the property voting system according to the designed technical framework.
[0036] Specifically, the technical framework divides the voting end into three modules: government end, property end and owner end.
[0037] In the specific implementation process, the government and property management sides are responsible for setting the specific types of voting items and implementing new voting; Preferably, it includes: formally publishing voting content, modifying voting, suspending voting at an appropriate time, publicizing voting results, and viewing voting details.
[0038] The property owner can participate in votes initiated by the government or property management, and has the right to modify his or her own voting results, and can also view the details of the vote.
[0039] The technical framework divides voting matters into three modules: voting on government matters, voting on property matters, and voting matters.
[0040] During the specific implementation process, the government has the authority to initiate voting on government matters related to public policies, changes in regulations, community development plans, etc.; the property side has the authority to initiate voting on property matters related to property management, service improvements, facility maintenance, etc.; both the government and the property side have the authority to initiate voting on voting matters, such as the election of owners' committees, community renovation plans, public facilities construction, property fee adjustments, and other voting matters based on actual conditions and needs.
[0041] The technical framework divides voting options into three modules: single choice, multiple choice and voting.
[0042] In the specific implementation process, single-choice allows only one option to be selected for each question, which is suitable for situations that require clear choices; multiple-choice allows multiple options to be selected for each question, which is suitable for situations that need to comprehensively consider multiple factors; voting on major issues has three options: approval, disapproval and abstention, which is suitable for situations that require collective decision-making.
[0043] In addition, the technical framework divides the voting content into four modules: voting topic, content description, upload attachments and editing options.
[0044] Step S2: Building an analysis and retrieval system based on the application requirements of the property voting system, specifically including the following steps: Step S201, deploying a large language model and a text embedding model in an analysis and retrieval system; In the specific implementation process, for the type of large language model, you can choose ChatGLM or Qianwen model; for the text embedding model, you can choose m3e-large, text-embedding-ada-002 and other models; As for the deployment method of large language models and text embedding models, you can deploy large language models and text embedding models on a cloud server, and then call the API interface through the HTTP service; you can also deploy large speech models and text embedding models locally and encapsulate the API interface.
[0045] Step S202: deploying the knowledge base of the property voting system in the analysis and retrieval system. The knowledge base construction process specifically includes: Step A1: Collect property-related documents.
[0046] During the specific implementation process, property management and operation-related documents, such as FAQs, service provider information, technical help documents, and interpretations of regulatory terms, can be used as document materials.
[0047] Step A2: Convert the collected document information into text vectors.
[0048] Specifically, the document data is converted into text vectors using a trained text embedding model. The text embedding model can learn how to map text vectors to index structures through training so that similar vectors can be found quickly during retrieval.
[0049] Step A3, storing the text vector into a vector database, so that the vector database after storing the text vector serves as a knowledge base.
[0050] Preferably, Chroma, Milvus, etc. can be selected as the vector database.
[0051] By converting documents related to property management and operations into text vectors and storing them in a vector database, it can provide support for subsequent professional knowledge retrieval.
[0052] Step S203: deploying the database of the property system in the analysis and retrieval system. The database construction process specifically includes: Step B1, design the table structure of the database according to the data of the property system; Optionally, the data of the property system includes building data, owner data, house data, administrator data, property management data, property operation data, etc.
[0053] For example, during the specific implementation process, tables can be created based on owner data, house data, property fee data, and maintenance data.
[0054] Step B2, select a database management system and create a database; Optionally, MySQL or SQLServer can be used as the database management system; Step B3, converting the indicator parameters in the database table structure into text vectors, storing the text vectors in a vector database, and using the vector database after storing the indicator parameter text vectors as a retrieval library; Step B4, inserting specific data into the database table according to the designed database table structure.
[0055] This step can use SQL statements to insert data into the corresponding entity table.
[0056] By storing various property management and operation data into the database and storing indicator parameters into the search library, it can not only provide support for subsequent knowledge base retrieval, but also quickly retrieve data related to the questionable indicator parameters.
[0057] Step S204, using the search enhancement generation framework to connect the large language model, the text embedding model, the knowledge base and the database, and connecting the search engine's API to the large language model; In the specific implementation process, the APIs of search engines such as Google and Bing can be used, so that information on the Internet can be retrieved through these search engines.
[0058] Optionally, a large language model programming framework can be used to connect large language models, knowledge bases, and databases.
[0059] Preferably, LangChain can be selected as the large language model programming framework.
[0060] For example, to convert text into SQL query statements and return chart information of structured data, LangChain is needed to organize and manage the query and response process of the large language model, and rely on prompt words and code calls to achieve SQL generation, SQL query, and structured data visualization.
[0061] Step S3, establish a communication connection between the property voting system and the analysis and retrieval system, and publish the property voting program in the property voting system.
[0062] Optionally, the HTTP protocol may be selected to establish a communication connection between the property voting system and the analysis and retrieval system.
[0063] Step S4, the property voting system and the analysis and retrieval system provide services to the property voting program to complete the property voting process.
[0064] Specifically, the services provided include answering voting questions and analyzing voting content.
[0065] Among them, answering voting topics is achieved by performing online search through the analysis and retrieval system. The specific process is as follows: Step C1, the property voting system obtains the property voting topic in the property voting program, and sends the obtained voting topic to the analysis and retrieval system; Step C2, the large language model in the analysis and retrieval system generates keywords related to the property voting topic based on the property voting topic; In the specific implementation process, prompt words of the large language model can be written so that the large language model has the ability to retrieve topic keywords.
[0066] Step C3, sending the keywords generated by the large language model to a search engine, and the search engine performs an online search for the keywords; Optionally, key information, historical cases, expert opinions, etc. of property voting issues can be searched online based on keywords, thereby helping voters to understand the issues more comprehensively.
[0067] In step C4, the content retrieved by the search engine is summarized by the large language model and sent to the property voting system, and then sent to the property voting program by the property voting system.
[0068] In the specific implementation process, prompt words can be written for the large language model so that the large language model can have the ability to summarize the text.
[0069] In view of the problem that voters may find it difficult to understand the voting topics and specific contents during the property voting process, which is difficult to be provided by the existing property voting system, the present invention proposes to obtain the property voting topics in the property voting program through the constructed property voting system, and send the obtained voting topics to the analysis and retrieval system, and then the large language model in the analysis and retrieval system performs an online search for the property voting topics, and summarizes the results of the online search, so as to provide detailed background information and explanations according to the property voting theme, so as to help voters better understand the voting topics and specific contents and avoid blind voting.
[0070] The analysis of voting content is achieved by performing knowledge base retrieval and database query through the analytical retrieval system. The knowledge base retrieval is to query and interpret the professional knowledge involved in the voting process; the database query is to query and visualize the various data involved; Specifically, the knowledge base retrieval process is as follows: Step D1, in the property voting procedure, the voting personnel intercept the doubtful professional knowledge in the property voting content and send the intercepted professional knowledge to the property voting system; Step D2, the property voting system receives the professional knowledge sent by the property voting program, and sends the received professional knowledge to the analysis and retrieval system; In step D3, the text embedding model in the analysis and retrieval system converts the professional knowledge into a text vector, matches the converted text vector in the knowledge base, and returns the matching result.
[0071] Specifically, the text vector of professional knowledge is used as the query vector, and the text vector stored in the knowledge base is used as the storage vector. The matching method is to calculate the similarity between the query vector and the storage vector, and sort the documents according to the similarity score, and finally return the first few documents with the highest similarity score as the matching result.
[0072] The similarity calculation formula is: Where α is the query vector, is the storage vector, i is the number of the storage vector, and sim is the similarity.
[0073] In step D4, the matching results are summarized and integrated by the large language model in the analysis and retrieval system and sent to the property voting system, which then sends them to the property voting program.
[0074] In the specific implementation process, this step can still be done by writing prompt words for the large language model, so that the large language model has the ability to summarize and integrate the text.
[0075] The database query process is as follows: Step E1, in the property voting procedure, the voting personnel intercept the index parameters in question in the property voting content, and send the intercepted index parameters to the property voting system; Step E2, the property voting system receives the index parameters sent by the property voting program, and sends the received index parameters to the analysis and retrieval system; Step E3, the text embedding model in the analysis and retrieval system converts the index parameters into text vectors, matches the converted text vectors in the retrieval library, and returns the matching results.
[0076] Specifically, the text vector of the indicator parameter is used as the query vector, and the text vector stored in the retrieval library is used as the storage vector. The matching method is to calculate the similarity between the query vector and the storage vector, and sort the documents according to the similarity score, and finally return the first few indicator parameters with the highest similarity score as the matching result.
[0077] Step E4, the large language model in the analysis and retrieval system generates an SQL statement according to the matching index parameters, sends the generated SQL statement to the database to execute data query, and returns the result of the data query to the large language model; Specifically, the SQL generation rules can be used as prompt words for the large language model, so that the large language model can generate SQL statements according to the SQL generation rules, thereby having the function of generating SQL statements according to indicator parameters.
[0078] Step E5, the large language model visualizes the data query results and sends the visualization results to the property voting system, and the property voting system sends the visualization results to the property voting program.
[0079] Among them, the visualization of data query results is achieved by calling the data-to-image service. Specifically, the data-to-image service is provided by the constructed http server. The http service has a built-in data processing function, which can parse the read data and draw the parsed data into a chart. Furthermore, the drawn chart is returned to the caller.
[0080] In view of the professional knowledge and data involved in the property voting process, such as policies and regulations, operational data, etc., it is difficult or impossible to obtain these professional knowledge and data for the existing property voting system. Voters will intercept the questionable professional knowledge and indicator parameters in the property voting content, and send the intercepted professional knowledge and indicator parameters to the property voting system, and then the analysis and retrieval system will perform knowledge base retrieval and database query on the professional knowledge and indicator parameters. Furthermore, the retrieved professional knowledge will be integrated and summarized, and the queried data will be visualized, thus realizing the transparency and intuitiveness of the professional knowledge and data involved in the voting process, making the voting process more rational and objective.
[0081] So far, according to the method disclosed in the present invention, the working process of the present invention has been implemented once.
[0082] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.
[0083] It should be noted that the serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments. And the terms "including", "comprising" or any other variants thereof in this article are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "including a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.
Claims
1. A property voting method based on retrieval-enhanced generation, It is characterized in that Includes steps: Step S1, build a property voting system according to the designed technical framework; Specifically, the technical framework divides the voting end into three modules: government end, property end, and owner end; divides voting items into three modules: government item voting, property item voting, and voting item voting; and divides voting options into three modules: single choice, multiple choice, and voting; Step S2, building an analysis and retrieval system according to the application requirements of the property voting system; Step S3, establishing a communication connection between the property voting system and the analysis and retrieval system, and publishing the property voting program in the property voting system; Step S4, the property voting system and the analysis and retrieval system provide services to the property voting program to complete the property voting process.
2. The property voting method based on retrieval enhancement generation according to claim 1, It is characterized in that Step S2 includes the steps Step S201, deploying a large language model and a text embedding model in an analysis and retrieval system; Step S202, deploying the knowledge base of the property voting system in the analysis and retrieval system; Step S203, deploying the database of the property system in the analysis and retrieval system; Step S204, using the retrieval enhancement generation framework to connect the large language model, the text embedding model, the knowledge base and the database, and connecting the API of the search engine to the large language model.
3. The property voting method based on retrieval enhancement generation according to claim 2, It is characterized in that The knowledge base for deploying the property voting system described in step S202 specifically includes the following steps: Step A1, collect property-related documents; Step A2, converting the collected document data into text vectors; Specifically, the document data is converted into text vectors using a trained text embedding model. The text embedding model can learn how to map text vectors to index structures through training, so that similar vectors can be found quickly during retrieval. Step A3, storing the text vector into a vector database, so that the vector database after storing the text vector serves as a knowledge base.
4. The property voting method based on retrieval enhancement generation according to claim 2, It is characterized in that The database of the property system deployed in step S203 specifically includes the following steps: Step B1, design the table structure of the database according to the data of the property system; Step B2, select a database management system and create a database; Step B3, converting the indicator parameters in the database table structure into text vectors, storing the text vectors in a vector database, and using the vector database after storing the indicator parameter text vectors as a retrieval library; Step B4, inserting specific data into the database table according to the designed database table structure.
5. The property voting method based on retrieval enhancement generation according to claim 1, It is characterized in that The property voting system and the analysis and retrieval system described in step S4 provide services to the property voting program. Specifically, the services provided include answering voting issues and analyzing voting content, wherein answering voting issues is achieved by performing online retrieval through the analysis and retrieval system; analyzing voting content is achieved by performing knowledge base retrieval and database query through the analysis and retrieval system.
6. The property voting method based on retrieval enhancement generation according to claim 5, It is characterized in that The analysis of voting content specifically includes the following steps: Step C1, the property voting system obtains the property voting topic in the property voting program, and sends the obtained voting topic to the analysis and retrieval system; Step C2, the large language model in the analysis and retrieval system generates keywords related to the property voting topic based on the property voting topic; Step C3, sending the keywords generated by the large language model to a search engine, and the search engine performs an online search for the keywords; In step C4, the content retrieved by the search engine is summarized by the large language model and sent to the property voting system, and then sent to the property voting program by the property voting system.
7. The property voting method based on retrieval enhancement generation according to claim 5, It is characterized in that The execution of knowledge base retrieval specifically includes the steps of: Step D1, in the property voting procedure, the voting personnel intercept the doubtful professional knowledge in the property voting content and send the intercepted professional knowledge to the property voting system; Step D2, the property voting system receives the professional knowledge sent by the property voting program, and sends the received professional knowledge to the analysis and retrieval system; Step D3, the text embedding model in the analysis and retrieval system converts the professional knowledge into a text vector, matches the converted text vector in the knowledge base, and returns the matching result; In step D4, the matching results are summarized and integrated by the large language model in the analysis and retrieval system and sent to the property voting system, which then sends them to the property voting program.
8. The property voting method based on retrieval enhancement generation according to claim 7, It is characterized in that The converted text vector is matched in the knowledge base in step D3. Specifically, the text vector of professional knowledge is used as the query vector, and the text vector stored in the knowledge base is used as the storage vector. The matching method is to calculate the similarity between the query vector and the storage vector, and sort the documents according to the similarity score, and finally return the first few documents with the highest similarity score as the matching result. Specifically, the similarity calculation formula is: Where α is the query vector, is the storage vector, i is the number of the storage vector, and sim is the similarity.
9. The property voting method based on retrieval enhancement generation according to claim 5, It is characterized in that The execution of database query specifically includes the steps of: Step E1, in the property voting procedure, the voting personnel intercept the index parameters in question in the property voting content, and send the intercepted index parameters to the property voting system; Step E2, the property voting system receives the index parameters sent by the property voting program, and sends the received index parameters to the analysis and retrieval system; Step E3, the text embedding model in the analysis and retrieval system converts the index parameter into a text vector, matches the converted text vector in the retrieval library, and returns the matching result; Step E4, the large language model in the analysis and retrieval system generates an SQL statement according to the matching index parameters, sends the generated SQL statement to the database to execute data query, and returns the result of the data query to the large language model; Step E5, the large language model visualizes the data query results and sends the visualization results to the property voting system, and the property voting system sends the visualization results to the property voting program.
10. The property voting method based on retrieval enhancement generation according to claim 9, It is characterized in that The visualization of the data query results described in step E5 is achieved by calling the data-to-image service. Specifically, the data-to-image service is provided by a constructed http server, and the http service has a built-in data processing function. The data processing function can parse the read data and draw the parsed data into a chart, and further return the drawn chart to the caller.
11. A property voting system based on retrieval-enhanced generation, It is characterized in that The operational implementation of the system relies on the property voting method based on retrieval-enhanced generation as described in any one of claims 1-10. Specifically, the system includes a property voting system and an analysis and retrieval system, wherein the property voting system is used to publish property voting procedures; the analysis and retrieval system is used to provide the property voting system with answers to voting issues, and perform knowledge base retrieval and database query services.
Citation Information
Patent Citations
LLM-based insurance professional knowledge management system
CN118035418A
Knowledge retrieval enhancement generation method and system based on large language model
CN118394890A
Knowledge management question-answering system construction method and device
CN118446296A
