Accurate data pushing method and system based on working scene

Abstract

The invention belongs to the field of natural language processing, and particularly relates to a precise data pushing method and system based on a working scene, and the method comprises the following steps: 1, creating the working scene; 2, assigning a work scene to ask questions; 3, preliminary question and answer pair matching; 4, understanding and supplementing problems; 5, deep retrieval and answer integration are carried out; 6, performing answer evaluation and feedback circulation; according to the method, a multi-dimensional working scene framework is constructed, the knowledge category is defined, and the carrying and utilization of working scene identifiers in the question and answer process are realized by collecting and integrating question and answer pair data only aiming at the scenes. The step ensures that the system can immediately identify the working environment behind the system when receiving the query, and further performs deep retrieval in a highly focused and highly targeted knowledge base, so as to output accurate answers highly matched with the query scene instead of wide and non-targeted information.

Description

Technical Field

[0001] The present invention belongs to the field of natural language processing, and relates to technical fields such as large language models, information extraction, and text classification, and in particular to a precise data push method based on a work scenario. Background Art

[0002] In the context of the information age, knowledge base question answering systems, as a key technology for information retrieval, are widely used in various application scenarios, aiming to quickly and accurately solve users' query needs. Despite significant progress, these systems still face challenges when dealing with highly specialized or industry-specific problems. Traditional knowledge bases are built on a wide range of data sources and are all-encompassing, but this breadth often sacrifices depth, resulting in feedback that may be too general and lacks specificity when answering questions in professional fields. For example, in fields such as medical diagnosis, financial analysis, or engineering, the questions asked by users are often closely related to specific workflows and industry specifications, and they expect to obtain precise answers that are directly applicable to actual operations rather than general theoretical overviews.

[0003] To make up for this shortcoming, researchers have tried to optimize the quality of answers by enhancing natural language understanding and context awareness, but this requires advanced algorithm support and is difficult to fully cover all professional details. Therefore, there is an urgent need for an innovative method that can retain the flexibility of the knowledge base question answering system while enhancing its expressiveness and accuracy in specific fields and situations. Summary of the invention

[0004] Technical problem solved by the present invention: The present invention is used to solve the problem that traditional knowledge bases are built on a wide range of data sources and are all-encompassing, but this breadth often sacrifices depth, resulting in feedback that is too general and lacks specificity when answering questions in professional fields.

[0005] The purpose of the present invention is to propose a method for pushing accurate data based on work scenarios. The core of this method is to build a multi-dimensional work scenario framework, which not only defines the scope of knowledge, but also collects and integrates question-answer data only for these scenarios, thereby realizing the carrying and utilization of work scenario identifiers in the question-answering process. This step ensures that when the system receives a query, it can instantly identify the work environment behind it, and then conduct in-depth searches in a highly focused and targeted knowledge base, thereby outputting accurate answers that are highly consistent with the query scenario, rather than broad and non-targeted information.

[0006] Technical solution of the present invention

[0007] A method for accurately pushing data based on a work scenario includes the following steps:

[0008] Step 1: Create a working scenario;

[0009] Step 2: Ask questions about specific work scenarios;

[0010] Step 3: Preliminary question-answer pair matching;

[0011] Step 4: Problem understanding and supplement;

[0012] Step 5: Deep retrieval and answer integration;

[0013] Step 6: Answer evaluation and feedback loop.

[0014] Further, step 1 specifically includes the following steps:

[0015] Step 1.1 Create a work scenario and assign an identifier;

[0016] In the knowledge engineering system, work scenarios for specific fields are created based on detailed analysis of business processes, business requirements, and work environments. Once created, unique identifiers are assigned to the work scenarios.

[0017] Step 1.2 Data collection and upload

[0018] Data collection is carried out through two channels: on the one hand, manually compiled question-answer pairs are collected; on the other hand, document data closely related to the scenario is collected. After uploading the system, the document content is extracted and preprocessed.

[0019] Furthermore, in step 1.1, a hash algorithm based on the key attributes of the scene is used to combine the scene name, main business content, and the role information of the personnel involved into a string, and a 256-bit hash value is generated as an identifier using the SHA-256 algorithm to ensure the uniqueness of each work scene in the system;

[0020] In step 1.2, the document content is preprocessed in the following ways: format conversion and data cleaning.

[0021] Further, step 2 specifically includes the following steps:

[0022] Step 2.1 Front-end display and data transmission

[0023] The front-end interface and the back-end interface are seamlessly connected. The front-end interface displays the established work scenario to the user. When the user enters a question, the front-end calls the back-end interface and transmits the question content itself, along with the work scenario identifier and system user information.

[0024] Step 2.2 Backend response and preparation

[0025] After receiving the data transmitted by the front end, the back end responds quickly, parses the work scenario identifier, and prepares the corresponding data processing channel based on the identifier, including configuring the corresponding database connection according to the work scenario and loading specific models and algorithms.

[0026] Further, step 3 is specifically as follows: after receiving a question with a scene identifier, the intelligent knowledge middle platform system performs a preliminary match in the knowledge base corresponding to the scene identifier; in the preliminary matching process, the intelligent knowledge middle platform system adopts approximate string matching and semantic similarity calculation, wherein, for approximate string matching, the edit distance algorithm Levenshtein is used to calculate the similarity between the question and the question part in the question-answer pair; for semantic similarity calculation, the pre-trained word vector model Word2Vec is used to convert the question and the question in the question-answer pair into a vector representation, and then the semantic similarity is measured by calculating the cosine similarity between the vectors;

[0027] Furthermore, in step 3, the edit distance algorithm Levenshtein supports fuzzy query and synonym replacement by constructing a synonym dictionary and fuzzy matching rules.

[0028] Further, step 4 includes: matching question processing: for questions that can be directly matched, in addition to matching based on question and answer pairs, a model-based question and answer method is used to semantically understand and analyze the questions;

[0029] Processing of unmatched questions: For questions that cannot be directly matched, the intelligent knowledge middle platform system builds the complete context of the question by analyzing the user's historical interaction data.

[0030] Furthermore, unmatched questions are specifically as follows: For questions that cannot be directly matched, the intelligent knowledge middle platform system determines the core intention of the user's question, extracts the key entities in the question, and analyzes the user's state changes throughout the conversation, so as to enhance the explanatory power of the question and the targeted nature of the retrieval.

[0031] Further, step 5 includes the following steps:

[0032] Step 5.1 Deep search

[0033] The search engine integrates an information retrieval model. During the retrieval process, for enhanced questions, the information retrieval model is used to extract key information from the documents and question-and-answer records associated with the work scenario. The information retrieval model calculates the relevance score based on the frequency of occurrence of the keywords in the question in the document and the length of the document, and measures the importance of the keywords by calculating the word frequency and inverse document frequency. The information retrieval model uses its powerful semantic understanding ability to perform deep matching between questions and document content.

[0034] Step 5.2 Answer Integration

[0035] After extracting key information, the answer is integrated, including sentence generation and logical coherence processing.

[0036] Further, step 6 includes the following steps:

[0037] Step 6.1 Answer Evaluation

[0038] After displaying the answer, the front end provides users with an answer evaluation channel. The answer evaluation is conducted from two dimensions: satisfaction and accuracy. Satisfaction is measured by the user's subjective evaluation, while accuracy is judged by comparing with the standard answer or the evaluation results of domain experts.

[0039] Step 6.2 Feedback Optimization

[0040] Adjust matching strategies and retrieval algorithms based on user feedback to continuously improve the performance and quality of question and answering.

[0041] A precise data push system based on a work scenario, applied to the precise data push method, the system comprising:

[0042] Work scenario creation module: configured to create work scenarios and assign identifiers and collect and upload data;

[0043] Questioning module: configured for front-end display and data transmission, and back-end response and preparation;

[0044] Preliminary question-answer pair matching module: After receiving a question with a scenario identifier, the intelligent knowledge middle platform system is configured to perform a preliminary match in the knowledge base corresponding to the scenario identifier;

[0045] Question understanding and supplementation module: It is configured to use a model-based question-answering method to perform semantic understanding and analysis on questions that can be directly matched, in addition to matching based on question-answer pairs; and for questions that cannot be directly matched, the intelligent knowledge middle platform system constructs the complete context of the question by analyzing the user's historical interaction data;

[0046] Deep retrieval and answer integration module: configured for deep retrieval and answer integration;

[0047] Answer evaluation and feedback loop module: configured for answer evaluation and answer feedback.

[0048] Beneficial Effects

[0049] The present invention proposes a method for pushing accurate data based on work scenarios. Through precise scenario positioning, it achieves accurate matching of user queries and professional field knowledge, ensuring that the returned answers are highly targeted and practical, thereby surpassing the general answers in traditional question-and-answer systems and greatly improving the accuracy and value of information. Specifically, it first optimizes the scope of knowledge through the detailed division of work scenarios, and effectively focuses retrieval on related fields, which not only significantly enhances search efficiency, but also makes the answers to professional questions more in-depth and accurate. In addition, the technology uses work scenario identification to intelligently identify the user's personalized demand background, and provides customized information services based on this. Especially when dealing with complex or highly specialized queries, it can fully meet the user's urgent needs for accurate information, greatly enhancing the user experience and satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is a method flow chart of the method of the present invention;

[0051] Figure 2 Data flow diagram for answer retrieval in the present invention. DETAILED DESCRIPTION

[0052] The features and illustrative embodiments of various aspects of the present invention will be described in detail below. In the detailed description below, many specific details are proposed to provide a comprehensive understanding of the present invention. However, it is obvious to those skilled in the art that the present invention can be implemented without the need for some of these specific details. The following description of the embodiments is only to provide a better understanding of the present invention by illustrating examples of the present invention. The present invention is by no means limited to any specific settings and methods proposed below, but covers any improvements, replacements and modifications of structures, methods, devices without departing from the spirit of the present invention. In the accompanying drawings and the following description, known structures and technologies are not shown to avoid unnecessary ambiguity in the present invention.

[0053] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other, and the embodiments can refer to and quote each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0054] An embodiment of the present invention is to provide a method for accurately pushing data based on a work scenario, comprising the following steps:

[0055] Step 1: Build a working scene

[0056] Create work scenarios and assign identifiers:

[0057] In the knowledge engineering system, work scenarios in specific fields are created based on detailed analysis of business processes, business requirements, and work environments. For example, for the creation of work scenarios in the field of aviation maintenance, factors such as the type of aircraft to be maintained, the different stages of maintenance (such as daily maintenance, fault repair), and the maintenance tools and equipment involved need to be considered. After creation, a unique identifier is assigned to it through a specific algorithm. A hash algorithm based on the key attributes of the scene can be used. For example, key information such as the scene name, main business content, and the roles of the personnel involved can be combined into a string, and a 256-bit hash value is generated using the SHA-256 algorithm as an identifier to ensure the uniqueness of each work scenario in the system.

[0058] Data collection and upload:

[0059] Data collection is mainly carried out through two channels. On the one hand, we collect carefully compiled question-answer pairs, which are written by experts in the field based on their long-term practical experience and professional knowledge, ensuring the authority and practicality of the data content. On the other hand, we collect documents closely related to the scenario, such as industry standard documents, operation manuals, fault case reports, etc. After uploading to the system, text extraction technology will be used to extract the document content, and necessary preprocessing will be performed, such as format conversion, data cleaning, etc., for subsequent processing.

[0060] Step 2: Ask questions in specific work scenarios

[0061] Front-end display and data transmission:

[0062] The front-end interface needs to be seamlessly connected with the back-end interface. The front-end interface presents the work scenarios established by the system to the user in a clear and intuitive way, such as through drop-down menus, classification lists, etc., to facilitate users to specify work scenarios. When the user enters a question, the front-end calls the back-end interface, in addition to passing the question content itself, it also needs to carry the work scenario identifier and system user information. This information will be encapsulated in a specific data format (such as JSON format), where user information includes the user's identity, permission level, and other content.

[0063] Backend response and preparation:

[0064] After receiving the data transmitted by the front end, the back end needs to respond quickly, parse the work scenario identifier, and prepare the corresponding data processing channel according to the identifier. This involves configuring the corresponding database connection according to the work scenario, loading specific models and algorithms, etc., to ensure the efficiency and accuracy of subsequent processing.

[0065] Step 3: Preliminary question-answer pair matching

[0066] After receiving a question with a scene identifier, the intelligent knowledge middle platform system performs a preliminary match in the knowledge base corresponding to the scene identifier. In this process, advanced technologies such as approximate string matching and semantic similarity calculation are used. For approximate string matching, the edit distance algorithm Levenshtein can be used to calculate the similarity between the question and the question part of the question and answer pair. For semantic similarity calculation, the pre-trained word vector model Word2Vec is used to convert the question and the question in the question and answer pair into a vector representation, and then the semantic similarity is measured by calculating the cosine similarity between the vectors. At the same time, the matching algorithm supports fuzzy query and synonym replacement, and improves the matching accuracy by building a synonym dictionary and fuzzy matching rules. For example, for the question of "aircraft landing gear failure repair", when there is a question and answer pair of "aircraft landing gear damage repair" in the knowledge base, it can be successfully matched through synonym replacement and fuzzy matching.

[0067] Step 4: Problem understanding and supplementation

[0068] Matching problem handling:

[0069] For questions that can be matched directly, in addition to question-answer pair matching, model-based question-answering technology is also used. For example, the pre-trained language model GPT is used to semantically understand and analyze the question to further confirm the accuracy of the match.

[0070] Unmatched problem handling:

[0071] For questions that are not directly matched, the intelligent knowledge middle platform system uses natural language understanding and dialogue management technology to build a complete context for the question by deeply analyzing the user's historical interaction data. Specifically, the core intention of the user's question is determined through intent recognition technology, and the key entities in the question (such as aircraft model, faulty parts, etc.) are extracted using entity extraction technology. In combination with dialogue state tracking technology, the user's state changes throughout the dialogue process are analyzed to enhance the explanatory power of the question and the pertinence of the retrieval. For example, if the user asks about different failure phenomena of an aircraft engine in succession, when asking related questions again, the system can better understand the background of the question based on the historical dialogue state.

[0072] Step 5: Deep retrieval and answer integration

[0073] Deep Search:

[0074] The search engine integrates advanced information retrieval models such as BM25, TF-IDF, BERT, etc. In the retrieval process, for enhanced questions, these models are used to extract key information from a large number of documents and question-and-answer records associated with the work scenario. It calculates the relevance score based on factors such as the frequency of occurrence of keywords in the question in the document and the length of the document; measures the importance of keywords by calculating the term frequency (TF) and inverse document frequency (IDF); and uses the powerful semantic understanding ability of the BERT model to deeply match the question and document content.

[0075] Answer integration:

[0076] After extracting key information, natural language generation technology is used to integrate answers. This involves operations such as sentence generation and logical coherence processing. For example, according to the different information fragments retrieved, they are organized into smooth and coherent sentences in a reasonable logical order to ensure the coherence and readability of the output answers, while ensuring that the answers are highly relevant to the questions and conform to the actual situation of the work scenario.

[0077] Step 6: Answer Evaluation and Feedback Loop

[0078] Answer evaluation:

[0079] After displaying the answer, the front end provides users with an answer evaluation channel. Answer evaluation is mainly carried out from two dimensions: satisfaction and accuracy. Satisfaction is measured by the user's subjective evaluation (such as satisfaction, dissatisfaction), and accuracy is judged by comparing with the standard answer (if it exists) or the evaluation results of domain experts. At the same time, the system will automatically collect some evaluation indicators, such as the completeness of the answer, logical clarity, etc.

[0080] Feedback optimization:

[0081] Based on user feedback, the system will adjust the matching strategy and retrieval algorithm. For example, if a user is not satisfied with the accuracy of an answer, the system will analyze the problem, which may be a semantic understanding deviation in the matching process or an incomplete retrieval result, and then optimize the relevant algorithms in a targeted manner, such as adjusting the weight of semantic similarity calculation, improving the parameters of the information retrieval model, etc., to continuously improve the performance and quality of question and answer.

[0082] The second embodiment of the present invention is a method for pushing accurate data based on work scenarios. Under the intelligent knowledge middle platform system, the technical solution for pushing accurate data based on work scenarios is carefully designed and implemented. Figure 1 and Figure 2 As shown below, we will take "travel reimbursement" as an example to specifically demonstrate the efficiency and delicacy of its operation process, which includes the following steps:

[0083] First, the technical team carefully constructed a work scenario called "Travel Expenses" in the middle office system and gave it a unique identifier "Travel expenses". This scenario not only defines the boundaries of the discussion, but also clarifies the scope of knowledge coverage. Key information such as "Business Trip Accommodation Standards" and "Business Trip Application Process" are included. In order to increase the direct hit rate of questions and answers, the team also preset a series of question and answer pairs for this scenario. For example, in response to the question "What are the accommodation reimbursement standards for senior engineers on business trips to Beijing?", the system is ready to respond directly: "XX yuan".

[0084] The user started the interaction in the "travel reimbursement" scenario, first asking the question: "Standards for accommodation reimbursement for senior engineers on business trips to Beijing." The system quickly identified the scenario identifier "Travel expenses" and directly pushed the answer "XX yuan" using the preset question-answer pair. Then, based on the answer to the previous question, the user further explored: "What about Tianjin?" This time, the question failed to find a direct match in the existing question-answer pairs.

[0085] Faced with this unpredictable question, the technical solution demonstrated its flexibility and intelligence. The question-answering search engine intervened, using the previous conversation history - namely "Senior Engineer's Accommodation Reimbursement Standards for Business Trips to Beijing" and its answers - as context clues, automatically completing and refining the new question into "Senior Engineer's Accommodation Reimbursement Standards for Business Trips to Tianjin". Subsequently, in the limited "travel reimbursement" scenario knowledge base and question-answer pairs, the search engine conducted in-depth mining and information integration, and finally refined and integrated the answer "The text does not provide relevant information on the accommodation standards for senior engineers on business trips to Tianjin. However, based on the accommodation standards for business trips, it can be inferred that the accommodation standards for senior engineers on business trips are generally. For Tianjin, it belongs to the "other regions" category, so the accommodation fee standard for senior engineers is XX yuan / day", and immediately fed back to the user through the interface. After getting the answer, the user evaluates and gives feedback on the satisfaction with the answer.

[0086] The whole process not only demonstrated the system's accurate grasp of the work scenario, but also demonstrated its intelligent completion and efficient retrieval capabilities when dealing with continuous and deep-seated problems. It truly provided personalized and high-precision information services in specific work scenarios, significantly improving user experience and work efficiency.

[0087] The present invention achieves accurate matching of user queries and professional field knowledge through scenario-based precise positioning, ensuring that the returned answers are highly targeted and practical, thus surpassing the general answers in traditional question-and-answer systems and greatly improving the accuracy and value of information. Specifically, it first optimizes the scope of knowledge through the detailed division of work scenarios, effectively focusing retrieval on related fields, which not only significantly enhances search efficiency, but also makes the answers to professional questions more in-depth and accurate. In addition, the technology uses work scenario identification to intelligently identify the user's personalized demand background, and provides customized information services accordingly. Especially when dealing with complex or highly specialized queries, it can fully meet the user's urgent needs for accurate information, greatly enhancing the user experience and satisfaction.

[0088] The third embodiment of the present invention provides a precise data push system based on a work scenario, the system comprising:

[0089] Work scenario creation module: configured to create work scenarios, assign identifiers, and collect and upload data;

[0090] Questioning module: configured for front-end display and data transmission, and back-end response and preparation;

[0091] Preliminary question-answer pair matching module: After receiving a question with a scenario identifier, the intelligent knowledge middle platform system performs a preliminary match in the knowledge base corresponding to the scenario identifier;

[0092] Question understanding and supplementation module: It is configured to use a model-based question-answering method to perform semantic understanding and analysis on questions that can be directly matched, in addition to matching based on question-answer pairs; and for questions that cannot be directly matched, the intelligent knowledge middle platform system constructs the complete context of the question by analyzing the user's historical interaction data;

[0093] Deep retrieval and answer integration module: configured for deep retrieval and answer integration;

[0094] Answer evaluation and feedback loop module: configured for answer evaluation and answer feedback.

[0095] In addition, the fourth embodiment of the present invention further provides a computer storage medium, in which computer executable instructions are stored, and the computer executable instructions are used to execute the aforementioned precise data push based on the work scenario.

[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the field can easily think of various equivalent modifications or substitutions within the technical scope disclosed by the present invention, and these modifications or substitutions should all be covered within the protection scope of the present invention.

Claims

1. A precise data push method based on work scenarios, characterized in that: The following steps are involved: Step 1: Create a working scenario; Step 2: Ask questions about specific work scenarios; Step 3: Preliminary question-answer pair matching; Step 4: Problem understanding and supplement; Step 5: Deep retrieval and answer integration; Step 6: Answer evaluation and feedback loop.

2. The method for pushing accurate data based on working scenarios according to claim 1 is characterized in that: Step 1 specifically includes the following steps: Step 1.1 Create a work scenario and assign an identifier; In the knowledge engineering system, work scenarios for specific fields are created based on detailed analysis of business processes, business requirements, and work environments. Once created, unique identifiers are assigned to the work scenarios. Step 1.2 Data collection and upload Data collection is carried out through two channels: on the one hand, manually compiled question-answer pairs are collected; on the other hand, document data closely related to the scenario is collected. After uploading the system, the document content is extracted and preprocessed.

3. The method for accurately pushing data based on working scenarios according to claim 2 is characterized in that: In step 1.1, a hash algorithm based on the key attributes of the scene is used to combine the scene name, main business content, and the role information of the personnel involved into a string, and the SHA-256 algorithm is used to generate a 256-bit hash value as an identifier to ensure the uniqueness of each work scene in the system; In step 1.2, the document content is preprocessed in the following ways: format conversion and data cleaning.

4. The method for pushing accurate data based on working scenarios according to claim 3 is characterized in that: Step 2 specifically includes the following steps: Step 2.1 Front-end display and data transmission The front-end interface and the back-end interface are seamlessly connected. The front-end interface displays the established work scenario to the user. When the user enters a question, the front-end calls the back-end interface and transmits the question content itself, along with the work scenario identifier and system user information. Step 2.2 Backend response and preparation After receiving the data transmitted by the front end, the back end responds quickly, parses the work scenario identifier, and prepares the corresponding data processing channel based on the identifier, including configuring the corresponding database connection according to the work scenario and loading specific models and algorithms.

5. The method for pushing accurate data based on working scenarios according to claim 4 is characterized in that: Step 3 is as follows: after receiving a question with a scene identifier, the intelligent knowledge middle platform system performs a preliminary match in the knowledge base corresponding to the scene identifier; in the preliminary matching process, the intelligent knowledge middle platform system uses approximate string matching and semantic similarity calculation. For approximate string matching, the edit distance algorithm Levenshtein is used to calculate the similarity between the question and the question part in the question-answer pair; for semantic similarity calculation, the pre-trained word vector model Word2Vec is used to convert the question and the question in the question-answer pair into a vector representation, and then the cosine similarity between the vectors is calculated to measure the semantic similarity; In step 3, the edit distance algorithm Levenshtein supports fuzzy query and synonym replacement by building a synonym dictionary and fuzzy matching rules.

6. The method for accurately pushing data based on working scenarios according to claim 5 is characterized in that: Step 4 includes: Matching question processing: For questions that can be directly matched, in addition to matching based on question and answer pairs, a model-based question and answer method is used to semantically understand and analyze the questions; Processing of unmatched questions: For questions that cannot be directly matched, the intelligent knowledge middle platform system builds the complete context of the question by analyzing the user's historical interaction data.

7. The method for accurately pushing data based on working scenarios according to claim 6 is characterized in that: Unmatched questions are specifically: For questions that cannot be directly matched, the intelligent knowledge middle platform system determines the core intention of the user's question, extracts the key entities in the question, and analyzes the user's state changes throughout the conversation, so as to enhance the explanatory power of the question and the targeted nature of the retrieval.

8. The method for accurately pushing data based on working scenarios according to claim 7 is characterized in that: Step 5 includes the following steps: Step 5.1 Deep search The search engine integrates an information retrieval model. During the retrieval process, for enhanced questions, the information retrieval model is used to extract key information from the documents and question-and-answer records associated with the work scenario. The information retrieval model calculates the relevance score based on the frequency of occurrence of the keywords in the question in the document and the length of the document, and measures the importance of the keywords by calculating the word frequency and inverse document frequency. The information retrieval model uses its powerful semantic understanding ability to perform deep matching between questions and document content. Step 5.2 Answer Integration After extracting key information, the answer is integrated, including sentence generation and logical coherence processing.

9. The method for accurately pushing data based on working scenarios according to claim 8 is characterized in that: Step 6 includes the following steps: Step 6.1 Answer Evaluation After displaying the answer, the front end provides users with an answer evaluation channel. The answer evaluation is conducted from two dimensions: satisfaction and accuracy. Satisfaction is measured by the user's subjective evaluation, while accuracy is judged by comparing with the standard answer or the evaluation results of domain experts. Step 6.2 Feedback Optimization Adjust matching strategies and retrieval algorithms based on user feedback to continuously improve the performance and quality of question and answering.

10. A precise data push system based on work scenarios, applied to the precise data push method according to claim 9, characterized in that: The system comprises: Work scenario creation module: configured to create work scenarios and assign identifiers and collect and upload data; Questioning module: configured for front-end display and data transmission, and back-end response and preparation; Preliminary question-answer pair matching module: After receiving a question with a scenario identifier, the intelligent knowledge middle platform system is configured to perform a preliminary match in the knowledge base corresponding to the scenario identifier; Question understanding and supplementation module: It is configured to use a model-based question-answering method to perform semantic understanding and analysis on questions that can be directly matched, in addition to matching based on question-answer pairs; and for questions that cannot be directly matched, the intelligent knowledge middle platform system constructs the complete context of the question by analyzing the user's historical interaction data; Deep retrieval and answer integration module: configured for deep retrieval and answer integration; Answer evaluation and feedback loop module: configured for answer evaluation and answer feedback.

Images