Large model session method and device based on data group and electronic equipment

Abstract

The invention discloses a large model session method and device based on a data group and electronic equipment. The method comprises the following steps: 1) a data group construction step: screening out data associated with a theme and a demand according to the theme and the demand of a session, establishing an association rule to associate the data, and temporarily combining to form a data group according to a preset rule; and 2) a large model session interaction step: retrieving data whose relevance with the user intention is greater than a preset value from the data group according to the user intention, constructing a frame structure of an answer in combination with the characteristics of the data group, and generating the answer output to the user by combining the frame structure with the detailed data in the data group. According to the large model session method based on the data group, multi-source and heterogeneous data are organically integrated into the data group according to a specific rule, so that the large model can fully absorb information nutrients from different levels, and a higher intelligent level is shown in session interaction.

Description

Technical Field

[0001] The present invention relates to the field of large model conversation technology, and in particular to a large model conversation method, device and electronic device based on data groups. Background Art

[0002] In today's digital age, big data and big model technologies are driving the transformation and development of various industries at an unprecedented speed. Big data is generated continuously from various fields with its massive, diverse and high-speed characteristics, becoming an indispensable strategic resource for enterprises, scientific research institutions and even the entire society. At the same time, big models, with their powerful learning ability, extensive knowledge reserves and excellent generalization performance, have demonstrated amazing strength in many fields such as natural language processing, image recognition, and intelligent decision-making, providing a new way to solve complex problems.

[0003] Traditional conversational methods often rely on simple rule matching or models trained with small-scale data, which are inadequate in the face of increasingly complex user needs and massive amounts of information. On the one hand, the rule matching method is less flexible and difficult to cope with diverse natural language expressions, which is prone to misunderstandings and leads to inaccurate or irrelevant answers; on the other hand, models trained with small-scale data have limited knowledge reserves and cannot fully tap into deep information associations, making it difficult to provide comprehensive and accurate answers.

[0004] For example, in the e-commerce field, when users inquire about detailed information, user experience, and comparison with similar products of a certain product, traditional conversational systems may only give general responses based on the basic description of the product, and are unable to provide in-depth insights by integrating multi-source data such as user reviews and market trends. In the medical and health field, when patients ask about treatment plans for their symptoms, traditional models based on limited medical knowledge find it difficult to combine the latest research results, clinical cases, etc. to give personalized and accurate suggestions, which may delay diagnosis and treatment of the disease.

[0005] In the early stages of data processing, manual screening of data is the most common method. Based on their own experience and established business needs, professionals screen one by one from numerous data tables and documents, and select the data they think is relevant to the problem. This process is like "looking for a needle in a haystack" in a vast database. Staff need to spend a lot of time familiarizing themselves with various data structures and contents before they can conduct targeted screening. For example, when financial institutions conduct risk assessments, analysts manually review transaction records, customer credit reports, market dynamic documents, etc. over the past few years to try to find clues that may indicate risks. After completing the screening, use basic data analysis tools or simple algorithm models to process these preliminarily selected data.

[0006] With the development of technology, general large models have begun to be widely used in data question-and-answer scenarios. Users directly input questions into general large models such as GPT-3 and Wenxin Yiyan, and the models directly generate answers based on the massive knowledge learned in the pre-training stage. When faced with some common sense questions, such as "What planets are there in the solar system?" and "When did Qin Shihuang unify the six kingdoms?", general large models can quickly give relatively accurate answers. When it comes to issues in the professional field within the enterprise, such as a manufacturing company asking "The trend and cause of the defective rate of production line A in the past quarter, refer to equipment maintenance records, raw material procurement batch data, and worker schedules", the general large model lacks an in-depth understanding of the specific data details and business logic of the enterprise, and can often only give broad and untargeted answers.

[0007] The disadvantages of manual data screening are very significant. On the one hand, the efficiency is extremely low. Faced with the current data volume of TB and PB, manual screening may take weeks or even months, seriously slowing down the business decision-making process. For example, when e-commerce companies conduct large-scale promotional activities, they need to analyze a large amount of user browsing, purchasing, and evaluation data. Manual operations are difficult to complete in a short period of time, and the best time to adjust the strategy is missed. On the other hand, manual screening is highly subjective. Different people have different judgment criteria for data relevance, and the selected data range and focus will also be biased, which makes the final data analysis results lack stability and reliability.

[0008] Although the general large model has strong generalization capabilities, it has obvious deficiencies in the application of professional fields. First, the knowledge of the general large model comes from a large-scale general corpus, and lacks in-depth understanding of the professional terminology, business processes and internal data of specific industries and enterprises, which leads to deviations or inaccuracies when answering professional questions. Secondly, it is difficult for the general large model to dynamically adjust according to the real-time data changes of the enterprise, and it is impossible to reflect the latest status of the business in a timely manner, and it is difficult to meet the enterprise's requirements for data timeliness.

[0009] In addition, general large models lack effective data integration and association capabilities when processing multi-source heterogeneous data, and are unable to fully tap the potential value between data, limiting their application effectiveness in complex business scenarios. Summary of the invention

[0010] In order to solve the difficulties of existing question-answering technology in professional field applications, the purpose of the present invention is to provide a large-model question-answering method based on domain business ontology enhancement, which provides users with accurate, efficient and explainable question-answering services by deeply integrating domain business ontology with the large model.

[0011] The purpose of the present invention and the solution to the technical problem are achieved by adopting the following technical solutions.

[0012] According to the first aspect of one or more embodiments of the present invention, a large model question answering method enhanced based on a domain business ontology includes the following steps:

[0013] 1) Data group construction step: According to the theme and requirements of the conversation, screen out the data associated with the theme and requirements, establish association rules according to the internal relationship and business logic between the data, associate the data, and then extract relevant data from the associated data, and temporarily combine them according to preset rules to form a data group;

[0014] 2) Large model conversation interaction step: Parse the input user statement, identify the user intention, associate the intention with the data group, retrieve data from the data group that has a relevance greater than a preset value with the user intention according to the user intention, analyze and mine the retrieved data to obtain the laws and trends of the data; According to the laws and trends of the data, combined with the characteristics of the data group, construct the framework structure of the answer, and generate the answer output to the user with the framework structure combined with the detailed data in the data group.

[0015] Optionally, in the large model conversation method based on a data group, in the data group construction step, screening out the data associated with the theme and requirements includes: integrating data from different departments, systems, and different formats, and screening out the data associated with the theme and requirements.

[0016] Optionally, in the data group construction step, after screening out the data associated with the theme and requirements and before establishing the association rules, it further includes: cleaning the screened data, and preprocessing the cleaned data so that the data is presented in a unified format and standard.

[0017] Optionally, preprocessing the cleaned data includes one or both of formatting and normalizing the data.

[0018] Optionally, in the large model conversation interaction step, parsing the input user statement and identifying the user intention includes: using natural language processing technology to parse the user input and identify the user intention.

[0019] Optionally, in the large model conversation interaction step, before identifying the user intention, it further includes: obtaining the context information of the conversation, parsing the background and purpose of the current question from the context information, and determining the user intention in combination with the parsing result of the input user statement.

[0020] Optionally, in the large model conversation interaction step, analyzing and mining the retrieved data to obtain the laws and trends of the data further includes: performing correlation analysis on data in different dimensions in the data to form a comprehensive analysis result.

[0021] Optionally, the framework structure for constructing an answer includes: based on the comprehensive analysis result and combined with the characteristics of the data group, constructing an answer framework.

[0022] According to a second aspect of one or more embodiments of the present application, a large model conversation device based on a data group, the device includes:

[0023] A data group construction unit, configured to screen out data associated with the topic and requirements according to the topic and requirements of the conversation, establish association rules according to the internal connection and business logic between the data, associate the data, and then extract relevant data from the associated data, and temporarily combine them according to preset rules to form a data group;

[0024] A large model conversation interaction unit, configured to parse the input user statement, identify the user intention, associate the intention with the data group, retrieve data from the data group that has a relevance greater than a preset value with the user intention according to the user intention, analyze and mine the retrieved data, and obtain the rules and trends of the data; according to the rules and trends of the data and combined with the characteristics of the data group, construct a framework structure for the answer, and generate an answer to be output to the user by combining the detailed data in the data group with the framework structure.

[0025] According to a third aspect of one or more embodiments of the present application, the present invention discloses an electronic device, including: a processor; and a memory for storing processor-executable instructions, wherein the processor realizes the method by running the executable instructions.

[0026] By means of the above technical solutions, the large model conversation method, device and electronic device based on the data group of the present invention at least have the following advantages:

[0027] 1) By organically integrating multi-source and heterogeneous data into a data group according to specific rules, the large model can fully absorb information nutrients from different levels by using the determined data range, the association relationship and fusion between the data, and thus show a higher intelligent level in the conversation interaction.

[0028] 2) Through this innovative method, not only can the user intention be accurately understood, but also detailed, accurate and personalized answers can be provided relying on rich data resources, effectively improving the quality of information services, improving the accuracy of data question answering, and supporting personalized answers according to the characteristics of the data group. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 Shows the overall flowchart of a large model conversation method based on a data group of the present invention.

[0030] Figure 2Shown is the structural diagram of a large model conversation device based on a data group according to the present invention.

[0031] Figure 3 Shown is a schematic structural diagram of an electronic device shown in an exemplary embodiment of the present invention.

[0032] Figure 4 Shown is a block diagram of a large model question and answer device based on a data group shown in an exemplary embodiment of the present application.

[0033] Figure 5 Shown is a schematic structural diagram of an electronic device shown in another embodiment of the present invention. Detailed implementation manners

[0034] The present invention will be further described in detail below through specific preferred embodiments in conjunction with the accompanying drawings, but the present invention is not limited to the following embodiments.

[0035] As Figure 1 shown, the present invention discloses a large model conversation method based on a data group, including the following steps:

[0036] 1) Data group construction step: According to the theme domain division, clarify the business scope of the conversation, and screen out various types of data related thereto from numerous data sources. These data may come from different departments, different systems, and even different formats. Establish association rules to associate the data, and temporarily combine them according to preset rules to form a data group;

[0037] 2) Large model conversation interaction step: Parse the input user statement, identify the user intention, associate the intention with the data group, retrieve data from the data group whose relevance to the user intention is greater than a preset value according to the user intention, analyze and mine the retrieved data to obtain the laws and trends of the data; According to the laws and trends of the data, combined with the characteristics of the data group, construct the framework structure of the answer, and generate the answer output to the user with the framework structure combined with the detailed data in the data group.

[0038] The large model conversation method based on a data group according to the present invention organically integrates multi-source and heterogeneous data into a data group according to specific rules, enabling the large model to fully absorb information nutrients from different levels, and thus showing a higher level of intelligence in conversation interaction.

[0039] "Data group" is an innovative concept born in the context of addressing the challenges of processing today's massive and complex data. It breaks the traditional way of processing individual data tables or documents in isolation, and temporarily combines multiple data tables and multiple documents into an organic whole according to specific rules. This means that instead of being limited to the limited information mining of a single data source, it focuses on integrating multi-source data and exploring the deep internal connections between data to maximize the value of data.

[0040] For example, in an e-commerce enterprise, data from different sources such as user order data tables, product detail documents, and user review documents are combined into a data group, which can comprehensively reflect comprehensive information such as user shopping behavior and product popularity, while a single order table can only present the thin dimension information of transaction records.

[0041] To bring light to solving the above dilemmas, the large model conversation method based on data groups of the present invention has emerged as the times require. This method organically integrates multi-source and heterogeneous data into a data group according to specific rules, enabling the large model to fully absorb information nutrients from different levels, and then demonstrating a higher level of intelligence in conversation interactions.

[0042] Through this innovative method, not only can it accurately understand user intentions, but also it can rely on rich data resources to provide detailed, accurate, and personalized answers, effectively improving the quality of information services.

[0043] In enterprise operations, it helps management make scientific decisions based on all-round data groups such as the market, competitors, and internal operations, promoting business development; in industries such as healthcare, education, and finance, it empowers professionals to deeply explore knowledge and provide customized services for customers, patients, students, etc., accelerating the process of digital transformation of the industry, which has extremely important practical significance and broad application prospects.

[0044] Example 1

[0045] A large model conversation method based on data groups specifically includes the following steps:

[0046] 1. Data group construction step

[0047] (1) Data source identification and integration

[0048] This step is the starting point of the entire data group construction, aiming to clarify which data needs to be processed. We will screen out various data related to the conversation topic and specific needs from numerous data sources. These data may come from different departments, different systems, and even different formats, such as text, tables, videos, etc. Identifying and integrating these scattered data sources together is for subsequent unified processing and analysis.

[0049] Function: It lays the foundation for subsequent data cleaning, preprocessing and correlation fusion, ensuring that we can fully obtain information related to the conversation topic and avoid inaccurate or incomplete analysis results due to data omissions.

[0050] (2) Data cleaning and preprocessing

[0051] In this step, we will clean the integrated data and remove invalid data, duplicate data, erroneous data, etc. to ensure the quality and accuracy of the data. Then we will pre-process the cleaned data, including formatting, normalization and other operations, so that the data can be presented in a unified format and standard, which is convenient for subsequent analysis and processing.

[0052] Function: Improve the quality and availability of data, create good conditions for subsequent data association, fusion and analysis, and avoid deviations in analysis results due to data quality issues.

[0053] (3) Data association and fusion

[0054] The purpose of this step is to associate and merge the cleaned and preprocessed data according to certain rules to form an organic whole. We will establish association rules based on the internal connections and business logic between the data, connect the data scattered in different data tables or documents, so that they can complement and verify each other, thus reflecting the overall picture of things more comprehensively and accurately.

[0055] Function: Break down barriers between data, realize data integration and sharing, provide a solid foundation for subsequent data group creation and analysis, and enable us to conduct in-depth mining and analysis of data from multiple dimensions and angles.

[0056] (4) Temporarily create data groups based on demand

[0057] During the conversation, according to the user's special needs, we will quickly extract relevant data from the already associated data and temporarily combine it into a data group according to specific rules. This data group is created to meet specific analysis and judgment needs. It contains the most relevant and valuable data for current needs, and can provide users with more accurate and targeted information support.

[0058] Function: Improve the flexibility and response speed of data processing, so that we can quickly generate the required data groups according to the real-time needs of users and provide users with more timely and effective services.

[0059] 2. Large model conversation interaction steps

[0060] (1) Understanding User Intent

[0061] This step is the first step for the large model to interact with users, aiming to accurately understand the intentions and needs of users. We will use natural language processing technology to parse the user's input, identify the user's intentions, and associate them with the data group, so as to provide accurate answers and services according to the user's intentions in the follow-up.

[0062] Function: Ensure that we can accurately grasp the needs of users, provide answers and solutions that better meet their expectations, and improve user satisfaction and interaction experience.

[0063] (2) Context Association and Data Group Confirmation

[0064] In this step, we will combine the context information of the conversation to more accurately understand the user's intentions and confirm the accuracy of the data group. Context information can help us understand what the user has asked before, as well as the background and purpose of the current question, so as to better understand the user's intentions and ensure that the data group can meet the actual needs of users.

[0065] Function: Improve the accuracy of understanding user intentions, ensure the accuracy and applicability of the data group, and provide more accurate and targeted answers and services for users.

[0066] (3) Precise Data Retrieval

[0067] In this step, we will accurately retrieve data highly relevant to the user's needs from the temporarily created data group according to the user's intentions and context association results. The purpose of this step is to ensure that we can find the data that best meets the user's needs and provide an accurate basis for subsequent analysis.

[0068] Function: Improve the accuracy and efficiency of data retrieval, ensure that we can quickly find the data required by users, and provide more timely and effective services for users.

[0069] (4) In-depth Analysis of the Data Group

[0070] In this step, we will conduct in-depth analysis on the retrieved data group to fully explore the deep connections and potential values among the data. We will use various data analysis techniques and methods, such as statistical analysis, machine learning, pattern recognition, etc., to conduct multi-dimensional analysis and mining on the data to reveal the laws and trends behind the data.

[0071] Function: Improve the depth and breadth of data analysis, enable us to more comprehensively and accurately understand the meaning and value of the data, and provide more in-depth and valuable insights and suggestions for users.

[0072] (5) Data Group Association Analysis

[0073] In this step, we will conduct correlation analysis on data from different dimensions to form a comprehensive analysis result. Through correlation analysis, we can discover the internal connections and mutual influences among different data, thereby understanding the overall picture of things more comprehensively and providing users with more comprehensive and integrated insights and suggestions.

[0074] Function: Improve the comprehensiveness and integrity of the analysis result, enabling us to deeply explore and analyze data from multiple dimensions and perspectives, and providing users with a more comprehensive and integrated solution.

[0075] (6) Construct a response framework based on the data group

[0076] In this step, we will construct the framework structure of the response according to the analysis result and in combination with the characteristics of the data group. The response framework is the skeleton of the response content, which stipulates the main content and structure of the response, enabling us to organize and present the response content in an orderly manner.

[0077] Function: Improve the logic and coherence of the response, enabling users to more clearly understand and grasp the main content and structure of the response, and enhancing the user's reading experience.

[0078] (7) Generate a detailed and personalized response

[0079] General description: In this step, we will generate a detailed, accurate and personalized response based on the response framework and in combination with the detailed data in the data group. We will according to the actual needs of the user.

[0080] Example 2

[0081] Figure 2 Shown is the structural diagram of a large model conversation device based on a data group of the present invention. As Figure 2 shown, the large model conversation device based on a data group includes: a data group construction unit 10 and a large model conversation interaction unit 20. Among them, the data group construction unit 10 is used to screen out data related to the theme and requirements of the conversation according to the theme and requirements of the conversation, establish association rules according to the internal connections and business logics among the data, associate the data, and then extract relevant data from the associated data and temporarily combine them according to preset rules to form a data group.

[0082] The large model conversation interaction unit 20 is used to parse the input user statements, identify the user intent, associate the intent with the data group, retrieve data from the data group that has a relevance greater than a preset value with the user intent according to the user intent, analyze and mine the retrieved data to obtain the rules and trends of the data; according to the rules and trends of the data, combined with the characteristics of the data group, construct the frame structure of the answer, and generate the answer output to the user with the detailed data in the data group combined with the frame structure.

[0083] For the specific implementation process of the functions and roles of each module in the above device, please refer to the implementation process of the corresponding steps in the above method for details, and will not be elaborated here.

[0084] Example 3

[0085] As Figure 3 shown, an embodiment of the present application further provides an electronic device, which includes a processor 301 and a memory 302. The memory 302 is used to store executable instructions that can be executed by the processor 301. Among them, the processor 301 realizes the above method by running the executable instructions.

[0086] Example 4

[0087] As Figure 4 shown, Figure 4 is a block diagram of a large model question and answer device based on a data group shown in an exemplary embodiment of the present application. The device mainly includes:

[0088] The data group construction module is used to temporarily combine according to preset rules to form a data group. The data group construction module includes data source identification and integration, data cleaning and preprocessing, data association and fusion, and temporarily creating a data group based on requirements.

[0089] The large model conversation interaction module is used to parse the statements input by the user and generate an answer based on the data group. The large model conversation interaction includes: user intent understanding, context association and data group confirmation, accurately retrieving data, deeply analyzing the data group, data group association analysis, constructing the answer frame of the data group, and generating a detailed and personalized answer. Among them, accurately retrieving data comes from temporarily creating a data group based on requirements in the data group construction module and deeply analyzing the data.

[0090] Example 5

[0091] As Figure 5As shown, an embodiment of the present application provides an electronic device, which includes an input device 501, a memory 502, a processor 503 and an output device 504. The input device 501 is used to obtain execution instructions, the memory 502 is used to store executable instructions of the processor 503, and the output device 504 is used to output execution results. The processor 503 implements the method described by running the executable instructions.

[0092] The devices or modules described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer, which may be in the form of a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email transceiver, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0093] The present invention introduces the concept of data groups, organically integrates multi-source, heterogeneous data into data groups according to specific rules, and uses the determined data range, the relationship and fusion between data to enable the large model to fully absorb information nutrients from different levels, thereby showing a higher level of intelligence in conversational interaction. Through this innovative approach, it can not only accurately understand user intentions, but also provide detailed, accurate and personalized answers based on rich data resources, effectively improve the quality of information services, improve the accuracy of data questions and answers, and support personalized answers based on the characteristics of data groups.

[0094] Application Example 1

[0095] Depth and breadth of data utilization:

[0096] Compared with the traditional way of processing a single data table or document in isolation, this method demonstrates excellent data integration capabilities. In the traditional model, data is often confined to their own "information islands", with limited analytical vision and difficulty in exploring deep connections. The large model conversation method based on data groups breaks down barriers and organically integrates multi-source and heterogeneous data according to specific rules to form a closely related organic whole.

[0097] In the enterprise decision-making support scenario, taking market strategy formulation as an example, the traditional method may only rely on simple sales reports and limited market research data for rough judgment, and the decision-making basis is weak. This method integrates the internal operation data of the enterprise, covering production, sales, inventory and other links, and at the same time widely absorbs external market dynamics, competitor intelligence, industry trends and other information to outline the overall market picture from all aspects.

[0098] Relying on such a rich data group, the large model deeply analyzes the interactions among various factors, accurately identifies market opportunities and potential risks, provides strong support for enterprises to formulate far-sighted market expansion, product R & D, and differential competition strategies, and helps enterprises move forward steadily in the complex and ever-changing market tide.

[0099] Application Example 2

[0100] Improvement in answer accuracy:

[0101] When traditional general large models face problems in professional fields, they often lack in-depth understanding of specific business logics, professional terms, and detailed data, resulting in broad and vague answers with poor accuracy. In contrast, the large model conversation method based on the data group is quite different. The data group and the large model work closely together, injecting strong impetus for accurate answers.

[0102] In the logistics industry, when users inquire about the reasons for cargo delays, the traditional method may only give a general speculation based on limited logistics track information, making it difficult to reach the root cause of the problem.

[0103] The data group constructed by the method of this embodiment covers multi-dimensional information such as order details, real-time status of transport vehicles, traffic conditions along the way, weather changes, and warehousing operation records. The large model deeply explores the associations among various links based on this.

[0104] By comprehensively analyzing the repair records of sudden vehicle breakdowns, the detention time in traffic congestion sections, the details of delays in warehouse goods in and out, and the specific impact of bad weather on the logistics process, the key factors causing delays are accurately located, and a detailed answer such as "Your goods were delayed because of heavy rain in [specific section], serious road waterlogging, blocked traffic for [X] hours, and the transport vehicle had a [specific type of breakdown] breakdown, taking [X] hours for repair, resulting in delays. It is expected that the subsequent delivery will be accelerated and it will be delivered at [specific time]" is given, with a significant improvement in accuracy and reliability.

[0105] The same is true in the financial field. When analyzing the risks of investment products, traditional methods mostly rely on basic materials such as product manuals and have insufficient insights into the ever-changing market fluctuations, the impacts of policy and regulatory adjustments, and the potential operational hazards within enterprises. This method integrates multiple data such as macroeconomic data, real-time financial market quotes, enterprise financial reports, regulatory policy dynamics, and industry risk cases to construct a data group. The large model accurately captures risk signals from it, comprehensively considers the interactive effects of various factors, and gives practical risk assessments and response strategies to protect investors and effectively avoid decision-making mistakes caused by one-sided information.

[0106] Application Example 3

[0107] Data integration and innovation:

[0108] The method of this embodiment breaks through the traditional data boundaries and realizes the organic integration of multi-source and heterogeneous data. Compared with the traditional mode of processing individual data tables or documents in isolation, it is a paradigm change in data utilization.

[0109] In the traditional way, data is scattered in various systems, the relevance is ignored, and the information value is greatly reduced. However, the method of this embodiment seamlessly integrates the data resources of the entire enterprise operation process, cross-departmental data, and even data of different themes and formats within the industry according to specific rules, broadening the breadth and depth of data utilization. Taking the supply chain management in the manufacturing industry as an example, the traditional mode only focuses on the bill of materials, inventory data in the production link, order information in the purchasing department, and order demand at the sales end. The data is isolated and it is difficult to collaborate, often resulting in inventory backlogs or shortages and delaying production and delivery.

[0110] The large model conversation method based on the data group is completely different. It organically integrates all-round data such as the quality data of raw materials from suppliers, the real-time operating parameters of production equipment, the logistics distribution track, the market demand forecast, and the customer after-sales feedback. When facing the risk of order delivery delay, the large model can quickly correlate the data of each link, accurately locate the root cause of the problem from multiple dimensions such as raw material supply delay, production equipment failure, logistics obstruction, and sudden change in market demand, and give a precise response strategy such as "due to the production process problem of the raw materials of supplier [specific supplier name], the delivery of [key raw material name] for this order is delayed by [X] days, affecting the production progress; at the same time, there is heavy rain on [specific section] during the logistics distribution, and the delay is expected to be [X] days. It is recommended to adjust the production plan in time, communicate with the customer and give priority to arranging urgent delivery", effectively improving the resilience and response speed of the supply chain and demonstrating the innovative value of data integration.

[0111] The above are only the preferred embodiments of the present invention, and do not impose any form of limitation on the present invention. Therefore, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention without departing from the content of the technical solution of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. A large model conversation method based on data groups, characterized in that: The following steps are involved: 1) Data group construction step: According to the topic and needs of the conversation, filter out the data associated with the topic and needs, establish association rules based on the inherent connection and business logic between the data, associate the data, and then extract relevant data from the associated data, and temporarily combine them according to the preset rules to form a data group; 2) Large model conversation interaction steps: parse the input user sentences, identify the user intention, associate the intention with the data group, retrieve data from the data group that has a correlation with the user intention greater than a preset value based on the user intention, analyze and mine the retrieved data to obtain the patterns and trends of the data; construct a framework structure for the answer based on the patterns and trends of the data and in combination with the characteristics of the data group, and use the framework structure in combination with the detailed data in the data group to generate an answer to be output to the user.

2. The large model conversation method based on data groups as claimed in claim 1, characterized in that: In the data group construction step, screening out data related to the subject and demand includes: integrating data from different departments, systems, and different formats, and screening out data related to the subject and demand.

3. The large model conversation method based on data groups as claimed in claim 2, characterized in that: In the data group construction step, after filtering out the data associated with the subject and demand and before establishing the association rules, it also includes: cleaning the filtered data and preprocessing the cleaned data so that the data is presented in a unified format and standard.

4. The large model conversation method based on data groups as claimed in claim 3, characterized in that: Preprocessing the cleaned data includes one or both of formatting and normalizing the data.

5. The large model conversation method based on data groups as claimed in claim 4, characterized in that: In the large model conversation interaction step, parsing the input user sentences and identifying the user intentions includes: parsing the user inputs using natural language processing technology and identifying the user intentions.

6. The large model conversation method based on data groups as claimed in claim 5, characterized in that: In the large model conversation interaction step, before identifying the user intention, it also includes: obtaining the context information of the conversation, parsing the background and purpose of the current question from the context information, and determining the user intention in combination with the parsing result of the input user sentence.

7. The large model conversation method based on data groups as claimed in claim 6, characterized in that: In the large model conversation interaction step, the retrieved data is analyzed and mined to obtain the patterns and trends of the data, which also includes: correlation analysis of data of different dimensions in the data to form a comprehensive analysis result.

8. The large model conversation method based on data groups as claimed in claim 7, characterized in that: The framework structure for constructing the answer includes: constructing the answer framework based on the comprehensive analysis results and combining the characteristics of the data group.

9. A large model conversation device based on a data group, characterized in that: The device comprises: The data group construction unit is used to filter out data associated with the topic and demand according to the topic and demand of the conversation, establish association rules according to the internal connection and business logic between the data, associate the data, and then extract relevant data from the associated data, and temporarily combine them according to preset rules to form a data group; The large model conversation interaction unit is used to parse the input user sentences, identify the user intention, associate the intention with the data group, retrieve the data with a correlation with the user intention greater than a preset value from the data group according to the user intention, analyze and mine the retrieved data to obtain the rules and trends of the data; according to the rules and trends of the data, combined with the characteristics of the data group, construct a framework structure of the answer, and use the framework structure combined with the detailed data in the data group to generate an answer output to the user.

10. An electronic device, characterized in that: include: processor; as well as A memory for storing processor-executable instructions, wherein the processor implements the method according to any one of claims 1 to 8 by executing the executable instructions.

Images