Large model industry chain intelligent mining algorithm and system based on credential executable environment

Abstract

The invention relates to a large model industry chain intelligent mining algorithm and system based on a credential executable environment. The system comprises a sensing module used for sensing a user problem and performing task splitting on the sensed user problem to obtain a target task used for analyzing upstream and downstream of each product in an industrial chain, and the target task is a task which can be directly processed by a large language model; the action module comprises calling tools, and the calling tools comprise a knowledge acquisition tool needing to be called when the large language model carries out task processing and a storage tool for storing industrial chain information obtained when the large language model processes the target tasks; and the brain module is used for processing each target task based on the big language model and each calling tool provided by the action module to obtain industrial chain information corresponding to the industrial chain. By the adoption of the system, complex tasks can be processed, and complete answers can be given.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to an intelligent mining algorithm and system for a large-model industrial chain based on an ICT executable environment. Background Art

[0002] The development of artificial intelligence (AI), particularly breakthroughs in large-scale pre-trained language models, has brought new solutions and perspectives to various industries. In the securities industry, investors face challenges such as accelerated market rotation and increasing demand for thematic investments. They urgently need efficient tools to quickly understand and identify various investment information across the industry chain. Large language models, with their ability to rapidly process large amounts of text and provide preliminary analytical results, are an ideal choice for assisting industry chain research.

[0003] Using large language models for industry chain analysis can significantly shorten response times and quickly generate preliminary results. While their accuracy may not be as high as manual analysis, manual review and correction based on the model's results can still significantly improve the speed and flexibility of investment decisions. More importantly, large language models can provide additional information dimensions based on big data, offering investors novel investment perspectives and strategies.

[0004] Although large language models have shown great potential, they face certain limitations in practical applications, especially when dealing with complex tasks, such as complex industries or providing detailed industrial chain details, where they may not be able to directly provide complete answers. Summary of the Invention

[0005] Based on this, it is necessary to provide a large-model industry chain intelligent mining algorithm and system based on an ICT executable environment that can process complex tasks and provide complete answers to the above technical problems.

[0006] In the first aspect, the present application provides a large-model industry chain intelligent mining algorithm and system based on an ICT executable environment, the system comprising:

[0007] The perception module is used to perceive user questions and split the perceived user questions into tasks to obtain target tasks for analyzing the upstream and downstream of each product in the industrial chain. The target tasks are tasks that can be directly processed by the large language model;

[0008] An action module, including various calling tools, including a knowledge acquisition tool required to be called when the large language model performs task processing and a storage tool for storing the industry chain information obtained by the large language model when processing each target task;

[0009] The brain module is used to process each of the target tasks based on the large language model and the calling tools provided by the action module to obtain the industrial chain information corresponding to the industrial chain.

[0010] In one embodiment, the brain module comprises:

[0011] A large language model processing unit, configured to process each target task based on the large language model, and obtain the industrial chain information corresponding to the target task if the industrial chain information corresponding to the target task can be obtained based on the large language model;

[0012] a knowledge supplement unit, configured to obtain knowledge related to the user question based on the knowledge acquisition tool when the industry chain information corresponding to the target task cannot be obtained based on the large language model, and input the user question and the acquired knowledge into the large language model processing unit;

[0013] The large language model processing unit is also used to obtain the industrial chain information corresponding to the target task based on the input user question and the acquired knowledge.

[0014] In one embodiment, the knowledge acquisition tool includes a news search tool and / or a knowledge base calling tool; the knowledge supplement unit is further used to retrieve knowledge related to the user question through the news search tool, and / or retrieve knowledge related to the user question from an externally provided knowledge base through the knowledge base calling tool.

[0015] In one embodiment, the news search tool is used to search for news to be searched, classify the news to be searched based on pre-set data categories, filter the classified news to be searched to obtain target news, embed the target news and user questions into vector format, and determine knowledge related to the user question by calculating the similarity of the embedding-converted vectors, wherein the filtering process includes at least one of filtering based on the quality of the news to be searched, filtering based on the data category of the news to be searched, and filtering based on the necessity of the news to be searched, the quality of the news to be searched is generated by a pre-trained quality assessment model, and the necessity of the news to be searched is generated by a pre-trained necessity assessment model;

[0016] The knowledge base calling tool converts the knowledge in the knowledge base and the user questions into vectors through the FAISS vector database, and determines the knowledge related to the user questions by calculating the similarity of the vectors converted from the FAISS vector database.

[0017] In one embodiment, the brain module is further configured to obtain task attributes of the target task, determine a target large language model based on the task attributes, and process the target task using the target large language model.

[0018] In one embodiment, the brain module is also used to determine Kimi as the target large language model when the task attribute is a text retrieval enrichment task; to determine ChatGPT as the target large language model when the task attribute is a logical reasoning task; and to determine DeepSeek as the target large language model when the task attribute is a processing task with a data volume greater than a data volume threshold.

[0019] In one embodiment, the brain module is also used to decompose the target tasks upstream, midstream and downstream to determine the products corresponding to each stage, and to mine and analyze the composition and importance of the products corresponding to each stage, and to sort out the investment targets based on preset rules and the importance of each product to obtain sorting results, and use the sorting results as industrial chain information.

[0020] In one embodiment, the perception module includes:

[0021] an initial product determining unit, configured to determine initial products based on the perceived user question, wherein each initial product is a product mentioned in the user question, or is obtained by analyzing product composition based on the industry chain name in the user question;

[0022] The upstream and downstream analysis unit is used to perform upstream and downstream relationship analysis based on each of the initial products to obtain each level in the industrial chain, and each of the levels is connected through products, and the upstream and downstream relationship is established through the supply and demand relationship of the products.

[0023] In one embodiment, the upstream and downstream analysis unit is also used to determine the supply and demand relationship of products in each level in the industrial chain based on the direct composition of each initial product, and when the product corresponding to the current level determined in the industrial chain meets the terminal condition, the product meeting the terminal condition is used as the terminal upstream product corresponding to the industrial chain, and the current level is used as the terminal level of the industrial chain; the terminal condition includes that the product corresponding to the current level determined in the industrial chain is a raw material and / or the product corresponding to the current level determined in the industrial chain determined from the entire industrial chain does not need to be further split.

[0024] In one embodiment, the apparatus further comprises:

[0025] The visualization module is used to store the stored industrial chain information in the corresponding industrial chain library, and display the industrial chain information stored in the industrial chain library in the form of a data dashboard.

[0026] The above-mentioned large-scale model industry chain intelligent mining algorithm and system based on the Xinchuang executable environment includes a perception module, an action module, and a brain module. Among them, the perception module is used to perceive user problems and perform task splitting on the perceived user problems to obtain target tasks for analyzing the upstream and downstream of each product in the industry chain, and the target tasks are tasks that can be directly processed by the large language model; the action module includes various calling tools, and the calling tools include the knowledge acquisition tools required to be called when the large language model performs task processing and the storage tools for storing the industry chain information obtained by the large language model processing each target task; the brain module is used to process each target task based on the large language model and the calling tools provided by the action module to obtain the industry chain information corresponding to the industry chain. In this way, splitting the user problems can obtain target tasks for analyzing the upstream and downstream of each product in the industry chain, and then directly processing these tasks through the large language model. When processing, the large language model processes each target task based on the large language model and the calling tools provided by the action module to obtain the industry chain information corresponding to the industry chain, thereby making up for the defects of the large language model and being able to process complex tasks and provide complete answers. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments of the present application or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying any creative work.

[0028] Figure 1 This is an application environment diagram of a large-model industry chain intelligent mining algorithm and system based on a trusted executable environment in one embodiment;

[0029] Figure 2 A schematic diagram of the process of generating a large-scale model industry chain map in one embodiment;

[0030] Figure 3 This is a schematic diagram showing the results of sorting out the large-scale model industry chain in an embodiment;

[0031] Figure 4 A schematic diagram of a data element industry chain map dashboard in one embodiment;

[0032] Figure 5FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0034] In one embodiment, Figure 1 As shown, a large-scale model industry chain intelligent mining algorithm and system based on the Xinchuang executable environment is provided. This embodiment uses the system as an example to illustrate the server. It can be understood that the system can also be a terminal, or a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services. In this embodiment, the system includes: a perception module, an action module, and a brain module.

[0035] The perception module is used to perceive user questions and perform task splitting on the perceived user questions to obtain target tasks for analyzing the upstream and downstream of each product in the industrial chain. The target tasks are tasks that can be directly processed by the large language model.

[0036] The action module includes various calling tools, including a knowledge acquisition tool that needs to be called when the large language model performs task processing and a storage tool that stores the industrial chain information obtained by the large language model when processing each target task;

[0037] The brain module is used to process each of the target tasks based on the large language model and the calling tools provided by the action module to obtain the industrial chain information corresponding to the industrial chain.

[0038] The rotation of current market hotspots is accelerating. The core of thematic investing lies in the rapid analysis and clarification of the relevant industry chains, and the rapid identification of relevant investment targets. Therefore, industry chain research is currently a hot topic. How to quickly understand an unfamiliar industry or investment hotspot, and even identify investment targets within it, is a pressing issue for investors.

[0039] To this end, this application proposes a large-model industry chain intelligent mining algorithm and system based on a trusted executable environment, including a perception module, an action module and a brain module.

[0040] The perception module is used to perceive user questions, wherein the user questions may include industry chain related content and corresponding target questions; wherein the industry chain related content includes the industry chain or the core products involved in the industry chain, for example, the core product involved in the "smartphone" industry chain is the smart phone, the industry chain includes multiple levels, and the user can also directly enter the corresponding industry chain. In this embodiment, the perception module can segment the user questions or process them through a large model to determine the industry chain related content and target questions in the user questions, and based on the industry chain related content and target questions, the user questions are task-splitting to obtain the target task of analyzing the upstream and downstream of each product in the industry chain. Specifically, the perception module can first obtain the corresponding industry chain based on the industry chain related content. The industry chain includes multiple levels, each level corresponds to a product, and different levels are connected by supply and demand relationships. Then, the levels involved are determined based on the target problem, and the target tasks are generated based on the target problem and the levels involved. In this way, the large model can analyze the upstream and downstream structure of the industry chain in detail, covering upstream raw material supply, midstream production and manufacturing, and downstream distribution channels, and reveal the connection and dependency between each link. Optionally, the objective task can include the upstream and downstream structure of the industrial chain, covering upstream raw material supply, midstream production and manufacturing, and downstream distribution channels, and revealing the connections and dependencies between each link. This objective task can also include the sorting of investment targets. Specifically, users will receive a detailed analysis of each component of the industrial chain, including their role in the overall industrial chain, their importance, and their influence on the entire industry. In addition, by analyzing market data and trends, the big model can also provide users with potential investment opportunities and identify leading companies and emerging companies in the industry. Through this highly integrated big model industrial chain mining tool, users can not only obtain rich information and data, but also enjoy personalized consulting services, so as to seize opportunities in a complex market environment and achieve precise investment.

[0041] The action module includes various calling tools. The main purpose of the calling tools is to supplement the brain module and to parse the target tasks that the large language model cannot handle and convert them into target tasks that the large language model can handle.

[0042] The calling tool includes a knowledge acquisition tool and a storage tool. The knowledge acquisition tool is used to retrieve knowledge based on target tasks that cannot be processed by the large language model, thereby supplementing the knowledge base of the large language model and solving the target tasks based on the retrieved knowledge. The storage tool is used to store the industrial chain information obtained by the large language model when processing each target task. In some optional embodiments, the storage tool also stores conversation records, which include corresponding user questions.

[0043] The brain module is the key module of the entire system. The brain module includes the industry chain knowledge base, the model base, the large language model and the sorting results of the investment targets. Among them, the industry chain knowledge base can be a knowledge base for training the large language model, and the knowledge base is also updated in real time through the knowledge acquisition tool of the action module. The knowledge base is also used by the perception module to build the industry chain, such as determining the upstream and downstream levels, the products corresponding to each level, the supply and demand relationship between each product, etc. The model base is used to store the basis of various large language models, including the DeepSeek model, etc. The large language model is used to process the target task to obtain the corresponding industry chain information, and the investment target sorting is used to process the industry chain information generated by the large language model to obtain the final sorting results for the convenience of user viewing.

[0044] Among them, the trusted executable environment refers to deploying the industry chain agent on the target's GPU resources, such as the GPU resources of Huawei 910A, to achieve the entire process of industry chain generation projects, including large language model inference deployment and knowledge base question and answer deployment.

[0045] The above-mentioned large-scale model industry chain intelligent mining algorithm and system based on the Xinchuang executable environment includes a perception module, an action module, and a brain module. Among them, the perception module is used to perceive user problems and perform task splitting on the perceived user problems to obtain target tasks for analyzing the upstream and downstream of each product in the industry chain, and the target tasks are tasks that can be directly processed by the large language model; the action module includes various calling tools, and the calling tools include the knowledge acquisition tools required to be called when the large language model performs task processing and the storage tools for storing the industry chain information obtained by the large language model processing each target task; the brain module is used to process each target task based on the large language model and the calling tools provided by the action module to obtain the industry chain information corresponding to the industry chain. In this way, splitting the user problems can obtain target tasks for analyzing the upstream and downstream of each product in the industry chain, and then directly processing these tasks through the large language model. When processing, the large language model processes each target task based on the large language model and the calling tools provided by the action module to obtain the industry chain information corresponding to the industry chain, thereby making up for the defects of the large language model and being able to process complex tasks and provide complete answers.

[0046] In one optional embodiment, the brain module includes: a large language model processing unit and a knowledge supplement unit, wherein the large language model processing unit is the large language model mentioned above, and the knowledge supplement unit is the industrial chain knowledge base mentioned above, wherein:

[0047] The large language model processing unit is used to process each of the target tasks based on the large language model, and obtain the industrial chain information corresponding to the target task when the industrial chain information corresponding to the target task can be obtained based on the large language model.

[0048] The knowledge supplement unit is used to obtain knowledge related to the user question based on the knowledge acquisition tool when the industrial chain information corresponding to the target task cannot be obtained based on the large language model, and input the user question and the acquired knowledge into the large language model processing unit.

[0049] The large language model processing unit is also used to obtain the industrial chain information corresponding to the target task based on the input user question and the acquired knowledge.

[0050] The large language model processing unit can process each target task based on the large language model, wherein the large language model processing unit can include decomposing the target tasks into upstream, midstream and downstream to determine the products corresponding to each stage, and mining and analyzing the composition and importance of the products corresponding to each stage, and sorting out the investment targets based on preset rules and the importance of each product to obtain sorting results, and use the sorting results as industrial chain information.

[0051] Among them, the industry chain agent brain module mainly adopts the current mainstream large language models, including DeepSeek, Qwen, etc.

[0052] The large language model processing unit obtains the industrial chain information corresponding to the target task when it is able to obtain the industrial chain information corresponding to the target task. Since the training text of the large language model contains sufficient industrial chain knowledge, it is capable of answering professional questions related to the industrial chain. At the same time, the large language model has powerful natural language processing and generation capabilities, so it can be used as a core component responsible for processing input information, making decisions and generating outputs.

[0053] Optionally, when the industrial chain information corresponding to the target task cannot be obtained based on the large language model, in order to realize the processing of the target task, the knowledge related to the user question is obtained through a knowledge acquisition tool, and the user question and the acquired knowledge are input into the large language model processing unit.

[0054] The acquisition of knowledge can be achieved by first semantically splitting the user question to obtain multiple objects that can represent specific semantics, and then obtaining the associated semantic objects of the objects. In this way, the knowledge acquisition tool performs knowledge retrieval based on the objects and associated semantic objects to obtain knowledge related to the user question. Finally, the user question and the acquired knowledge are input into the large language model processing unit, thereby supplementing the knowledge base of the large language model. The large language model can then process the corresponding tasks to obtain the final industrial chain information.

[0055] In some optional embodiments, the position of the industrial chain corresponding to the user question can be determined, and then the obtained industrial chain information is stored in the industrial chain based on the position of the industrial chain, thereby facilitating subsequent processing.

[0056] In one of the optional embodiments, the knowledge acquisition tool includes a news search tool and / or a knowledge base calling tool; the knowledge supplement unit is further used to retrieve knowledge related to the user question through the news search tool, and / or retrieve knowledge related to the user question from an externally provided knowledge base through the knowledge base calling tool.

[0057] The news search tool is used to search news to obtain knowledge related to the user's question, and the knowledge base calling tool allows the user to call the corresponding knowledge base to obtain at least one question related to the user's question.

[0058] In one optional embodiment, the news search tool is used to search for news to be searched, classify the news to be searched based on pre-set data categories, filter the classified news to be searched to obtain target news, embed the target news and user questions into vector format, and determine knowledge related to the user question by calculating the similarity of the embedding-converted vectors, wherein the filtering process includes at least one of filtering based on the quality of the news to be searched, filtering based on the data category of the news to be searched, and filtering based on the necessity of the news to be searched, the quality of the news to be searched is generated by a pre-trained quality assessment model, and the necessity of the news to be searched is generated by a pre-trained necessity assessment model;

[0059] The knowledge base calling tool converts the knowledge in the knowledge base and the user questions into vectors through the FAISS vector database, and determines the knowledge related to the user questions by calculating the similarity of the vectors converted from the FAISS vector database.

[0060] Among them, the news search tool can search for news to be searched from various sources and classify the news to be searched based on the source of the news to be searched. For specific data categories, please refer to Figure 2As shown, the classified news to be searched is then filtered to obtain target news, wherein the filtering process can include at least one of filtering based on the quality of the news to be searched, filtering based on the data category of the news to be searched, and filtering based on the necessity of the news to be searched. The quality of the news to be searched is generated by a pre-trained quality assessment model, and the necessity of the news to be searched is generated by a pre-trained necessity assessment model. The quality of the news to be searched can be obtained based on the source of the news to be searched, the time of the news to be searched, etc. The filtering of the data category of the news to be searched is to ensure that at least one news to be searched is included in each data category, or the data category of the news to be searched includes at least a target number of categories, wherein the target number can be set based on needs, and the setting of the target number can be performed based on the data category. For example, for a key data category, it can be one, that is, as long as the target news of one key data category is included, it is sufficient. For non-key data categories, the more the better, that is, for non-key data categories, diversity needs to be ensured to ensure the accuracy of subsequent industry chain information. The necessity of the news to be searched is generated by a pre-trained necessity assessment model, such as the data categories that must be included, such as at least one key data category.

[0061] After obtaining the target news, the similarity between the target news and the user question can be calculated. This similarity can be calculated using an embedding-based algorithm, for example. First, the target news and user question need to be embedded and converted into vector format to quickly calculate text similarity. During this process, the "text2vec-large-chinese" model is used for embedding conversion. Other agent projects, such as Langchain, use the FAISS vector database to implement the retrieval step in RAG. This database is an efficient similarity search and clustering library that can quickly process large-scale data and perform similarity searches in high-dimensional space. Similarity searches calculate the "text distance" between two vectors. The smaller the distance, the higher the similarity, thus supplementing the model with professional knowledge. For example, the knowledge base call tool converts the knowledge in the knowledge base and user questions into vectors using the FAISS vector database and determines the knowledge related to the user question by calculating the similarity of the vectors converted from the FAISS vector database.

[0062] Among them combined Figure 2 As shown, Figure 2This is the process of generating a large-scale industrial chain map in an embodiment, in which the target is to integrate the target's multi-dimensional data, realize the target chain analysis, enhance the dimension of related understanding, and assist in investment research and decision-making. The industrial chain includes industrial chain mining and real-time update algorithms to analyze the degree of correlation between the industrial chain and the target. The data categories include basic information, market information, business information, intellectual property, operating information, and public opinion data. Basic information includes registration information and company profiles, market information includes main business, industry research reports, and company research reports, business information includes business scope, WeChat official account, etc., intellectual property includes patents and software copyrights, etc., and operating information includes special licenses, scientific and technological achievements, standard setting, bidding, recruitment information, etc. Public opinion data includes corporate public opinion, industry public opinion, financial information and capital markets, etc. The knowledge acquisition tool of the action module can acquire the corresponding knowledge, classify the acquired knowledge based on these categories, and then filter the acquired knowledge. The filtering can be performed through a pre-trained data filtering model, and then the filtered data is subjected to industrial information extraction. For example, the industrial information extraction model performs fine-grained configuration of data sources, information extraction and industrial mapping, and finally obtains an industrial chain map generation model. The industrial chain map generation model knows the preset information of the industrial chain, and executes a large-model industrial chain mining algorithm, relevant target extraction and correlation analysis, and industrial chain construction and real-time update.

[0063] The Big Model Industry Chain Mining tool provides users with an advanced language-based interactive interface. Users can easily interact with the Big Model to delve deeper into key targets within the industry chain. This interactive experience provides users with a range of in-depth analyses and information. First, the Big Model provides a detailed analysis of the upstream and downstream structure of the industry chain, encompassing upstream raw material supply, midstream manufacturing, and downstream distribution channels, revealing the connections and dependencies between each link. Second, users receive a comprehensive analysis of each component of the industry chain, including their role, importance, and impact on the industry as a whole. Furthermore, by analyzing market data and trends, the Big Model provides users with potential investment opportunities and identifies leading companies and emerging companies within the industry. This highly integrated Big Model industry chain mining tool not only provides users with a wealth of information and data, but also offers personalized consulting services, enabling them to seize opportunities and achieve targeted investment in complex market environments.

[0064] In this embodiment, the model may have less knowledge about products with less relevant information and may not be able to provide answers based on its own knowledge. A solution based on retrieval-enhanced generation (RGA) is proposed, which configures a news retrieval tool for the industry chain agent. Retrieval-enhanced generation includes two steps: retrieval and generation. The retrieval step is to search for the knowledge that is most relevant to the question to be answered in an external knowledge base or database based on a natural language model, such as documents; the generation step is to combine the user question with the retrieved information and put it into a large language model to generate a response.

[0065] To further enhance the large model's knowledge base, the system utilizes news retrieval tools and integrates information from multiple sources, including news, research reports, and company announcements. This not only effectively expands the large model's knowledge base but also improves the quality of the final retrieved text through a hybrid retrieval approach. The invocation of external tools allows the industry chain agent to obtain a more detailed industry chain structure and access a wider range of news data.

[0066] In addition, the retrieval enhancement generation step is a key link in the entire process. First, the news data needs to be embedded and converted into a vector format in order to quickly calculate the text similarity. In this process, the "text2vec-large-chinese" model is selected for the embedding conversion. Other agent projects such as Langchain use the FAISS vector database to implement the retrieval step in RAG. This is an efficient similarity search and clustering library that can quickly process large-scale data and perform similarity searches in high-dimensional space. Similarity search calculates the "text distance" between two vectors. The smaller the distance, the higher the similarity, thus providing professional knowledge to supplement the model.

[0067] In one of the optional embodiments, the brain module is further used to obtain task attributes of the target task, and determine a target large language model based on the task attributes, and process the target task through the target large language model.

[0068] In one of the optional embodiments, the brain module is also used to determine Kimi as the target large language model when the task attribute is a text retrieval enrichment task; to determine ChatGPT as the target large language model when the task attribute is a logical reasoning task; and to determine DeepSeek as the target large language model when the task attribute is a processing task with a data volume greater than a data volume threshold.

[0069] The core brain module is provided by currently available large-scale models, each with its own unique characteristics and advantages. For example, the Kimi model excels at processing long Chinese texts and is therefore recommended for retrieval and text enrichment tasks. The ChatGPT series excels in logical reasoning and is suitable for industrial chain analysis tasks. For lower-level, high-volume text processing tasks, the DeepSeek model is preferred to ensure cost control and high efficiency. Selecting the most appropriate large-scale model based on the specifics of each task maximizes the unique advantages of each model while balancing accuracy and cost control.

[0070] In one of the optional embodiments, the perception module includes: an initial product determination unit, which is used to determine each initial product based on the perceived user question, and each initial product is a product carried in the user question, or the initial product is obtained by performing a product composition analysis based on the name of the industrial chain in the user question; an upstream and downstream analysis unit, which is used to perform an upstream and downstream relationship analysis based on each initial product to obtain each level in the industrial chain, and each level is connected through a product, and the upstream and downstream relationship is established through the supply and demand relationship of the product.

[0071] In one of the optional embodiments, the upstream and downstream analysis unit is also used to determine the supply and demand relationship of products in each level in the industrial chain based on the direct composition of each of the initial products, and when the product corresponding to the current level determined in the industrial chain meets the terminal condition, the product meeting the terminal condition is used as the terminal upstream product corresponding to the industrial chain, and the current level is used as the terminal level of the industrial chain; the terminal condition includes that the product corresponding to the current level determined in the industrial chain is a raw material and / or the product corresponding to the current level determined in the industrial chain determined from the entire industrial chain does not need to be further split.

[0072] The perception module is used to determine target tasks. The perception module of the industry chain agent detects user questions and breaks down tasks into sub-tasks through prompt engineering. First, it's clear that each level of the industry chain is connected by products, and upstream and downstream relationships are established through product supply and demand. The entire industry chain can be broken down into upstream and downstream analyses of each product, enabling refined management.

[0073] The perception module includes an initial product determination unit. Because user questions can include industry chains or core products, product composition analysis can be performed based on the industry chain name or the initial product can be directly derived based on the products in the user's question. Product composition analysis can be based on direct components, such as the smartphone industry chain, where "direct components" include displays, lithium batteries, processors, and storage.

[0074] The upstream and downstream analysis unit is used to perform upstream and downstream relationship analysis based on each of the initial products to obtain each level in the industrial chain, and each of the levels is connected through products, and the upstream and downstream relationship is established through the supply and demand relationship of the products. The industrial chain corresponds to multiple levels, each level can correspond to at least one product, and the products between different levels have a supply and demand relationship. In this application, the upstream and downstream products of the initial product can be mined, and the final key level and the terminal upstream product can be determined. For example, when the product corresponding to the current level determined in the industrial chain meets the terminal conditions, the product that meets the terminal conditions is used as the terminal upstream product corresponding to the industrial chain, and the current level is used as the terminal level of the industrial chain; the terminal conditions include that the product corresponding to the current level determined in the industrial chain is raw material and / or the product corresponding to the current level determined in the industrial chain determined from the entire industrial chain does not need to be further split.

[0075] To facilitate understanding, combine Figure 3 As shown, taking the smartphone industry chain as an example, its "direct components" include displays, lithium batteries, processors, and storage, which are also the "upstream" products of the phone. Analyzing the display's direct components reveals upstream components such as the LCD panel, touch module, and diode panel. Following this upward trajectory, we can eventually identify a subset of products that meet the "end condition": the product itself is a raw material, or there's no need to further separate products from the overall industry chain. For example, for lithium batteries, the traceability ultimately goes back to cathode raw materials like nickel and cobalt, but for displays, the traceability to liquid crystal materials, glass substrates, and other components is already sufficiently detailed.

[0076] In one of the optional embodiments, the device further includes: a visualization module, configured to store each of the stored industrial chain information in a corresponding industrial chain library, and to display the industrial chain information stored in the industrial chain library in the form of a data dashboard.

[0077] The investment target visualization module generates a large-scale industrial chain target. After the target is generated, mature and high-quality industrial chain information will be aggregated and added to the industrial chain library. Through the data dashboard, users can obtain rich industrial chain statistical data.

[0078] The visualization module is used to aggregate and add mature, high-quality industry chain information to the industry chain library after the large-scale model industry chain targets are generated. The industry chain library is a dynamically updated resource that contains rigorously screened and verified industry chain data, providing users with a comprehensive and in-depth industry perspective. Through the data dashboard, users can access rich industry chain statistics, intuitively perceive the composition of industry chain targets, and understand the value creation and potential risks of each link. Currently, high-quality data element industry chain, wine industry chain, and synthetic materials industry chain maps have been constructed in the industry chain map library and presented in the form of data dashboards.

[0079] Among them, combined Figure 4 As shown, Figure 4 The data element industry chain map dashboard is shown. After the large-scale industry chain target is generated, mature, high-quality industry chain information will be aggregated and added to the industry chain library. The industry chain library is a dynamically updated resource that contains rigorously screened and verified industry chain data, providing users with a comprehensive and in-depth industry perspective. Through the data dashboard, users can access rich industry chain statistics, intuitively perceive the composition of industry chain targets, and understand the value creation and potential risks of each link.

[0080] For ease of understanding, the present application is based on a large-scale model industry chain intelligent mining algorithm and system in a trusted and innovative executable environment. The system mainly includes four core modules: a brain module, a perception module, an action module, and an investment target visualization module. As the core of the entire Agent, the brain module uses a general large model to handle complex analysis tasks. The perception module is responsible for receiving and parsing questions raised by users, and decomposing them into a series of specific subtasks through the prompt project to form a complete set of thinking links. The action module focuses on calling a variety of tools to expand the professional knowledge base of the large model, including news search tools, knowledge base access tools, industry chain database writing tools, etc., to significantly expand the model's professional knowledge, thereby improving the overall performance of the system. After the large model generates the industry chain target, the investment target visualization module aggregates high-quality industry chain information into the industry chain library, and displays rich statistical data through an intuitive data dashboard. The entire system is deployed in a trusted and innovative environment to ensure efficiency and security.

[0081] The industry chain agent perception module is primarily used to enable industry chain agents to perceive user questions and, through prompt engineering, break down tasks into specific subtasks, forming a chain of thought. Furthermore, the entire industry chain is broken down into upstream and downstream analyses of each product, enabling detailed industry chain analysis.

[0082] The brain module of the industrial chain agent relies on the current mainstream large language model, and flexibly selects the most suitable large language model according to the needs of specific tasks to ensure efficient and accurate task processing.

[0083] The industrial chain agent action module, that is, the agent's ability to call external tools, solves problems such as lack of relevant knowledge of large language models and poor real-time performance by calling news retrieval, knowledge base retrieval, industrial chain database writing, and other tools.

[0084] The investment target visualization module enables users to obtain rich industry chain statistical data in the form of a data dashboard, intuitively perceive the composition of industry chain targets, and understand the value creation and potential risks of each link.

[0085] The system is designed to help investors quickly understand and analyze unfamiliar industries or investment hotspots, and accurately identify relevant investment targets. The system utilizes a general large language model as its brain module for information processing, task execution, and decision-making. The perception module captures user questions and employs prompt engineering to break them down into more detailed tasks, forming a thought chain to guide subsequent operations. The action module integrates multiple tools, such as news search and knowledge base Q&A, to expand professional knowledge and ensure real-time data. It also supports writing to the industry chain database and preserving conversation records. The investment target visualization module displays generated industry chain information in the form of a data dashboard, enabling users to intuitively understand the industry chain structure, value creation, and potential risks at each link, thereby assisting investment decisions. The entire system is deployed in a trusted innovation environment, ensuring security and efficiency. Through efficient task decomposition and resource integration, the system has significantly improved the efficiency and quality of industry chain research, meeting the market's demand for fast and accurate analysis.

[0086] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0087] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Figure 5As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. Among them, the processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store information related to the large-model industrial chain intelligent mining algorithm and system during the processing based on the credible executable environment. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements a method involved in a large-model industrial chain intelligent mining algorithm and system based on the credible executable environment.

[0088] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0089] In one embodiment, a computer device is also provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps in the method involved in the large-model industrial chain intelligent mining algorithm and system based on the ICT executable environment are implemented.

[0090] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the method involved in the above-mentioned large-model industrial chain intelligent mining algorithm and system based on the ICT executable environment are implemented.

[0091] In one embodiment, a computer program product is provided, including a computer program, which, when executed by a processor, implements the steps in the method involved in the above-mentioned large-model industrial chain intelligent mining algorithm and system based on the ICT executable environment.

[0092] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0093] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile memory and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processor involved in the various embodiments provided herein may be, but are not limited to, a general-purpose processor, a central processing unit (CPU), a graphics processing unit (GPU), a digital signal processor (DSP), a programmable logic unit (PLC), a data processing logic unit based on quantum computing, an artificial intelligence (AI) processor, and the like.

[0094] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0095] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A large-scale model industry chain intelligent mining algorithm and system based on the Xinchuang executable environment, characterized by: The system comprises: The perception module is used to perceive user questions and split the perceived user questions into tasks to obtain target tasks for analyzing the upstream and downstream of each product in the industrial chain. The target tasks are tasks that can be directly processed by the large language model; An action module, including various calling tools, including a knowledge acquisition tool required to be called when the large language model performs task processing and a storage tool for storing the industry chain information obtained by the large language model when processing each target task; The brain module is used to process each of the target tasks based on the large language model and the calling tools provided by the action module to obtain the industrial chain information corresponding to the industrial chain.

2. The system according to claim 1, wherein: The brain module includes: A large language model processing unit, configured to process each target task based on the large language model, and obtain the industrial chain information corresponding to the target task if the industrial chain information corresponding to the target task can be obtained based on the large language model; a knowledge supplement unit, configured to obtain knowledge related to the user question based on the knowledge acquisition tool when the industry chain information corresponding to the target task cannot be obtained based on the large language model, and input the user question and the acquired knowledge into the large language model processing unit; The large language model processing unit is also used to obtain the industrial chain information corresponding to the target task based on the input user question and the acquired knowledge.

3. The system according to claim 2, characterized in that The knowledge acquisition tool includes a news search tool and / or a knowledge base calling tool; the knowledge supplement unit is also used to retrieve knowledge related to the user question through the news search tool, and / or retrieve knowledge related to the user question from an externally provided knowledge base through the knowledge base calling tool.

4. The system according to claim 3, characterized in that The news search tool is used to search for news to be searched, classify the news to be searched based on pre-set data categories, filter the classified news to be searched to obtain target news, embed the target news and user questions into vector format, and determine knowledge related to the user question by calculating the similarity of the embedding-converted vectors, wherein the filtering process includes at least one of filtering based on the quality of the news to be searched, filtering based on the data category of the news to be searched, and filtering based on the necessity of the news to be searched, wherein the quality of the news to be searched is generated by a pre-trained quality assessment model, and the necessity of the news to be searched is generated by a pre-trained necessity assessment model; The knowledge base calling tool converts the knowledge in the knowledge base and the user questions into vectors through the FAISS vector database, and determines the knowledge related to the user questions by calculating the similarity of the vectors converted from the FAISS vector database.

5. The system according to claim 1, wherein: The brain module is further configured to obtain task attributes of the target task, determine a target large language model based on the task attributes, and process the target task using the target large language model.

6. The system according to claim 5, characterized in that The brain module is also used to determine Kimi as the target large language model when the task attribute is a text retrieval enrichment task; to determine ChatGPT as the target large language model when the task attribute is a logical reasoning task; and to determine DeepSeek as the target large language model when the task attribute is a processing task with a data volume greater than a data volume threshold.

7. The system according to claim 1, wherein: The brain module is also used to decompose the target tasks upstream, midstream and downstream to determine the products corresponding to each stage, and to explore and analyze the composition and importance of the products corresponding to each stage, and to sort out the investment targets based on preset rules and the importance of each product to obtain sorting results, and use the sorting results as industrial chain information.

8. The system according to claim 1, wherein: The perception module includes: an initial product determining unit, configured to determine initial products based on the perceived user question, wherein each initial product is a product mentioned in the user question, or is obtained by analyzing product composition based on the industry chain name in the user question; The upstream and downstream analysis unit is used to perform upstream and downstream relationship analysis based on each of the initial products to obtain each level in the industrial chain, and each of the levels is connected through products, and the upstream and downstream relationship is established through the supply and demand relationship of the products.

9. The system according to claim 8, characterized in that The upstream and downstream analysis unit is also used to determine the supply and demand relationship of products in each level in the industrial chain based on the direct composition of each initial product, and when the product corresponding to the current level determined in the industrial chain meets the terminal condition, the product that meets the terminal condition is used as the terminal upstream product corresponding to the industrial chain, and the current level is used as the terminal level of the industrial chain; the terminal condition includes that the product corresponding to the current level determined in the industrial chain is a raw material and / or the product corresponding to the current level determined in the industrial chain determined from the entire industrial chain does not need to be further split.

10. The system according to claim 1, wherein: The device further comprises: The visualization module is used to store the stored industrial chain information in the corresponding industrial chain library, and display the industrial chain information stored in the industrial chain library in the form of a data dashboard.

Images