Steel e-commerce platform shopping guide Agent interaction method and system based on large model

Abstract

The invention relates to a steel e-commerce platform shopping guide Agent interaction method and system based on a large model, and the method comprises the following steps: obtaining user information, product information and market information of a steel e-commerce platform, carrying out the data cleaning, structuring and enhancement processing, and obtaining the preprocessed data; based on the preprocessed data, multi-dimensional features of the user and attributes and relation information of the steel products are extracted, and a knowledge graph is constructed; constructing prompt of the large model, and realizing fine tuning of the large model in multiple scenes; and based on the knowledge graph and the fine-tuned large model, interaction between the user and the steel e-commerce platform shopping guide Agent is realized. Compared with the prior art, the shopping guide service can be provided for customers in a targeted mode, and the method has the advantages of being high in adaptation degree to the steel industry, accurately guiding user requirements and the like.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a shopping guide Agent interaction method and system for a steel e-commerce platform based on a large model. Background Art

[0002] At present, steel e-commerce platforms have become an important channel for steel product transactions. They aggregate a large number of suppliers and buyers through online platforms and provide convenient online transaction services. However, the existing steel e-commerce platform shopping guides still have the following problems:

[0003] (1) Inadequate user experience: The amount of product information on e-commerce platforms is huge, and users often feel confused when filtering and comparing products. The existing search and filtering mechanism requires multiple searches and filters to find the required products, making it difficult to quickly find products that suit them and inefficient.

[0004] (2) Lack of personalized shopping guides: Products and services are similar and lack differentiated competitive advantages, resulting in low user loyalty. The existing shopping guide system is difficult to handle complex steel product consultations and cannot deeply understand users' personalized purchasing needs and provide in-depth technical support.

[0005] (3) Low operational efficiency: When handling a large number of inquiries (such as during promotional periods), human customer service staff are overwhelmed and costly, making it difficult to meet users’ demands for immediate responses.

[0006] To sum up, the steel industry has not yet seen any intelligent application that can fully integrate user needs, product information and corporate services to provide personalized and efficient shopping guide services. It is necessary to propose a new model of smarter shopping guide method for steel e-commerce platforms. Summary of the invention

[0007] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a large-model-based steel e-commerce platform shopping guide Agent interaction method and system to solve or partially solve the problem that in the steel e-commerce platform scenario, it is impossible to provide targeted shopping guide services to customers.

[0008] The purpose of the present invention can be achieved by the following technical solutions:

[0009] One aspect of the present invention provides a steel e-commerce platform shopping guide Agent interaction method based on a large model, comprising the following steps:

[0010] Obtain user information, product information and market information from the steel e-commerce platform, perform data cleaning, structuring and enhancement processing, and obtain pre-processed data;

[0011] Based on the preprocessed data, a knowledge graph is constructed by extracting the multi-dimensional features of users and the attributes and relationship information of steel products;

[0012] Build a prompt for a large model and fine-tune the large model in multiple scenarios;

[0013] Based on the knowledge graph and the fine-tuned large model, the interaction between users and the shopping guide agent of the steel e-commerce platform is realized.

[0014] As a preferred technical solution, the process of data cleaning, structuring and enhancing the user information, product information and market information obtained from the steel e-commerce platform includes the following steps:

[0015] Through rule-based outlier detection, we can identify and delete erroneous values, remove noise, and unify naming and description;

[0016] Through structured processing, the data is converted into a form that matches the large model, and data enhancement is achieved by filling in missing data.

[0017] As a preferred technical solution, the process of constructing a knowledge graph includes the following steps:

[0018] Identify entities in preprocessed data based on BERT-CRF and generate relationship descriptions between entities through relationship extraction;

[0019] Build a knowledge graph based on the identified entities and extracted relationship descriptions.

[0020] As a preferred technical solution, the prompt of the large model includes one or more of the tool, tool parameters, format requirements for tool calling, large model role positioning, and Agent capability range.

[0021] As a preferred technical solution, fine-tuning of the large model is achieved in multiple scenarios, wherein the multiple scenarios include recommending suitable steel products according to user needs, answering user questions about steel specifications, and querying one or more of steel application scenarios.

[0022] As a preferred technical solution, the process of realizing the interaction between the user and the shopping guide Agent of the steel e-commerce platform based on the knowledge graph and the fine-tuned large model includes the following steps:

[0023] Obtain the user's question information, configure the big model role as a professional engineer who rewrites the user's question, perform big model intent recognition based on the user's conversation context, rewrite the question information based on the user's intent, and determine whether the Agent needs to be called after correcting spelling errors and supplementing missing information;

[0024] In response to calling the Agent, the prompt parameter is extracted based on the question information, and the recommendation result is obtained based on the knowledge graph and the fine-tuned large model, and the shopping guide suggestion information is obtained by combining it with the user dialogue context;

[0025] In response to not calling the Agent or obtaining the shopping guide suggestion information, the large model role is updated to be a professional shopping guide that further actively guides based on the recommendation results. Based on the shopping guide suggestion information, guiding questions are generated for products or topics of interest to the user to obtain formatted guiding words.

[0026] As a preferred technical solution, the shopping guide suggestion information includes a recommended product list and / or price information.

[0027] As a preferred technical solution, the user information includes one or more of the user's basic industrial and commercial information, transaction information, behavioral data, preference characteristics, and corporate qualifications; the product information includes one or more of the basic properties of steel, warranty, inventory information, and price system; the market information includes one or more of supply and demand relationships, price indexes, industry trends, and competitive product analysis.

[0028] Another aspect of the present invention provides a steel e-commerce platform shopping guide Agent interactive system based on a large model, comprising:

[0029] The data collection and preprocessing module is used to obtain user information, product information and market information of the steel e-commerce platform, and perform data cleaning, structuring and enhancement processing to obtain preprocessed data;

[0030] A user-product knowledge graph construction module, which is used to construct a knowledge graph based on the preprocessed data by extracting the multi-dimensional features of users and the attributes and relationship information of steel products;

[0031] Large model LoRA fine-tuning training module, used to build the prompt of the large model and implement fine-tuning of the large model in multiple scenarios;

[0032] The shopping guide agent configuration and scheduling module is used to realize the interaction between the user and the shopping guide agent of the steel e-commerce platform based on the knowledge graph and the fine-tuned large model.

[0033] As a preferred technical solution, the shopping guide Agent configuration and scheduling module includes:

[0034] The user question rewriting unit is used to obtain the user's question information, configure the big model role as a professional engineer who rewrites the user's question, perform big model intent recognition based on the user conversation context, rewrite the question information based on the user's intent, and determine whether the Agent needs to be called after correcting spelling errors and supplementing missing information;

[0035] A shopping guide agent scheduling unit, which is used to extract prompt parameters based on question information in response to calling the agent, obtain recommendation results based on the knowledge graph and the fine-tuned large model, and obtain shopping guide suggestion information by combining the results with the user dialogue context;

[0036] The user demand refinement guidance unit is used to update the large model role to a professional shopping guide that further actively guides based on the recommendation results in response to not calling the Agent or obtaining the shopping guide suggestion information, and generate guiding questions for the products or topics that the user is interested in based on the shopping guide suggestion information to obtain formatted guiding words.

[0037] Compared with the prior art, the present invention has at least one of the following beneficial effects:

[0038] (1) Providing targeted shopping guide services to customers: The present invention constructs a multi-dimensional knowledge graph that integrates user characteristics and product attributes, breaking through the limitation of the separation of user and product information in traditional systems, and providing more comprehensive knowledge support for personalized recommendations. Through personalized intelligent shopping guides, the present invention greatly improves the decision-making efficiency of users. Users can find the required steel products more quickly and accurately, thereby shortening the procurement cycle and reducing the time and opportunity costs of enterprises.

[0039] (2) High adaptability to the steel industry: The present invention achieves optimized adaptation of the large model in the steel field through LoRA fine-tuning, while maintaining general capabilities, it improves the ability to understand steel industry knowledge.

[0040] (3) Realizing question rewriting: The present invention introduces a large model-based question rewriting, which can more accurately understand the user's intention and supplement the missing information based on the context, thereby improving the accuracy and completeness of the query.

[0041] (4) Accurately guiding user needs: The present invention realizes refined guidance of user needs by utilizing a large model, and can actively guide users to express more specific needs. This not only improves the user experience, but also helps the system collect more valuable user preference information, providing data support for subsequent personalized services and product recommendation strategy optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a schematic diagram of the steel e-commerce platform shopping guide Agent interaction method based on the large model in the embodiment;

[0043] Figure 2 It is a schematic diagram of the steel e-commerce platform shopping guide Agent interaction system based on the large model in the embodiment;

[0044] Figure 3It is a schematic diagram of the workflow of the shopping guide agent configuration and scheduling module in the embodiment. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0046] Example 1

[0047] In view of the problems existing in the above-mentioned prior art, this embodiment provides a shopping guide agent interaction method for a steel e-commerce platform based on a large model, see Figure 1 The method includes steps S1-S4, each of which is described in detail as follows:

[0048] Step S1, data collection and preprocessing: Based on the characteristics of the steel industry, collect and organize multi-source heterogeneous data such as user information, product information, market data, etc., and perform cleaning, standardization and structured processing to provide a high-quality data foundation for building a user-product knowledge graph.

[0049] Specifically, step S1 may include steps S101-S103:

[0050] Step S101, multi-source data collection: design an API interface to query and obtain historical user information (including basic industrial and commercial information, transaction information, behavioral data, preference characteristics, enterprise qualifications, etc.), product information (including basic properties of steel, warranty, inventory information, price system, etc.) and market data (including supply and demand relationships, price index, industry trends, competitive product analysis, etc.) from the steel e-commerce platform database.

[0051] Step S102, data cleaning and standardization: through the rule-based outlier detection algorithm, identify and process the noise and errors in the data, and unify the naming descriptions from different sources.

[0052] Step S103, structured processing and enhancement: extract structured information and generate question-answer pairs based on the big model to provide corpus that meets the requirements for knowledge graph construction and big model LoRA fine-tuning training. At the same time, data completion is performed to partially fill in missing product attributes and market information to ensure the integrity and accuracy of key data.

[0053] Step S2, user-product knowledge graph construction: Based on the preprocessed data, the large model is used to extract the multi-dimensional characteristics of users and the core attributes and relationships of steel products, and a multi-dimensional knowledge graph is constructed to provide query support for subsequent shopping guide agents.

[0054] Specifically, step S2 may include steps S201-S203:

[0055] Step S201, ontology design: Based on the expert knowledge and data analysis in the steel field, a multi-level ontology model for the steel industry is designed. A high-quality ontology model is the key to building a high-quality knowledge graph. The large model is used to assist in the generation of ontology concepts and relationships, which are reviewed and optimized by experts to standardize the terms and concepts in the industry, reduce misunderstandings and communication barriers caused by inconsistent terminology, and improve the efficiency and accuracy of information exchange in the industry. At the same time, it provides important guidance information for the subsequent entity recognition and relationship extraction process.

[0056] Step S202, entity recognition and relationship extraction: further identify entities such as user names and product names in the collected data based on the entity recognition algorithm BERT-CRF. Use the large model to extract relationships and generate relationship descriptions between entities.

[0057] Step S203, knowledge graph interaction and visualization: Use common knowledge graph construction tools (such as Neo4j, etc.) to build a knowledge graph, design a multi-dimensional knowledge exploration interface, support multi-angle selection of sub-graphs by user attributes, product attributes, etc., and provide services for subsequent shopping guide agents.

[0058] Step S3, fine-tuning training of the large model LoRA: Use professional corpus from the steel industry to fine-tune the pre-trained large model for domain adaptability, and improve the model's understanding and generation capabilities in the steel field as well as its shopping guide agent scheduling capabilities.

[0059] Specifically, step S3 may include steps S301-S304.

[0060] Step S301, first use the big model to build a unified beginning of the dialogue records in all fine-tuning corpora based on the beginning of the sample corpus (i.e., prompt, including available tools and their parameters, format requirements for tool calls, big model role positioning and specific functions, and Agent's capabilities). The beginning format of the shopping guide Agent sample corpus is as follows:

[0061] "\n#Tools\n\n##You have the following tools:\n\nsteel_product_recommend: Steel product recommendation API. Input parameters:{\"type\":\"object\",\"properties\":{\"product_type\":{\"type\":\"string\",\"description\":\"Steel type, such as H-beam, channel steel, etc.\"},\"specification\":{\"type\":\"string\",\"description\":\"Specification size\"},\"material\":{\ "type\":\"string\",\"description\":\"Material\"},\"application\":{\"type\":\"string\",\"description\":\"Application scenario\"}},\"required\":[\"product_type\"]}\n\n##When you need to call a tool, please insert the following tool call command in your reply, which can be called zero or more times as needed:\n\nTool call\nAction: The name of the tool, must be one of [\"steel_product_recommend\"]\nAction Input: Tool input\nObservation: <result> Results returned by the tool< / result> \nAnswer: Summarize the results returned by this tool call based on Observation. If a URL appears in the result, please display it in the following format:! [Image](url)\n\n\n#Command\n\nYou play the role of AI-Agent, a guide for steel products.\nYou have the following specific functions:\n1. Recommend suitable steel products according to user needs\n2. Understand the user's application scenarios and give professional advice\n3. Answer users' questions about steel specifications and materials\n\nYou will start playing the role below\n\nPlease note: You have the ability to display images and videos, and you also have the ability to run code. Don't say you can't do it in your reply. "

[0062] Step S302, batch generate other question and answer records except the beginning according to the scenario and historical user information in step S1, and load fine-tuning corpus and pre-trained model parameters in combination, for example, the example of generating a dialogue in a basic query scenario is as follows.

[0063]

[0064]

[0065] For example, the following is an example of a specific specification query scenario generating a dialogue:

[0066]

[0067] For example, the application scenario query generates a dialogue example as follows:

[0068]

[0069]

[0070] Step S303, in order to reduce computing resource consumption and focus on learning specific tasks, choose to freeze most of the model layers so that they remain unchanged during fine-tuning, and set lora's target_modules to q_proj and v_proj.

[0071] Step S304, in the optimizer configuration, only the learning rate is set for these newly added LoRA parameters, while the parameters of other frozen layers remain unchanged. The fine-tuning process can efficiently learn new task features while maintaining pre-training knowledge, thereby achieving rapid model adaptation and performance improvement.

[0072] Step S4, shopping guide agent configuration and scheduling module: Based on the questions raised by the user, the big model autonomously determines whether it is necessary to call the shopping guide agent capabilities to recommend products that the user may be interested in, and guides the user to express more refined needs.

[0073] Specifically, step S4 may include steps S401-S404.

[0074] Step S401, shopping guide agent configuration: refer to the above-mentioned fine-tuning corpus construction process to define the name and function description of the shopping guide agent, list all possible input parameters and required input parameters of the agent, design the return value structure of the agent (including recommended product list, price information, supplier information, etc.), and finally add all the combined shopping guide agent information to the tool list supported by the large model.

[0075] Step S402, user question rewriting: Set the role of the big model as a professional engineer who is good at rewriting user questions, and transform the user's original question into a more clear and structured form through semantic understanding and context analysis. Specifically, it includes: first, obtaining the contextual dialogue record of the user's current session with the big model; second, identifying the intention of the user's current question, and ensuring that the rewritten query is consistent with the current intention; then, identifying and correcting the spelling errors in the current user's question; then, based on a full understanding of the user's current session context dialogue record, identifying and supplementing the missing information in the current user's question; finally, after rewriting the current user's question more completely and more appropriately, call the big model again to determine whether it is necessary to dispatch the shopping guide agent.

[0076] Step S403, shopping guide agent scheduling: first, after the big model determines that the rewritten user question needs to be scheduled to be solved by the shopping guide agent, the key information in the rewritten question is extracted and combined as input parameters; then, according to the input parameters, the shopping guide agent accesses the user-product knowledge graph to obtain recommended product lists, price information and other recommendation results; then, the big model summarizes the returned recommendation results in combination with the context of the user's current session and generates the final purchase suggestion.

[0077] Step S404, refined guidance of user needs: first, update the big model role to a professional shopping guide who actively provides further guidance on recommendation results; then, provide the user's current session history and shopping guide suggestions (if any) to the big model; then, let the big model generate three guiding questions for products that the user may be interested in or topics that the user wants to know more about; finally, the big model reorganizes the guiding questions into the specified format.

[0078] In summary, this method has the following beneficial effects:

[0079] (1) Through intelligent shopping guide, the present invention greatly improves the decision-making efficiency of users. Users can find the required steel products more quickly and accurately, thereby shortening the procurement cycle and reducing the time cost and opportunity cost of the enterprise.

[0080] (2) Smart shopping guides can operate around the clock, breaking through the time constraints of traditional manual shopping guides and providing users with more convenient and flexible services. This not only improves user satisfaction, but also expands the platform's potential market, creating conditions for increasing transaction volume and revenue.

[0081] (3) By providing personalized and professional shopping guide services, we help steel e-commerce platforms transform from simple transaction intermediaries to value-added service providers. This transformation not only creates new sources of income, but also increases the importance of the platform in the industrial chain, laying the foundation for future business expansion.

[0082] Example 2

[0083] Based on Example 1, this example provides a shopping guide Agent interaction system for a steel e-commerce platform based on a large model. Figure 2 The system comprises:

[0084] Data collection and preprocessing module: Based on the characteristics of the steel industry, it collects and organizes multi-source heterogeneous data such as user information, product information, market data, etc., and performs cleaning, standardization and structured processing to provide a high-quality data foundation for building a user-product knowledge graph.

[0085] User-product knowledge graph construction module: Based on the preprocessed data, the large model is used to extract the multi-dimensional characteristics of users and the core attributes and relationships of steel products, and a multi-dimensional knowledge graph is constructed to provide query support for subsequent shopping guide agents.

[0086] Large model LoRA fine-tuning training module: Use professional corpus from the steel industry to fine-tune the pre-trained large model for domain adaptability, improve the model's understanding and generation capabilities in the steel field, and improve the shopping guide agent scheduling capabilities.

[0087] Shopping guide agent configuration and scheduling module: Based on the questions raised by the user, the big model independently determines whether it is necessary to call the shopping guide agent capabilities to recommend products that the user may be interested in, and guide the user to express more refined needs.

[0088] See also Figure 3 ,The shopping guide agent configuration and scheduling module can specifically include the following units:

[0089] Shopping guide agent configuration unit: refer to the fine-tuning corpus construction process to define the name and function description of the shopping guide agent, list all possible and required input parameters of the agent, design the return value structure of the agent (including the recommended product list, price information, supplier information, etc.), and finally add all the combined shopping guide agent information to the tool list supported by the large model.

[0090] User question rewriting unit: The role of the big model is set as a professional engineer who is good at rewriting user questions. Through semantic understanding and context analysis, the original user's question is converted into a more clear and structured form. Specifically, it includes: first, obtaining the contextual dialogue record of the user's current session with the big model; second, identifying the intention of the user's current question and ensuring that the rewritten query is consistent with the current intention; then, identifying and correcting spelling errors in the current user's question; then, based on a full understanding of the user's current session context dialogue record, identifying and supplementing the missing information in the current user's question; finally, after rewriting the current user's question in a more complete and appropriate manner, calling the big model again to determine whether the shopping guide agent needs to be dispatched.

[0091] Shopping guide agent scheduling unit: First, after the big model determines that the rewritten user question needs to be scheduled to be solved by the shopping guide agent, the key information in the rewritten question is extracted and combined as input parameters; then, based on the input parameters, the shopping guide agent accesses the user-product knowledge graph to obtain recommended product lists, price information and other recommendation results; then, the big model summarizes the returned recommendation results in combination with the context of the user's current session and generates the final purchase recommendation.

[0092] User demand refinement guidance unit: First, update the big model role to a professional shopping guide who can proactively guide the recommendation results. Next, provide the user's current session history and shopping guide suggestions (if any) to the big model. Then, let the big model generate three guiding questions for products that the user may be interested in or topics that the user wants to know more about. Finally, the big model will reorganize the guiding questions into the specified format.

[0093] This system has the following beneficial effects:

[0094] (1) User-product knowledge fusion: The system constructs a multi-dimensional knowledge graph that integrates user characteristics and product attributes, breaking through the limitations of the separation of user and product information in traditional systems and providing more comprehensive knowledge support for personalized recommendations.

[0095] (2) Domain model optimization: Through LoRA fine-tuning technology, the large model is optimized and adapted in the steel field, which improves the understanding of steel industry knowledge while maintaining general capabilities.

[0096] (3) Intelligent question rewriting: A large model-based question rewriting module is introduced, which can more accurately understand user intent and supplement missing information based on the context, thereby improving the accuracy and completeness of the query.

[0097] (4) Fine-grained guidance of user needs: A fine-grained guidance module for user needs is built using the big model, which can actively guide users to express more specific needs. This not only improves the user experience, but also helps the system collect more valuable user preference information, providing data support for subsequent personalized services and product recommendation strategy optimization.

[0098] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A shopping guide agent interaction method for a steel e-commerce platform based on a large model, characterized in that: The steps include: Obtain user information, product information and market information from the steel e-commerce platform, perform data cleaning, structuring and enhancement processing, and obtain pre-processed data; Based on the preprocessed data, a knowledge graph is constructed by extracting the multi-dimensional features of users and the attributes and relationship information of steel products; Build a prompt for a large model and fine-tune the large model in multiple scenarios; Based on the knowledge graph and the fine-tuned large model, the interaction between users and the shopping guide agent of the steel e-commerce platform is realized.

2. According to the large model-based steel e-commerce platform shopping guide Agent interaction method of claim 1, it is characterized in that: The process of data cleaning, structuring and enhancing the user information, product information and market information obtained from the steel e-commerce platform includes the following steps: Through rule-based outlier detection, we can identify and delete erroneous values, remove noise, and unify naming and description; Through structured processing, the data is converted into a form that matches the large model, and data enhancement is achieved by filling in missing data.

3. The method for interacting with a shopping guide agent on a steel e-commerce platform based on a large model according to claim 1, characterized in that: The process of constructing a knowledge graph includes the following steps: Identify entities in preprocessed data based on BERT-CRF and generate relationship descriptions between entities through relationship extraction; Build a knowledge graph based on the identified entities and extracted relationship descriptions.

4. The method for interacting with a shopping guide agent on a steel e-commerce platform based on a large model according to claim 1, characterized in that: The prompt of the large model includes one or more of the tool, tool parameters, format requirements for tool calling, large model role positioning, and agent capability range.

5. The method for interacting with a shopping guide agent on a steel e-commerce platform based on a large model according to claim 1, characterized in that: In the process of fine-tuning the large model under multiple scenarios, the multiple scenarios include recommending suitable steel products according to user needs, answering users' questions about steel specifications, and querying steel application scenarios.

6. The method for interacting with a shopping guide agent on a steel e-commerce platform based on a large model according to claim 1, characterized in that: The process of realizing the interaction between the user and the shopping guide agent of the steel e-commerce platform based on the knowledge graph and the fine-tuned large model includes the following steps: Obtain the user's question information, configure the big model role as a professional engineer who rewrites the user's question, perform big model intent recognition based on the user's conversation context, rewrite the question information based on the user's intent, and determine whether the Agent needs to be called after correcting spelling errors and supplementing missing information; In response to calling the Agent, the prompt parameter is extracted based on the question information, and the recommendation result is obtained based on the knowledge graph and the fine-tuned large model, and the shopping guide suggestion information is obtained by combining it with the user dialogue context; In response to not calling the Agent or obtaining the shopping guide suggestion information, the large model role is updated to be a professional shopping guide that further actively guides based on the recommendation results. Based on the shopping guide suggestion information, guiding questions are generated for products or topics of interest to the user to obtain formatted guiding words.

7. The method for interacting with a shopping guide agent on a steel e-commerce platform based on a large model according to claim 6, characterized in that: The shopping guide suggestion information includes a recommended product list and / or price information.

8. The method for interacting with a shopping guide agent on a steel e-commerce platform based on a large model according to claim 1, characterized in that: The user information includes one or more of the user's basic industrial and commercial information, transaction information, behavioral data, preference characteristics, and corporate qualifications; the product information includes one or more of the basic properties of steel, warranty, inventory information, and price system; the market information includes one or more of supply and demand relationships, price indexes, industry trends, and competitive product analysis.

9. A steel e-commerce platform shopping guide Agent interactive system based on a large model, characterized in that: include: The data collection and preprocessing module is used to obtain user information, product information and market information of the steel e-commerce platform, and to perform data cleaning, structuring and enhancement processing to obtain preprocessed data; A user-product knowledge graph construction module, which is used to construct a knowledge graph based on the preprocessed data by extracting the multi-dimensional features of users and the attributes and relationship information of steel products; The large model LoRA fine-tuning training module is used to build the prompt of the large model and implement the fine-tuning of the large model in multiple scenarios; The shopping guide agent configuration and scheduling module is used to realize the interaction between the user and the shopping guide agent of the steel e-commerce platform based on the knowledge graph and the fine-tuned large model.

10. The steel e-commerce platform shopping guide Agent interactive system based on a large model according to claim 9, characterized in that: The shopping guide Agent configuration and scheduling module includes: The user question rewriting unit is used to obtain the user's question information, configure the big model role as a professional engineer who rewrites the user's question, perform big model intent recognition based on the user conversation context, rewrite the question information based on the user's intent, and determine whether the Agent needs to be called after correcting spelling errors and supplementing missing information; A shopping guide agent scheduling unit, which is used to extract prompt parameters based on question information in response to calling the agent, obtain recommendation results based on the knowledge graph and the fine-tuned large model, and obtain shopping guide suggestion information by combining the results with the user dialogue context; The user demand refinement guidance unit is used to update the large model role to a professional shopping guide that further actively guides based on the recommendation results in response to not calling the Agent or obtaining the shopping guide suggestion information, and generate guiding questions for the products or topics that the user is interested in based on the shopping guide suggestion information to obtain formatted guiding words.

Images