Demand-based multi-agent e-commerce recommendation methods, systems, devices, and media

Abstract

This invention relates to the field of e-commerce technology, disclosing a demand-based multi-agent e-commerce recommendation method, comprising: acquiring user needs, profiles, and historical behaviors; extracting demand types and acquiring corresponding agents; loading user data into a context management system to construct a multi-agent system containing scheduling, understanding, recommendation, and explanation agents; the scheduling agent parsing the needs and session state to determine whether to invoke the understanding agent; if invoked, the understanding agent semantically mines and generates structured needs; the recommendation agent, based on the structured needs, obtains a candidate set from a product library and combines and optimizes it to obtain product solutions; the explanation agent generates personalized explanation content based on the solutions and profiles and outputs it to the terminal to complete the recommendation. This invention also proposes a demand-based multi-agent e-commerce recommendation system, device, and storage medium. This invention can improve the accuracy of user need understanding in e-commerce recommendations.

Description

Technical Field

[0001] This invention relates to the field of e-commerce technology, and in particular to a demand-based multi-agent e-commerce recommendation method, system, device, and medium. Background Technology

[0002] Traditional e-commerce platform technologies primarily rely on keyword search and collaborative filtering recommendations based on user profiles. Keyword search technology is based on text matching and relevance ranking, requiring users to explicitly input product names or attribute keywords. The search process is simplistic and cannot create a continuous, multi-round iterative search experience. When users only know their problem (e.g., "the air in my home is not fresh") but not specific products, this technology struggles to directly match suitable items. While collaborative filtering recommendations based on user profiles can push products based on historical behavior, the results are typically static lists or periodically updated, lacking real-time context awareness. Furthermore, it is limited to single-product or single-category recommendations, making it difficult to generate cross-category or cross-scenario combined recommendation schemes. This results in low decision-making efficiency and insufficient information utilization for users facing complex shopping scenarios.

[0003] In recent years, intelligent dialogue systems based on large models have shown application potential in the e-commerce field, capable of understanding user intent through natural language interaction. However, existing technologies typically employ a single agent to handle all requests or only implement simple task distribution mechanisms, lacking deep collaboration and context sharing capabilities among multiple agents. A single agent, limited by its fixed functional boundaries, struggles to simultaneously handle complex tasks such as demand mining, product retrieval, combination recommendations, and personalized explanations. Furthermore, it is prone to losing historical information during long conversations, leading to insufficient service continuity. In addition, existing solutions fail to build a unified context management system; data is fragmented between functional modules, hindering the dynamic fusion and real-time synchronization of user profiles, historical behavior, and real-time conversation information, thus limiting the level of service personalization and response efficiency. Summary of the Invention

[0004] This invention provides a demand-based multi-agent e-commerce recommendation method, system, device, and medium, the main purpose of which is to improve the accuracy of understanding user needs in e-commerce recommendations.

[0005] To achieve the above objectives, the present invention provides a demand-based multi-agent e-commerce recommendation method, comprising: Acquire the natural language needs, user profiles, and historical behavior data of the target user; extract the need types of the natural language needs; and acquire multiple intelligent agents based on the need types. The user profile and the historical behavior data are loaded into a preset context management system to obtain a standard management system. A multi-agent system is constructed based on the standard management system and the multiple agents. The multi-agent system includes a task scheduling agent, a demand understanding agent, a search and recommendation agent, and a product explanation agent. Acquire the session state data in the multi-agent system, use the task scheduling agent to parse the natural language requirement and the session state data to obtain the parsing result, and determine whether it is necessary to call the requirement understanding agent based on the parsing result; When a call is needed, the intelligent agent that understands the requirements performs semantic mining on the natural language requirements to obtain supplementary semantic elements, and generates structured requirement information based on the supplementary semantic elements. Based on the structured demand information, the search and recommendation agent obtains a set of candidate products from a preset product library, and optimizes the combination of the candidate product set to obtain a product combination scheme. Based on the product explanation intelligent agent, personalized explanation content is generated according to the product combination scheme and user profile, and the product combination scheme and the personalized explanation content are output to the preset user terminal to complete the e-commerce recommendation.

[0006] Optionally, the step of extracting the demand type of the natural language requirement and obtaining multiple agents based on the demand type includes: The natural language requirements are semantically understood using a pre-defined large language model to obtain semantic understanding results; The semantic understanding results are mapped to a preset scene classification system to obtain the demand type corresponding to the natural language demand; Obtain the scene tags, function tags, and domain knowledge tags associated with the corresponding demand type for the natural language demand, and select the corresponding intelligent agents from the preset intelligent agent library according to the scene tags, function tags, and domain knowledge tags to obtain the multiple intelligent agents.

[0007] Optionally, loading the user profile and the historical behavior data into a preset context management system to obtain a standard management system includes: The user profile and the historical behavior data are standardized to obtain standardized user data. The preset context management system is initialized and configured to obtain the initial context management system, and the standardized user data is written into the initial context management system to obtain the standard management system.

[0008] Optionally, the step of using the task scheduling agent to parse the natural language requirement and the session state data to obtain the parsing result includes: Obtain the dialogue state parsing model in the task scheduling agent, and standardize the natural language requirements and the conversation state data to obtain standard language requirements and standard conversation state data. The dialogue state parsing model is used to perform semantic parsing on the standard language requirements to obtain semantic parsing results, and the dialogue state parsing model is used to perform context parsing on the standard conversation state data to obtain context parsing results. The semantic parsing result and the context parsing result are fused and verified to determine the consistency and information completeness of the natural language requirement and the session state data, and the verification result is obtained. Based on the verification results, the semantic parsing results and the context parsing results are structurally integrated and validated to obtain the parsing results.

[0009] Optionally, the step of performing semantic mining on the natural language requirement through the requirement understanding agent to obtain supplementary semantic elements includes: The intelligent agent that understands the requirements extracts the core semantics of the natural language requirements to obtain basic semantic elements. The basic semantic elements are fused with the user profile and the historical behavior data to obtain fusion requirement information; The fusion requirement information is inferred and completed using a pre-set common sense base and a pre-set domain knowledge base to obtain full semantic information; The full set of semantic information is structured and validated to obtain the supplementary semantic elements.

[0010] Optionally, generating structured requirement information based on the supplementary semantic elements includes: Obtain a preset structured requirement template and the dimension system of the structured requirement template. Based on the dimension system of the structured requirement template, decompose the supplementary semantic elements and fill the slots to obtain the filling template information. The filled template information is fused with the user profile and the historical behavior data to obtain fused requirement information. The fused requirement information is then quantified and logically verified to obtain preliminary structured requirement information. The preliminary structured requirement information is standardized to obtain the structured requirement information.

[0011] Optionally, the step of obtaining a set of candidate products from a preset product library using the search recommendation agent based on the structured demand information includes: Extract the requirement feature vector, label system, and constraints from the structured requirement information; Based on the demand feature vector, the tag system, and the constraints, products corresponding to the structured demand information are obtained from the preset product library to obtain an initial candidate product set; The initial candidate product set is deduplicated and its validity is verified to obtain the candidate product set.

[0012] To address the aforementioned problems, the present invention also provides a demand-based multi-agent e-commerce recommendation system, the system comprising: The system construction module is used to acquire the natural language needs, user profiles and historical behavior data of the target user, extract the need types of the natural language needs, and acquire multiple intelligent agents based on the need types; The user profile and the historical behavior data are loaded into a preset context management system to obtain a standard management system. A multi-agent system is constructed based on the standard management system and the multiple agents. The multi-agent system includes a task scheduling agent, a demand understanding agent, a search and recommendation agent, and a product explanation agent. The requirement analysis module is used to acquire the session state data in the multi-agent system, use the task scheduling agent to parse the natural language requirement and the session state data to obtain the parsing result, and determine whether it is necessary to call the requirement understanding agent based on the parsing result. When a call is needed, the intelligent agent that understands the requirements performs semantic mining on the natural language requirements to obtain supplementary semantic elements, and generates structured requirement information based on the supplementary semantic elements. The product optimization module is used to obtain a set of candidate products from a preset product library based on the structured demand information using the search recommendation agent, and to combine and optimize the set of candidate products to obtain a product combination scheme. The product recommendation module is used to generate personalized explanation content based on the product explanation intelligent agent, the product combination scheme and user profile, and output the product combination scheme and the personalized explanation content to the preset user terminal to complete the e-commerce recommendation.

[0013] To address the above problems, the present invention also provides an electronic device, the electronic device comprising: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the demand-based multi-agent e-commerce recommendation method as described above.

[0014] To address the aforementioned problems, the present invention also provides a computer-readable storage medium, including a data storage area and a program storage area. The data storage area stores created data, and the program storage area stores a computer program. When the computer program is executed by a processor, it implements the demand-based multi-agent e-commerce recommendation method described above.

[0015] This invention acquires user needs, profiles, and historical behaviors, extracts need types, and obtains corresponding intelligent agents. User data is loaded into a context management system to construct a multi-agent system containing scheduling, understanding, recommendation, and explanation agents. The scheduling agent parses the needs and session state to determine whether to invoke the understanding agent. If invoked, the understanding agent performs semantic mining to generate structured needs. The recommendation agent retrieves a candidate set from the product library based on the structured needs and combines and optimizes it to obtain product solutions. The explanation agent generates personalized explanation content based on the solutions and profiles and outputs it to the terminal to complete the recommendation. Therefore, the need-based multi-agent e-commerce recommendation method, system, electronic device, and computer-readable storage medium proposed in this invention, by constructing a consultative e-commerce service architecture of "multi-agent collaboration + context sharing," enables the system to deeply mine real needs from fuzzy user input and dynamically generate scenario-based product combination solutions. This solves the technical problems of traditional e-commerce search processes being simplistic, recommendations lacking scenario awareness, and difficulty in cross-category combinations, thus improving the accuracy of user need understanding in e-commerce recommendations. Attached Figure Description

[0016] Figure 1 A flowchart illustrating a demand-based multi-agent e-commerce recommendation method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a demand-based multi-agent e-commerce recommendation system provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the internal structure of an electronic device that implements a demand-based multi-agent e-commerce recommendation method according to an embodiment of the present invention.

[0017] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0019] This application provides a demand-based multi-agent e-commerce recommendation method. The executing entity of the demand-based multi-agent e-commerce recommendation method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application embodiment: a server, a terminal, etc. The server can be an independent server or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms. In other words, the demand-based multi-agent e-commerce recommendation method can be executed by software or hardware installed on remote devices or server-side devices, and the software can be a blockchain platform. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0020] Reference Figure 1 The diagram shown illustrates a flowchart of a demand-based multi-agent e-commerce recommendation method according to an embodiment of the present invention. In this embodiment, the demand-based multi-agent e-commerce recommendation method includes the following steps S1-S6: S1. Obtain the natural language needs, user profile and historical behavior data of the target user, extract the need type of the natural language needs, and obtain multiple intelligent agents according to the need type.

[0021] Understandably, by parsing the type of demand from user input and matching it with the corresponding intelligent agent role, it can be ensured that subsequent steps are handled by intelligent agents with the corresponding domain capabilities, thereby avoiding the efficiency loss and adaptation deviation caused by generalized processing and laying the foundation for personalized services.

[0022] User profiles refer to the tagged descriptions of users' static attributes and long-term preferences, covering basic attributes (age, gender, region), interests and preferences (favorite categories, brands), spending power (average order value, price sensitivity), and identity tags (such as "outdoor enthusiast"). They reflect the user's macro-level characteristics, providing background information for each agent to offer personalized services, and ensuring that recommendation strategies match user identities.

[0023] Historical behavioral data refers to records of users' past interactions with the platform, including browsing history, search history, transaction history, favorites and add-to-cart entries, and past conversations. This dynamic data reveals users' true preferences and behavioral patterns, serving as contextual references for understanding user needs and enabling refined operations such as avoiding duplicate recommendations and identifying repeat purchase needs.

[0024] Among them, the demand type refers to the standardized label obtained by mapping the natural language demand to the preset classification system after semantic parsing. It usually includes multiple dimensions such as scenario, category, and intent (such as "Scenario: Outdoor Camping" and "Intent: Equipment Procurement"). It is a bridge connecting the user's fuzzy input with the system's preset intelligent agent combination and is used to determine which professional intelligent agents to call in the future.

[0025] Here, "multiple agents" specifically refers to a set of agent types matched from a pre-defined agent role library based on the type of need (such as a need understanding agent, a search and recommendation agent, a product explanation agent, etc.). At this stage, they have not yet been instantiated or loaded with user data; they serve only as a "job list" to provide a role framework for building a personalized service team later.

[0026] In this context, an intelligent agent refers to an autonomous processing unit within a multi-agent system that possesses specific professional expertise. It can independently undertake service tasks at a particular stage and collaborate with other intelligent agents through a context management system. Each intelligent agent possesses domain-specific processing capabilities, enabling real-time decision-making and dynamic responses based on user data, collectively completing the entire service process from understanding user needs to recommending products.

[0027] For example, when a user's demand type is "outdoor camping equipment purchase", the system will match intelligent agent types such as demand understanding, search recommendation, and product explanation, laying the foundation for subsequent instantiation and collaborative services based on user data.

[0028] Furthermore, the step of extracting the demand type of the natural language requirement and obtaining multiple agents based on the demand type includes: The natural language requirements are semantically understood using a pre-defined large language model to obtain semantic understanding results; The semantic understanding results are mapped to a preset scene classification system to obtain the demand type corresponding to the natural language demand; Obtain the scene tags, function tags, and domain knowledge tags associated with the corresponding demand type for the natural language demand, and select the corresponding intelligent agents from the preset intelligent agent library according to the scene tags, function tags, and domain knowledge tags to obtain the multiple intelligent agents.

[0029] Among them, the large language model refers to a deep learning model that has been pre-trained on massive amounts of text data and is able to understand and generate natural language. In this step, it is used to perform semantic parsing on the natural language requirements input by the user, extract the core intent and key information, and provide a basis for determining the type of requirement in the future.

[0030] The semantic understanding result refers to the structured or semi-structured information output by the large language model after parsing the natural language requirements. It includes the user's core intent, key entities, intent tendencies, etc. It is a deep processing product of the original input and is used to map it to the scene classification system.

[0031] The scenario classification system refers to a pre-built standardized classification framework that covers various business scenarios of the platform. It includes multiple scenario categories and their hierarchical relationships. Each scenario category is associated with a corresponding tag system and intelligent agent configuration rules, which are used to transform semantic understanding results into standardized demand types.

[0032] Among them, the demand type corresponding to natural language demand refers to the standardized scene label obtained after mapping the semantic understanding results to the scene classification system, which identifies the business scene category to which the user's current demand belongs (such as "outdoor camping", "beauty and skin care", "digital office"), and is the core basis for matching intelligent agent roles in the future.

[0033] Among them, the scenario tags, function tags, and domain knowledge tags associated with the demand type refer to a set of multi-dimensional tags that are preset and bound to a specific demand type. The scenario tag identifies the major category of the demand, the function tag identifies the specific function to be implemented (such as "demand understanding" or "product retrieval"), and the domain knowledge tag identifies the professional knowledge domain to be invoked (such as "outdoor equipment knowledge" or "cosmetic ingredient knowledge"), which are used to accurately match intelligent agent roles with corresponding capabilities from the intelligent agent library.

[0034] For example, user Zhang San inputs a natural language request on an e-commerce platform: "What do I need to prepare for a camping trip during the holidays?" The system immediately retrieves Zhang San's user profile (30 years old, male, outdoor enthusiast, residing in Beijing) and historical behavioral data (purchased a single tent last year, frequently browses outdoor equipment, and searched for "camping guides"). Using a large language model, the system analyzes the request and extracts the request type as "outdoor camping equipment purchase." Based on this request type, the system matches a request understanding agent, a search and recommendation agent, a product explanation agent, and an after-sales agent from a pre-set agent role library, forming a list of agents for this service.

[0035] S2. Load the user profile and the historical behavior data into a preset context management system to obtain a standard management system. Construct a multi-agent system based on the standard management system and the multiple agents, wherein the multi-agent system includes a task scheduling agent, a demand understanding agent, a search and recommendation agent, and a product explanation agent.

[0036] Understandably, by injecting user data to activate the context management system, a set of static intelligent agent roles are instantiated into a dynamic multi-agent system with unified memory, clear division of labor, and the ability to collaborate in real time through a data hub. This lays the architectural foundation and data cornerstone for the subsequent realization of personalized and coherent consultative e-commerce services.

[0037] Among them, a multi-agent system refers to a collaborative system composed of multiple agents with different functions. Each agent shares user data and session state in real time through a context management system, and works together to complete the entire service process from demand understanding to product recommendation in a collaborative manner.

[0038] Among them, the task scheduling agent, as the central coordinator of the multi-agent system, is responsible for receiving user requests, parsing natural language requirements and the current conversation state, intelligently deciding which agent to call next, and monitoring the execution status and task completion of each agent to ensure that the entire collaborative process proceeds efficiently and orderly.

[0039] For example, if a user says, "What do I need to prepare for a camping trip during the holidays?", the task scheduling agent first takes this question and then checks the "work log" (i.e., the context management system) to see the current state—has the user said anything else? What stage were we at before? Is the information sufficient? Then, it analyzes the user's new question together with the existing logs and determines: "The user only said they want to go camping, but didn't say how many people they'll be with or how much it will cost; the information is insufficient." Therefore, it decides: "In this case, the 'requirement understanding agent' needs to step in and find out." Then, it notifies the requirement understanding agent to begin its work.

[0040] Among them, the demand understanding agent is used for deep mining and accurate understanding of user needs, and can handle users' ambiguous or cross-category shopping problems. The demand understanding agent performs semantic parsing of natural language needs through a large language model, and combines user profiles, historical behavior and multi-turn interaction information to transform users' implicit needs into a structured demand vector and tag system, providing accurate basis for subsequent retrieval and recommendation.

[0041] For example, in a camping scenario, the agent asks the user questions such as "How many people are traveling? What is the budget? Where are you going camping?" and combines the user profile (such as having bought a single tent last year) to infer that the user may need family equipment, and finally generates structured demand information containing "3 people, budget 5000, camping in the suburbs".

[0042] Among them, the search recommendation agent is responsible for the core functions of product retrieval and combination recommendation. It adopts a hybrid strategy that combines vector retrieval and text retrieval to obtain a set of candidate products that match the structured demand information from the preset product library. It also combines user profiles and consumption scenarios to optimize cross-category and multi-product combinations and generate multiple scenario-based product combination schemes.

[0043] For example, based on structured demand information, it begins searching for suitable products in the "warehouse" (product library): first, it looks for semantically similar products based on meanings like "camping" and "family"; then, it further filters using specific conditions such as "3 people" and "under 5000 yuan". After finding a batch of tents, sleeping bags, and sleeping mats, it removes duplicates and filters out unsuitable ones, resulting in a batch of candidate products. Next, it combines these candidate products into several complete solutions: an "economy" solution: selecting entry-level models with high cost-performance ratios; a "comfort" solution: selecting reputable brand sets; and a "luxury" solution: selecting lightweight and feature-rich upgraded models. Finally, these three solutions are handed over to the next intelligent agent to be displayed to the user.

[0044] The product explanation AI is responsible for providing multi-dimensional and personalized presentations and explanations of recommended product combinations. It dynamically adjusts the focus of the explanation based on user profiles (e.g., emphasizing nutritional value for health-conscious users, and highlighting packaging and brand for gift-givers), enhances user perception of the products through various formats such as images, text, and videos, and dynamically optimizes the explanation content through interaction with user behavior, thereby improving purchase conversion rates.

[0045] Furthermore, the task scheduling agent acts as the central coordinator for all aspects, the demand understanding agent is responsible for converting the user's fuzzy input into precise instructions, the search and recommendation agent completes product retrieval and combination optimization based on the instructions, and the product explanation agent presents the recommendation results to the user in a personalized way. The four agents work together seamlessly through the context management system to achieve a complete consultative service loop from the user's fuzzy needs to personalized product recommendations.

[0046] Furthermore, the step of loading the user profile and the historical behavior data into a preset context management system to obtain a standard management system includes: The user profile and the historical behavior data are standardized to obtain standardized user data. The preset context management system is initialized and configured to obtain the initial context management system, and the standardized user data is written into the initial context management system to obtain the standard management system.

[0047] The context management system refers to a central data platform pre-set in a multi-agent system for centralized storage and management of user data and session states. It has a data storage structure, agent communication interface and session state management module. It can synchronize user profiles, historical behavior data, current session dynamic information and task execution status of each agent in real time among multiple agents, and provide unified data access and sharing services for all agents, ensuring information consistency, real-time performance and traceability in the multi-agent collaboration process.

[0048] For example, the system loads Zhang San's user profile and historical behavior data into a pre-defined context management system. After data cleaning, format conversion, and feature extraction, the context management system is initialized, resulting in a standard management system containing Zhang San's personalized data. Subsequently, based on this standard management system and the list of agents obtained in the first step, the system instantiates and constructs a multi-agent system. This system includes a task scheduling agent, a demand understanding agent, a search and recommendation agent, and a product explanation agent. These agents share user data and interaction states in real time through the context management system.

[0049] S3. Obtain the session state data in the multi-agent system, use the task scheduling agent to parse the natural language requirement and the session state data to obtain the parsing result, and determine whether it is necessary to call the requirement understanding agent based on the parsing result.

[0050] Understandably, by jointly parsing the user's current input and the existing conversation context through the task scheduling agent, the system can intelligently judge the completeness of the demand information and autonomously decide whether to activate the demand understanding agent for in-depth mining. This allows the system to skip redundant interactions when information is sufficient and to promptly supplement information when it is insufficient, thus achieving a dynamic balance between response efficiency and service depth.

[0051] Session state data refers to the collection of all dynamic information and state markers accumulated and stored in the context management system during the process of a multi-agent system serving the current user. This includes supplementary user information collected in the current session, the task execution status of each agent, the history of this round of interactions, and context markers identifying the session phase. This data provides the task scheduling agent with real-time decision-making support, enabling it to accurately grasp the current service progress, assess the completeness of information, and intelligently determine which agent to invoke next or whether supplementary information is needed, ensuring the continuity of multi-round interactions and the orderly collaboration of agents.

[0052] The parsing result refers to the standardized output generated by the task scheduling agent after jointly parsing the user's natural language needs and the current session state data. It includes the user's core intent, information completeness assessment, agent invocation decision, and items to be supplemented. This result serves as the decision instruction for the task scheduling agent, clearly indicating whether the next step is to invoke the needs understanding agent for in-depth analysis or directly proceed to the retrieval and recommendation stage. This avoids blind invocation or repeated queries, improving system response efficiency and resource utilization while ensuring information completeness.

[0053] Furthermore, the step of using the task scheduling agent to parse the natural language requirements and the session state data to obtain the parsing result includes: Obtain the dialogue state parsing model in the task scheduling agent, and standardize the natural language requirements and the conversation state data to obtain standard language requirements and standard conversation state data. The dialogue state parsing model is used to perform semantic parsing on the standard language requirements to obtain semantic parsing results, and the dialogue state parsing model is used to perform context parsing on the standard conversation state data to obtain context parsing results. The semantic parsing result and the context parsing result are fused and verified to determine the consistency and information completeness of the natural language requirement and the session state data, and the verification result is obtained. Based on the verification results, the semantic parsing results and the context parsing results are structurally integrated and validated to obtain the parsing results.

[0054] The semantic parsing result refers to the structured information output by the large language model after semantically understanding the standardized natural language requirements. It includes the user's current core intent, key entities, demand tendencies, and sentiment tendencies. It is a deep analysis product of the user's current expression and is used for subsequent fusion and verification with contextual information.

[0055] The dialogue state parsing model refers to a lightweight processing module within the task scheduling agent used to analyze the current conversation state. It is responsible for classifying and parsing conversation progress, information completeness, and the task states of each agent, outputting structured state labels. The difference between this model and the large language model mentioned earlier is that the dialogue state parsing model does not undertake semantic understanding tasks; it only performs state classification and rule matching, resulting in low computational overhead and fast response time.

[0056] Among them, the context parsing result refers to the structured information output by the large language model after parsing the standardized session state data. It includes the user supplementary information accumulated in the current session, the task execution status of each agent, the interaction history, and the session stage marker. It is a quantitative expression of the current service progress status and is used to evaluate the completeness of the existing information.

[0057] The verification result refers to the judgment conclusion obtained after comparing the consistency and completeness of the semantic parsing result and the context parsing result. It includes whether there is a contradiction between the user's current needs and the existing context, whether the current information is sufficient to support the next operation, whether there is any missing information and the specific dimensions of the missing information, which serve as the basis for subsequent structured integration and agent invocation decisions.

[0058] Furthermore, the task scheduling agent retrieves the current session state data (currently an empty session) from the context management system, while simultaneously receiving the user's original request, "What do I need to prepare for a camping trip?" It jointly analyzes the natural language request and the session state, discovering that the user has only expressed a vague intent, lacking key information such as the number of travelers, budget, and destination, resulting in insufficient information completeness. Therefore, the analysis result determines that a request understanding agent needs to be invoked for in-depth analysis.

[0059] S4. When a call is required, the intelligent agent that understands the requirements performs semantic mining on the natural language requirements to obtain supplementary semantic elements, and generates structured requirement information based on the supplementary semantic elements.

[0060] Understandably, by using a demand-understanding intelligent agent to perform multi-dimensional semantic analysis and multi-round interactive guidance on users' vague expressions, the system can transform users' implicit needs into explicit structured information, fill the information blind spots in the initial input, and thus transform vague intentions such as "I want to buy something" into precise instructions that can be executed in subsequent searches, significantly improving the accuracy and completeness of demand understanding.

[0061] Among them, supplementary semantic elements refer to the key information units extracted by the demand understanding agent after semantic mining of the user's original natural language demand, multi-round interactive guidance, and inference by combining user profiles and historical behavior data. These elements are used to fully describe the user's intentions. They include the content explicitly expressed by the user, the implicit intentions, and the missing information supplemented by the interaction. They expand the information dimensions and refine the content of the original fuzzy input, providing basic materials for the subsequent generation of structured demand information.

[0062] Among them, structured demand information refers to the standardized demand data package generated by the demand understanding agent based on supplementary semantic elements, after dimensional decomposition, slot filling, quantification and logical verification according to the preset demand template. It includes multiple representation forms such as demand feature vectors, tag system, and constraints. It has the characteristics of unified format and strong executability, and can be directly used as input instructions for search recommendation agents to perform product retrieval and combination optimization.

[0063] Furthermore, the process of semantically mining the natural language requirement through the requirement-understanding agent to obtain supplementary semantic elements includes: The intelligent agent that understands the requirements extracts the core semantics of the natural language requirements to obtain basic semantic elements. The basic semantic elements are fused with the user profile and the historical behavior data to obtain fusion requirement information; The fusion requirement information is inferred and completed using a pre-set common sense base and a pre-set domain knowledge base to obtain full semantic information; The full set of semantic information is structured and validated to obtain the supplementary semantic elements.

[0064] Among them, basic semantic elements refer to the key information units obtained by the demand understanding agent after extracting the core semantics of the user's original natural language demand and removing redundant expressions. They include the user's core intent, key entities and demand backbone. They are a refined expression of the original input and serve as the basic input for subsequent integration of user personalized data.

[0065] Among them, the fusion of demand information refers to the intermediate information representation generated by the demand understanding agent after fusing basic semantic elements with user profiles, historical behavior data and multi-round interaction information obtained from the context management system. This realizes the initial combination of current demand and user personalized characteristics, and provides an information foundation for subsequent knowledge base reasoning and completion.

[0066] Among them, the preset common sense base and the preset domain knowledge base refer to the structured knowledge resource collections pre-built by the system. The common sense base stores general knowledge (such as common sense about life and common matching rules), while the domain knowledge base stores professional knowledge in specific fields (such as outdoor equipment parameters and the efficacy of cosmetic ingredients). They are used to supplement and enhance the reasoning of integrated demand information, thereby improving the depth and accuracy of demand understanding.

[0067] Among them, full semantic information refers to the complete semantic representation generated by the demand understanding agent after calling the common sense base and domain knowledge base to reason and complete the fused demand information. It includes the user's original intent, personalized features, and implicit demands and related information supplemented by knowledge reasoning. It is a product of comprehensive and in-depth understanding of user needs and provides a foundation for subsequent structured processing.

[0068] Furthermore, after the demand understanding agent is invoked, it initiates a multi-turn dialogue based on the user profile and historical behavior in the context management system (e.g., knowing that Zhang San bought a single tent last year): "How many people are you planning to travel with? What is your approximate budget? Where do you plan to camp?" Zhang San answers in turn: "3 people, budget around 5000 yuan, camping in the suburbs." The demand understanding agent integrates the above supplementary information with historical data, calls the common sense base and outdoor domain knowledge base for reasoning and completion, and generates structured demand information containing demand feature vectors (camping, family, beginner), tag system (scenario: outdoor camping; group: family trip; experience: beginner), and constraints (budget ≤ 5000 yuan, 3 people, need tent / sleeping bag / moisture-proof mat, etc.), and stores it in the context management system.

[0069] Furthermore, the generation of structured requirement information based on the supplementary semantic elements includes: Obtain a preset structured requirement template and the dimension system of the structured requirement template. Based on the dimension system of the structured requirement template, decompose the supplementary semantic elements and fill the slots to obtain the filling template information. The filled template information is fused with the user profile and the historical behavior data to obtain fused requirement information. The fused requirement information is then quantified and logically verified to obtain preliminary structured requirement information. The preliminary structured requirement information is standardized to obtain the structured requirement information.

[0070] Among them, the structured requirement template refers to a pre-defined standardized framework used to standardize supplementary semantic elements into a unified format. It defines the organizational structure, field names, and data format of the requirement information, ensuring that each intelligent agent can correctly parse and process the requirement information.

[0071] The dimensional system of the structured requirement template refers to the multi-dimensional information organization framework defined within the structured requirement template, which includes core dimensions such as requirement feature vectors, label system, and constraints, as well as their hierarchical relationships, and is used to guide the decomposition and classification of supplementary semantic elements.

[0072] Among them, filling template information refers to the intermediate information representation obtained after decomposing and filling the slots of the supplementary semantic elements according to the dimensional system of the structured requirement template. At this time, the information has been organized according to the template but has not yet been deeply integrated and verified with user profiles and historical behavior data.

[0073] Among them, the integrated demand information refers to the enhanced demand information obtained by integrating the fill template information with user profiles and historical behavior data. This achieves a preliminary combination of current demand and user personalized characteristics, providing an information foundation for subsequent quantitative processing and logical verification.

[0074] Among them, preliminary structured requirement information refers to requirement information generated after quantitative processing and logical verification of integrated requirement information, but which has not yet undergone final standardization. It already has basic structured characteristics and logical consistency, and will be further standardized to form the final structured requirement information.

[0075] S5. Based on the structured demand information, the search recommendation agent obtains a set of candidate products from the preset product library, and optimizes the combination of the candidate product set to obtain a product combination scheme.

[0076] Understandably, by employing a hybrid strategy that combines vector retrieval and text retrieval through the search recommendation agent, the system can complement each other's advantages in both semantic understanding and precise matching. It can quickly obtain relevant candidate sets from a massive number of products and optimize cross-category combinations by combining user profiles and scenario features, thereby improving the relevance and usability of the recommendation results.

[0077] The pre-built product database refers to a structured product data set that the system builds in advance and updates continuously. It includes product attribute information (such as title, description, price, and inventory), semantic vector representation, scene tags, and relationships between products, providing a unified data access and retrieval foundation for the search and recommendation agent.

[0078] Among them, the candidate product set refers to the relay result set composed of products that match the user's needs, which are initially selected from the preset product library by the search and recommendation agent based on the demand feature vector, tag system and constraints in the structured demand information, and through a hybrid retrieval strategy. It serves as the input object for subsequent combination optimization operations.

[0079] Among them, the product combination scheme refers to the set of product recommendations generated by the search recommendation agent after performing cluster analysis, cross-category matching, and sorting optimization on the candidate product set. These recommendations meet the specific needs of users and usually include multiple schemes with different price ranges or functional focuses (such as economy, comfort, and luxury) to suit the diverse preferences and spending power of users.

[0080] Furthermore, the step of obtaining a set of candidate products from a preset product library using the search and recommendation agent based on the structured demand information includes: Extract the requirement feature vector, label system, and constraints from the structured requirement information; Based on the demand feature vector, the tag system, and the constraints, products corresponding to the structured demand information are obtained from the preset product library to obtain an initial candidate product set; The initial candidate product set is deduplicated and its validity is verified to obtain the candidate product set.

[0081] The demand feature vector, tag system, and constraints are three core retrieval criteria extracted from structured demand information. The demand feature vector maps user demands into a numerical representation in a high-dimensional semantic space and is used for vector similarity retrieval. The tag system is a standardized set of classification tags used to accurately match product categories, scenario attributes, and functional features. The constraints are explicit or implicit filtering restrictions imposed by users, including quantifiable filtering rules such as price range, brand preference, and inventory status. The three together constitute the input parameters of the hybrid retrieval strategy.

[0082] The initial candidate product set refers to the product set initially obtained by the search recommendation agent from the preset product library based on demand feature vectors, label system and constraints. The products in this set have met the basic requirements of semantic relevance, label matching degree and constraints, but have not yet undergone duplicate item removal and validity verification. It serves as an intermediate result set for subsequent fine-tuning.

[0083] Furthermore, the search and recommendation agent reads structured demand information from the context management system, extracting demand feature vectors, tag systems, and constraints. It employs a hybrid strategy combining vector retrieval and text retrieval to recall candidate products such as tents, sleeping bags, and sleeping mats that match the demand from a pre-set product library. After deduplication and validity verification, a set of candidate products is obtained. Subsequently, based on Zhang San's user profile (preference for lightweight items and brand inclination) and the camping scenario, the candidate products are clustered, ranked, and cross-category-based, generating three product combination schemes: economy (entry-level equipment), comfort (branded sets), and luxury (high-end lightweight equipment), which are then written back to the context management system.

[0084] S6. Based on the product explanation intelligent agent, generate personalized explanation content according to the product combination scheme and user profile, and output the product combination scheme and the personalized explanation content to the preset user terminal to complete the e-commerce recommendation.

[0085] Understandably, by having the product explanation agent dynamically adjust the focus and presentation method based on user profiles, the system can highlight the corresponding product value based on the different user's concerns (such as health, cost-effectiveness, brand, etc.), transforming static product parameters into personalized reasons for purchase, thereby enhancing the user's perception and trust in the product and effectively promoting purchase conversion.

[0086] Personalized explanation content refers to the multi-dimensional display and explanation information dynamically generated by the product explanation AI based on product combination schemes and target user profiles. This content can automatically adjust the focus of the explanation based on the user's historical preferences, consumption scenarios, and real-time interaction behavior (e.g., highlighting nutritional value for health-conscious individuals, emphasizing performance parameters for tech enthusiasts, and focusing on packaging and brand for gift-givers), and is presented in various formats such as text, images, and videos, aiming to enhance users' perception and trust in the product, and improve purchase conversion rates.

[0087] Furthermore, the product explanation agent, by acquiring three product combination schemes from the context management system and Zhang San's user profile, generates personalized explanation content for each product in each scheme. For example, for the tent, considering Zhang San's existing single-person tent, it explains: "This family tent has more space than your current single-person tent, suitable for 3 people, and only takes 5 minutes to set up." For the sleeping bag, it emphasizes temperature suitability based on the climate of Beijing's suburbs. Finally, the three product combination schemes and their personalized explanations are output to Zhang San's mobile terminal, presented in text, images, and voice formats for comparison and selection, completing this consultative e-commerce recommendation service.

[0088] This invention acquires user needs, profiles, and historical behaviors, extracts need types, and obtains corresponding intelligent agents. User data is loaded into a context management system to construct a multi-agent system containing scheduling, understanding, recommendation, and explanation agents. The scheduling agent parses the needs and session state to determine whether to invoke the understanding agent. If invoked, the understanding agent performs semantic mining to generate structured needs. The recommendation agent retrieves a candidate set from the product library based on the structured needs and combines and optimizes it to obtain product solutions. The explanation agent generates personalized explanation content based on the solutions and profiles and outputs it to the terminal to complete the recommendation. Therefore, the need-based multi-agent e-commerce recommendation method, system, electronic device, and computer-readable storage medium proposed in this invention, by constructing a consultative e-commerce service architecture of "multi-agent collaboration + context sharing," enables the system to deeply mine real needs from fuzzy user input and dynamically generate scenario-based product combination solutions. This solves the technical problems of traditional e-commerce search processes being simplistic, recommendations lacking scenario awareness, and difficulty in cross-category combinations, thus improving the accuracy of user need understanding in e-commerce recommendations.

[0089] like Figure 2 The diagram shown is a module schematic of the demand-based multi-agent e-commerce recommendation system of the present invention.

[0090] The demand-based multi-agent e-commerce recommendation system 100 described in this invention can be installed in an electronic device. Depending on the functions implemented, the demand-based multi-agent e-commerce recommendation system may include a system construction module 101, a demand analysis module 102, a product optimization module 103, and a product recommendation module 104. The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and are stored in the memory of the electronic device.

[0091] In this embodiment, the functions of each module / unit are as follows: The system construction module 101 is used to acquire the natural language needs, user profiles and historical behavior data of the target user, extract the need types of the natural language needs, and acquire multiple intelligent agents according to the need types; The user profile and the historical behavior data are loaded into a preset context management system to obtain a standard management system. A multi-agent system is constructed based on the standard management system and the multiple agents. The multi-agent system includes a task scheduling agent, a demand understanding agent, a search and recommendation agent, and a product explanation agent. The requirement analysis module 102 is used to obtain the session state data in the multi-agent system, use the task scheduling agent to parse the natural language requirement and the session state data to obtain the parsing result, and determine whether it is necessary to call the requirement understanding agent based on the parsing result. When a call is needed, the intelligent agent that understands the requirements performs semantic mining on the natural language requirements to obtain supplementary semantic elements, and generates structured requirement information based on the supplementary semantic elements. The product optimization module 103 is used to obtain a set of candidate products from a preset product library based on the structured demand information using the search recommendation agent, and to combine and optimize the set of candidate products to obtain a product combination scheme. The product recommendation module 104 is used to generate personalized explanation content based on the product explanation intelligent agent, according to the product combination scheme and user profile, and output the product combination scheme and the personalized explanation content to the preset user terminal to complete the e-commerce recommendation.

[0092] In detail, the modules in the demand-based multi-agent e-commerce recommendation system 100 described in this embodiment of the invention employ the same methods as described above. Figure 1 The aforementioned demand-based multi-agent e-commerce recommendation method employs the same technical means and produces the same technical effects, so it will not be elaborated further here.

[0093] like Figure 3The diagram shown is a structural schematic of an electronic device that implements the demand-based multi-agent e-commerce recommendation method of the present invention.

[0094] The electronic device may include a processor 10, a memory 11, a communication bus 12, and a communication interface 13. It may also include a computer program stored in the memory 11 and capable of running on the processor 10, such as a demand-based multi-agent e-commerce recommendation program.

[0095] In some embodiments, the processor 10 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 10 is the control unit of the electronic device, connecting various components of the entire electronic device through various interfaces and lines. It executes programs or modules stored in the memory 11 (e.g., executing demand-based multi-agent e-commerce recommendation programs) and calls data stored in the memory 11 to perform various functions of the electronic device and process data.

[0096] The memory 11 includes at least one type of readable storage medium, including flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 11 can be an internal storage unit of an electronic device, such as a portable hard drive. In other embodiments, the memory 11 can be an external storage device of the electronic device, such as a plug-in portable hard drive, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, etc. Furthermore, the memory 11 can include both internal and external storage units of the electronic device. The memory 11 can be used not only to store application software and various types of data installed on the electronic device, such as the code of a demand-based multi-agent e-commerce recommendation program, but also to temporarily store data that has been output or will be output.

[0097] The communication bus 12 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be divided into an address bus, a data bus, a control bus, etc. The bus is configured to enable communication between the memory 11 and at least one processor 10, etc.

[0098] The communication interface 13 is used for communication between the aforementioned electronic device and other devices, including a network interface and a user interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a Wi-Fi interface, Bluetooth interface, etc.), typically used to establish communication connections between the electronic device and other electronic devices. The user interface may be a display, an input unit (such as a keyboard), or optionally, a standard wired or wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen, etc. The display may also be appropriately referred to as a screen or display unit, used to display information processed in the electronic device and to display a visual user interface.

[0099] Figure 3 Only electronic devices with components are shown; it will be understood by those skilled in the art that... Figure 3 The structure shown does not constitute a limitation on the electronic device and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0100] For example, although not shown, the electronic device may also include a power supply (such as a battery) to power the various components. Preferably, the power supply can be logically connected to the at least one processor 10 through a power management device, thereby enabling functions such as charging management, discharging management, and power consumption management. The power supply may also include one or more DC or AC power supplies, recharging devices, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be described in detail here.

[0101] It should be understood that the embodiments described are for illustrative purposes only and are not limited to this structure in the scope of the patent application.

[0102] The demand-based multi-agent e-commerce recommendation program stored in the memory 11 of the electronic device is a combination of multiple computer programs. When run in the processor 10, it can achieve the following: Acquire the natural language needs, user profiles, and historical behavior data of the target user; extract the need types of the natural language needs; and acquire multiple intelligent agents based on the need types. The user profile and the historical behavior data are loaded into a preset context management system to obtain a standard management system. A multi-agent system is constructed based on the standard management system and the multiple agents. The multi-agent system includes a task scheduling agent, a demand understanding agent, a search and recommendation agent, and a product explanation agent. Acquire the session state data in the multi-agent system, use the task scheduling agent to parse the natural language requirement and the session state data to obtain the parsing result, and determine whether it is necessary to call the requirement understanding agent based on the parsing result; When a call is needed, the intelligent agent that understands the requirements performs semantic mining on the natural language requirements to obtain supplementary semantic elements, and generates structured requirement information based on the supplementary semantic elements. Based on the structured demand information, the search and recommendation agent obtains a set of candidate products from a preset product library, and optimizes the combination of the candidate product set to obtain a product combination scheme. Based on the product explanation intelligent agent, personalized explanation content is generated according to the product combination scheme and user profile, and the product combination scheme and the personalized explanation content are output to the preset user terminal to complete the e-commerce recommendation.

[0103] Specifically, the processor 10's implementation method of the above-mentioned computer program can be found in [reference needed]. Figure 1 The descriptions of the relevant steps in the corresponding embodiments are not repeated here.

[0104] Furthermore, if the modules / units integrated into the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, or a read-only memory (ROM).

[0105] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor of an electronic device, can perform the following: Acquire the natural language needs, user profiles, and historical behavior data of the target user; extract the need types of the natural language needs; and acquire multiple intelligent agents based on the need types. The user profile and the historical behavior data are loaded into a preset context management system to obtain a standard management system. A multi-agent system is constructed based on the standard management system and the multiple agents. The multi-agent system includes a task scheduling agent, a demand understanding agent, a search and recommendation agent, and a product explanation agent. Acquire the session state data in the multi-agent system, use the task scheduling agent to parse the natural language requirement and the session state data to obtain the parsing result, and determine whether it is necessary to call the requirement understanding agent based on the parsing result; When a call is needed, the intelligent agent that understands the requirements performs semantic mining on the natural language requirements to obtain supplementary semantic elements, and generates structured requirement information based on the supplementary semantic elements. Based on the structured demand information, the search and recommendation agent obtains a set of candidate products from a preset product library, and optimizes the combination of the candidate product set to obtain a product combination scheme. Based on the product explanation intelligent agent, personalized explanation content is generated according to the product combination scheme and user profile, and the product combination scheme and the personalized explanation content are output to the preset user terminal to complete the e-commerce recommendation.

[0106] In the several embodiments provided by this invention, it should be understood that the disclosed devices, systems, and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0107] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0108] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0109] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0110] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within the invention. No appended diagram markings in the claims should be construed as limiting the scope of the claims.

[0111] The blockchain referred to in this invention is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Essentially, a blockchain is a decentralized database, a chain of data blocks linked together using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (anti-counterfeiting) and generate the next block. A blockchain can include an underlying blockchain platform, a platform product service layer, and an application service layer.

[0112] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0113] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or systems stated in a system claim may also be implemented by a single unit or system through software or hardware. The term "second class" is used to indicate names and does not indicate any specific order.

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A demand-based multi-agent e-commerce recommendation method, characterized in that, The method includes: Acquire the natural language needs, user profiles, and historical behavior data of the target user; extract the need types of the natural language needs; and acquire multiple intelligent agents based on the need types. The user profile and the historical behavior data are loaded into a preset context management system to obtain a standard management system. A multi-agent system is constructed based on the standard management system and the multiple agents. The multi-agent system includes a task scheduling agent, a demand understanding agent, a search and recommendation agent, and a product explanation agent. Acquire the session state data in the multi-agent system, use the task scheduling agent to parse the natural language requirement and the session state data to obtain the parsing result, and determine whether it is necessary to call the requirement understanding agent based on the parsing result; When a call is needed, the intelligent agent that understands the requirements performs semantic mining on the natural language requirements to obtain supplementary semantic elements, and generates structured requirement information based on the supplementary semantic elements. Based on the structured demand information, the search and recommendation agent obtains a set of candidate products from a preset product library, and optimizes the combination of the candidate product set to obtain a product combination scheme. Based on the product explanation intelligent agent, personalized explanation content is generated according to the product combination scheme and user profile, and the product combination scheme and the personalized explanation content are output to the preset user terminal to complete the e-commerce recommendation.

2. The demand-based multi-agent e-commerce recommendation method of claim 1, wherein, The step of extracting the demand type of the natural language requirement and obtaining multiple agents based on the demand type includes: The natural language requirements are semantically understood using a pre-defined large language model to obtain semantic understanding results; The semantic understanding results are mapped to a preset scene classification system to obtain the demand type corresponding to the natural language demand; Obtain the scene tags, function tags, and domain knowledge tags associated with the corresponding demand type for the natural language demand, and select the corresponding intelligent agents from the preset intelligent agent library according to the scene tags, function tags, and domain knowledge tags to obtain the multiple intelligent agents.

3. The demand-based multi-agent e-commerce recommendation method of claim 1, wherein, The step of loading the user profile and the historical behavior data into a preset context management system to obtain a standard management system includes: The user profile and the historical behavior data are standardized to obtain standardized user data. The preset context management system is initialized and configured to obtain the initial context management system, and the standardized user data is written into the initial context management system to obtain the standard management system.

4. The demand-based multi-agent e-commerce recommendation method of claim 1, wherein, The step of using the task scheduling agent to parse the natural language requirements and the session state data to obtain the parsing results includes: Obtain the dialogue state parsing model in the task scheduling agent, and standardize the natural language requirements and the conversation state data to obtain standard language requirements and standard conversation state data. The dialogue state parsing model is used to perform semantic parsing on the standard language requirements to obtain semantic parsing results, and the dialogue state parsing model is used to perform context parsing on the standard conversation state data to obtain context parsing results. The semantic parsing result and the context parsing result are fused and verified to determine the consistency and information completeness of the natural language requirement and the session state data, and the verification result is obtained. Based on the verification results, the semantic parsing results and the context parsing results are structurally integrated and validated to obtain the parsing results.

5. The demand-based multi-agent e-commerce recommendation method of claim 1, wherein, The process of semantically mining the natural language requirement through the requirement-understanding agent to obtain supplementary semantic elements includes: The intelligent agent that understands the requirements extracts the core semantics of the natural language requirements to obtain basic semantic elements. The basic semantic elements are fused with the user profile and the historical behavior data to obtain fusion requirement information; The fusion requirement information is inferred and completed using a pre-set common sense base and a pre-set domain knowledge base to obtain full semantic information; The full set of semantic information is structured and validated to obtain the supplementary semantic elements.

6. The demand-based multi-agent e-commerce recommendation method of claim 1, wherein, The generation of structured requirement information based on the supplementary semantic elements includes: Obtain a preset structured requirement template and the dimension system of the structured requirement template. Based on the dimension system of the structured requirement template, decompose the supplementary semantic elements and fill the slots to obtain the filling template information. The filled template information is fused with the user profile and the historical behavior data to obtain fused requirement information. The fused requirement information is then quantified and logically verified to obtain preliminary structured requirement information. The preliminary structured requirement information is standardized to obtain the structured requirement information.

7. The demand-based multi-agent e-commerce recommendation method of claim 1, wherein, The step of obtaining a set of candidate products from a preset product library based on the structured demand information using the search recommendation agent includes: Extract the requirement feature vector, label system, and constraints from the structured requirement information; Based on the demand feature vector, the tag system, and the constraints, products corresponding to the structured demand information are obtained from the preset product library to obtain an initial candidate product set; The initial candidate product set is deduplicated and its validity is verified to obtain the candidate product set.

8. A demand-based multi-agent e-commerce recommendation system, characterized in that, The system includes: The system construction module is used to acquire the natural language needs, user profiles and historical behavior data of the target user, extract the need types of the natural language needs, and acquire multiple intelligent agents based on the need types; The user profile and the historical behavior data are loaded into a preset context management system to obtain a standard management system. A multi-agent system is constructed based on the standard management system and the multiple agents. The multi-agent system includes a task scheduling agent, a demand understanding agent, a search and recommendation agent, and a product explanation agent. The requirement analysis module is used to acquire the session state data in the multi-agent system, use the task scheduling agent to parse the natural language requirement and the session state data to obtain the parsing result, and determine whether it is necessary to call the requirement understanding agent based on the parsing result. When a call is needed, the intelligent agent that understands the requirements performs semantic mining on the natural language requirements to obtain supplementary semantic elements, and generates structured requirement information based on the supplementary semantic elements. The product optimization module is used to obtain a set of candidate products from a preset product library based on the structured demand information using the search recommendation agent, and to combine and optimize the set of candidate products to obtain a product combination scheme. The product recommendation module is used to generate personalized explanation content based on the product explanation intelligent agent, the product combination scheme and user profile, and output the product combination scheme and the personalized explanation content to the preset user terminal to complete the e-commerce recommendation.

9. An electronic device, comprising: The electronic device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the demand-based multi-agent e-commerce recommendation method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It includes a data storage area and a program storage area. The data storage area stores the created data, and the program storage area stores the computer program. When the computer program is executed by the processor, it implements the demand-based multi-agent e-commerce recommendation method as described in any one of claims 1 to 7.

Images