An online intelligent shopping guide method, device and medium based on a large model

Abstract

The application discloses an online intelligent shopping guide method and device based on a large model, and a medium, and belongs to the technical field of the cross of artificial intelligence and e-commerce services, and is used for solving the technical problem that the existing online shopping guide mode is dependent on fixed dialogue templates, it is difficult to accurately find products meeting the own needs, and the shopping experience of consumers is poor. The method comprises the following steps: performing multi-modal demand sensing and analysis processing on shopping guide commodity demand information to obtain real-time demand context information; constructing and correlating commodity information in an e-commerce platform with a knowledge graph to obtain a commodity knowledge graph; constructing and iterating a user portrait label system for consumers; performing personalized recommendation and information interaction processing on the shopping guide commodity demand information of the consumers based on the real-time demand context information, the commodity knowledge graph and real-time user portrait information; and performing dynamic optimization processing on the commodity personalized recommendation result to obtain an optimized commodity personalized recommendation strategy.

Description

Technical Field

[0001] This application relates to the intersection of artificial intelligence and e-commerce services, and in particular to an online intelligent shopping guide method, device and medium based on a large model. Background Technology

[0002] Against the backdrop of the booming e-commerce industry, traditional online shopping guide models face numerous pain points. Consumers struggle to quickly and accurately find products that meet their needs amidst a vast array of goods, often falling into a decision-making dilemma due to information overload. Shopping guide services rely heavily on fixed script templates, failing to flexibly adjust recommendation strategies and communication methods based on consumers' personalized needs, preferences, and scenarios. Furthermore, the lack of in-depth exploration of consumers' potential needs during the shopping process hinders the provision of proactive and forward-looking shopping advice, resulting in poor consumer shopping experiences and low conversion rates for merchants. Summary of the Invention

[0003] This application provides an online intelligent shopping guide method, device, and medium based on a large model to solve the following technical problem: existing online shopping guide models mostly rely on fixed script templates, making it difficult to accurately find products that meet one's own needs, resulting in a poor shopping experience for consumers.

[0004] The embodiments of this application adopt the following technical solutions: On the one hand, embodiments of this application provide an online intelligent shopping guide method based on a large model, including: performing multimodal demand perception and parsing processing on shopping guide product demand information to obtain real-time demand context information; constructing and associating product information in the e-commerce platform into a knowledge graph to obtain a product knowledge graph; constructing and iterating a user profile tag system for consumers based on their historical consumption record data to obtain real-time user profile information; performing personalized recommendation and information interaction processing on the consumer's shopping guide product demand information based on the real-time demand context information, the product knowledge graph, and the real-time user profile information to obtain personalized product recommendation results; and dynamically optimizing the personalized product recommendation results through user shopping guide feedback data to obtain an optimized personalized product recommendation strategy.

[0005] This application's embodiments utilize a large language model for deep semantic analysis of consumers' multimodal needs, compressing the traditional process of repeatedly asking questions manually to the second level of demand insight. This significantly improves demand identification efficiency and dramatically increases the accuracy of demand capture. It also identifies potential related needs, enabling businesses to significantly reduce customer churn due to inaccurate recommendations and improve user decision-making efficiency. Furthermore, the intelligent interaction engine can generate reassuring messages, differentiated advantage interpretations, or incentive programs in real time, increasing user purchase intention conversion rates and significantly improving order conversion rates. By integrating knowledge graphs and user profiles, the system can identify the preferences of high-potential customers and optimal product combinations, providing data support for product selection and promotions. Intelligent add-to-cart and fulfillment tracking functions improve order completion rates and reduce sales losses due to abandoned purchases.

[0006] In one feasible implementation, multimodal demand perception and parsing processing is performed on the shopping guide product demand information to obtain real-time demand context information. Specifically, this includes: using a multimodal information acquisition submodule to collect and process the shopping guide product demand information under multi-dimensional interactive modes to obtain multimodal information; wherein the multi-dimensional interactive modes include: text information, voice information, and image information; using a large-model semantic understanding submodule to extract key information related to consumer demand types from the text information in the multimodal information; performing emotion recognition and analysis processing on the voice information in the multimodal information; identifying key product attributes from the image information in the multimodal information; obtaining multimodal demand element information based on hierarchical processing of the multimodal information; and constructing a demand element tagging system based on the multimodal demand element information and through a demand element standardization submodule; wherein the demand element tagging system includes at least: product category, core function, budget range, user group, usage scenario, preferred brand, and quality requirements; and using the demand element tagging system to perform contextual recognition of key information in the real-time shopping guide product demand information to obtain the real-time demand context information.

[0007] In one feasible implementation, a product knowledge graph is constructed and associated from product information on an e-commerce platform. Specifically, this includes: a product information collection and structuring submodule, which collects and processes multi-dimensional data from the e-commerce platform's product data sources to obtain multi-dimensional product data; wherein the multi-dimensional product data includes: basic product information, functional attributes, relationships, and reputation data; a knowledge graph construction and association submodule, which performs hierarchical association processing on the attribute data in the multi-dimensional product data and constructs an original associated product knowledge graph; and a knowledge graph dynamic update submodule, which updates the product data between corresponding nodes and relationship attributes in the original associated product knowledge graph to obtain the final product knowledge graph.

[0008] In one feasible implementation, based on consumers' historical consumption record data, a user profile tagging system is constructed and iterated to obtain real-time user profile information. Specifically, this includes: collecting and integrating the consumer's historical multi-dimensional consumption data through a user data collection and integration submodule to obtain historical consumption record data; wherein the historical consumption record data includes: historical shopping records, interaction records, basic information, and real-time demand elements; establishing a multi-dimensional user profile tagging system through a user profile tagging system construction submodule; wherein the multi-dimensional user profile tagging system includes: static tags, dynamic tags, behavioral tags, and evaluation tags; classifying the historical consumption record data by attribute tags through the multi-dimensional user profile tagging system to obtain user-product interaction profile features; and iteratively updating the user-product interaction profile features in real-time based on the consumer's latest needs and preferences under new user interaction behaviors through a real-time user profile real-time iteration submodule to generate the real-time user profile information.

[0009] In one feasible implementation, based on the real-time demand context information, the product knowledge graph, and the real-time user profile information, personalized recommendation processing is performed on the consumer's product demand information to obtain personalized product recommendation results. Specifically, this includes: using a personalized product recommendation submodule, and based on the real-time demand context information and the real-time user profile information, filtering the product knowledge graph to obtain a set of candidate products that best matches; and intelligently sorting the candidate products in the set of candidate products based on comprehensive multi-dimensional product features to obtain a list of recommended products; using an intelligent product comparison submodule, comparing the multi-dimensional association features of each product in the list of recommended products to obtain product association feature comparison information; wherein, the multi-dimensional association features include: core parameters, functional characteristics, price, user ratings, and after-sales service policies; and using the product recommendation list information and the product association feature comparison information, personalized recommendation processing is performed on the consumer's product demand information to obtain personalized product recommendation results.

[0010] In one feasible implementation, the consumer's shopping guide product demand information is processed through information interaction to obtain personalized product recommendation results. This includes: adding corresponding products from the personalized product recommendation results to the shopping cart under multiple instructions via a contextualized shopping cart addition submodule to obtain shopping cart product information based on the user's e-commerce platform account; querying the order integration information corresponding to the shopping guide product demand information under user intent parsing via an integrated order query and management submodule to obtain order status information; and associating both the shopping cart product information and the order status information with the personalized product recommendation results.

[0011] In one feasible implementation, the process of interacting with the consumer's product demand information to obtain personalized product recommendation results further includes: using a dynamic interaction strategy submodule to perform progressive hierarchical demand parsing processing on the product demand information using a relevant language analysis model to obtain real-time user demand intent information; using a potential demand mining submodule to mine potential user demand information from the product demand information, and generating potential product recommendation information based on the mined potential demand intent information; and associating both the real-time user demand intent information and the potential product recommendation information with the personalized product recommendation results.

[0012] In one feasible implementation, the personalized product recommendation results are dynamically optimized using user feedback data to obtain an optimized personalized product recommendation strategy. Specifically, this includes: collecting multi-dimensional feedback data from consumers through a feedback data collection and analysis submodule, identifying problem information in the multi-dimensional feedback data, and obtaining feedback data analysis results; wherein the multi-dimensional feedback data includes: direct consumer feedback, behavioral feedback, and transaction result data; and using a model and strategy optimization submodule, incrementally optimizing the feedback optimization features from the feedback data analysis results into the personalized product recommendation results, and adaptively adjusting the weights of the recommendation algorithm and matching rules for the personalized product recommendation results to obtain the optimized personalized product recommendation strategy.

[0013] Secondly, embodiments of this application also provide an online intelligent shopping guide device based on a large model, the device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, enabling the at least one processor to execute an online intelligent shopping guide method based on a large model as described in any of the above embodiments.

[0014] Thirdly, embodiments of this application also provide a non-volatile computer storage medium, wherein the storage medium is a non-volatile computer-readable storage medium, and the non-volatile computer-readable storage medium stores at least one program, each program including instructions, wherein when the instructions are executed by a terminal, the terminal executes an online intelligent shopping guide method based on a large model as described in any of the above embodiments.

[0015] This application provides an online intelligent shopping guide method, device, and medium based on a large model. Compared with the prior art, the embodiments of this application have the following beneficial technical effects: 1. By leveraging a large language model for deep semantic analysis of consumers' multimodal needs, the traditional process of repeatedly asking questions manually to gain insights into needs is compressed to the second level, significantly improving the efficiency of need identification and the accuracy of need capture. It also identifies potential related needs, enabling businesses to significantly reduce customer churn caused by inaccurate recommendations and improve user decision-making efficiency.

[0016] 2. When the system detects user hesitation, price comparison, or negative emotions, the intelligent interaction engine can generate reassuring messages, explanations of differentiated advantages, or preferential incentive plans in real time, thereby increasing the conversion rate of user purchase intention and significantly improving the order conversion rate.

[0017] 3. By integrating knowledge graphs and user profiles, the system can identify the preferences of high-potential customers and the optimal product combinations, providing data support for product selection and promotions. Intelligent add-to-cart and fulfillment tracking functions improve order completion rates and reduce sales losses due to abandoned purchases. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 A flowchart illustrating an online intelligent shopping guide method based on a large model, provided for embodiments of this application; Figure 2 This is a schematic diagram of the structure of an online intelligent shopping guide device based on a large model, provided as an embodiment of this application. Detailed Implementation

[0019] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application.

[0020] It should be noted that the embodiments of this application provide an online intelligent shopping guide system based on a large model, including: a multimodal demand perception and analysis module, a product knowledge graph construction and management module, a user profile construction and iteration module, an intelligent recommendation and interaction module, and a dynamic optimization and feedback module.

[0021] The online intelligent shopping guide system based on a large model adopts a microservice architecture. Each module achieves loosely coupled communication and asynchronous data processing through message queues and API calls. The multimodal demand perception module serves as the front-end interaction interface, receiving and processing user input in real time. The product knowledge graph and user profile module acts as the core data platform, providing real-time data support for recommendations and interactions. The intelligent recommendation and interaction module serves as the business platform, encapsulating various shopping guide capabilities and providing them externally through a unified service gateway. The dynamic optimization and feedback module acts as the back-end optimization engine, enabling continuous model iteration and strategy updates. The system relies on a distributed computing and storage framework, supporting horizontal scaling and high-availability deployment to meet the high-concurrency, low-latency online shopping guide requirements of e-commerce platforms.

[0022] Meanwhile, the online intelligent shopping guide system based on a large model integrates large language models, knowledge graphs, multimodal data processing, and user profiling technologies. By constructing a full-process intelligent shopping guide framework, it achieves end-to-end intelligent services from consumer demand perception, intelligent analysis, accurate product recommendations to dynamic optimization of shopping guide strategies. Core technologies include: deep semantic understanding capabilities based on large language models fine-tuned in the e-commerce field, supporting accurate parsing of consumer multimodal demand information; product knowledge graph construction and dynamic updating technology, enabling structured management of product attributes, relationships, and reputation data; real-time iteration of user profiles and personalized matching algorithms, ensuring the accuracy and timeliness of recommended content; and a model dynamic optimization mechanism based on feedback data, ensuring continuous evolution of shopping guide capabilities. It can be widely applied to various online e-commerce platforms, brand official stores, and other scenarios, addressing pain points such as low efficiency, insufficient personalization, and difficulty in uncovering potential needs in traditional shopping guides, improving the quality of shopping guide services and commercial conversion results through intelligent technologies.

[0023] This application provides an online intelligent shopping guide method based on a large model, such as... Figure 1 As shown, the online intelligent shopping guide method based on a large model specifically includes steps S101-S105: S101. Perform multimodal demand perception and parsing processing on the product demand information of the shopping guide to obtain real-time demand context information.

[0024] Specifically, the multimodal information acquisition submodule collects and processes product demand information through multi-dimensional interactive methods to obtain multimodal information. These multi-dimensional interactive methods include text, voice, and image information.

[0025] As a feasible implementation method, the multimodal information acquisition submodule supports information acquisition through various interactive methods such as text, voice, and images. For text information, it connects to the e-commerce platform's chat window via API to acquire consumer input text in real time. For voice information, it integrates a speech recognition engine to convert consumer speech content into text data while preserving emotional characteristics. For image information, it utilizes image recognition technology to extract product features, scene elements, and other information from images, such as images of desired products uploaded by consumers or photos of usage scenarios. During the acquisition process, the data is cleaned in real time to remove noise and ensure data quality.

[0026] Furthermore, through the large-scale model semantic understanding submodule, key information related to consumer demand types is extracted from the text information in the multimodal information. Emotion recognition and analysis are performed on the speech information in the multimodal information; key product attributes are identified from the image information in the multimodal information; and multimodal demand element information is obtained based on the hierarchical processing of the multimodal information.

[0027] As a feasible implementation method, the large-scale model semantic understanding submodule employs a large language model fine-tuned for e-commerce applications to perform deep semantic understanding on the collected multimodal information. For textual information, it automatically identifies consumer demand types (clearly stating purchase needs, inquiring about product features, comparing similar products, etc.) and extracts key information. For example, for "wanting a phone suitable for students, with a budget under 2000 yuan and good camera performance," it can accurately extract elements such as "user group: students," "budget: under 2000 yuan," "core need: good camera performance," and "product category: mobile phone." For speech-to-text conversion and emotional features, it combines the large-scale model's sentiment analysis capabilities to determine the consumer's emotional state and adjust subsequent communication tone and recommendation strategies accordingly. For image parsing results, it associates with a product knowledge graph to identify the product category, brand, and key attributes in the image, understanding the consumer's potential preferences.

[0028] Furthermore, based on multimodal demand element information and through a demand element standardization sub-module, a demand element tagging system is constructed. This demand element tagging system includes at least: product category, core functions, budget range, user group, usage scenario, preferred brands, and quality requirements.

[0029] Furthermore, through the demand element tagging system, the context of key information in the real-time shopping guide product demand information is identified to obtain real-time demand context information.

[0030] As a feasible implementation method, the demand element standardization submodule transforms the unstructured demand elements parsed from the large model into a standardized format, establishing a demand element tagging system, including dimensions such as product category, core functions, budget range, user group, usage scenario, preferred brand, and quality requirements. It supplements and refines ambiguous demands; for example, when a consumer only states "I want to buy a good laptop," it proactively asks questions (such as "Will you mainly use the laptop for office work, study, or gaming? What is your approximate budget range?") to obtain key information, forming a complete and standardized set of demand elements. This provides a unified data foundation for subsequent accurate recommendations, ultimately completing the contextual recognition of key information to obtain real-time demand context information.

[0031] S102. Construct and associate product information from e-commerce platforms into a knowledge graph to obtain a product knowledge graph.

[0032] Specifically, the product information collection and structuring submodule collects and processes multi-dimensional data from e-commerce platform product data sources to obtain multi-dimensional product data. This multi-dimensional product data includes: basic product information, functional attributes, relationships, and customer reputation data.

[0033] As a feasible implementation method, the product information collection and structuring submodule connects to e-commerce platform product databases, brand official data sources, and third-party product review platforms to collect basic product information, functional attributes, relationships, and reputation data. Utilizing information extraction technology (based on a combination of pre-trained models and rules), unstructured product description text and user reviews are transformed into structured data. For example, attributes such as "processor model: Snapdragon 8 Gen2," "screen size: 6.7 inches," and "battery capacity: 5000mAh" are extracted from mobile phone product descriptions. Positive tags such as "clear camera," "long battery life," and "smooth system" and negative tags such as "severe overheating" and "insufficient battery life" are extracted from user reviews.

[0034] Furthermore, through the knowledge graph construction and association submodule, the attribute data in the multidimensional data of products are hierarchically associated, and the original associated product knowledge graph is constructed.

[0035] As a feasible implementation method, the knowledge graph construction and association submodule uses a graph database to construct a product knowledge graph, defining entities such as "product," "attribute," "category," "user reviews," and "brand," as well as relationships such as "has attribute," "belongs to category," "related product," and "user reviews." For example, "product-phone A" establishes an "has attribute" relationship with "attribute-processor model," with the attribute value being "Snapdragon 8Gen2"; "product-phone A" establishes a "complementary product" relationship with "product-earphone B." Simultaneously, it integrates a product classification system to achieve hierarchical associations between different product categories, such as "phone" belonging to the "digital products" category, and "digital products" belonging to the "electronic products" category, providing support for cross-category recommendations and product association analysis.

[0036] Furthermore, through the knowledge graph dynamic update submodule, the original related product knowledge graph is updated with product data between corresponding nodes and relational attributes to obtain the product knowledge graph.

[0037] As a feasible implementation method, the knowledge graph dynamic update submodule establishes a real-time update mechanism, connecting to the e-commerce platform's product data update interface. When information such as product price, inventory, specifications, and reviews changes, the corresponding nodes and relationship attributes of the knowledge graph are updated synchronously in a timely manner. For example, when product inventory is insufficient, the "inventory status" attribute of the "product" entity is updated; when a large number of negative reviews are added, the "review tag" and "rating" attributes of the relationship between "product" and "user reviews" are updated. Simultaneously, the completeness and consistency of the knowledge graph are periodically verified, and data conflicts are detected and corrected through rule-based reasoning to ensure the accuracy and timeliness of the knowledge graph data.

[0038] S103. Based on consumers' historical consumption records, construct and iterate on the relevant user profile tag system to obtain real-time user profile information.

[0039] Specifically, the user data collection and integration submodule collects and integrates consumers' historical multi-dimensional consumption data to obtain historical consumption record data. This historical consumption record data includes: historical shopping records, interaction records, basic information, and real-time demand elements.

[0040] As a feasible implementation method, the user data collection and integration submodule collects multi-dimensional consumer data, including historical shopping records, interaction records, basic information, and real-time demand elements. Through data fusion technology, user data scattered across different systems is integrated under a unified user identifier, solving the "data silo" problem and forming a complete user data set.

[0041] Furthermore, a multi-dimensional user profile tagging system is established through a user profile tagging system construction sub-module. This multi-dimensional user profile tagging system includes: static tags, dynamic tags, behavioral tags, and evaluation tags.

[0042] As a feasible implementation method, the user profile tagging system construction submodule establishes a multi-dimensional user profile tagging system, including static tags, dynamic tags, behavioral tags, and evaluation tags. Machine learning algorithms are used to analyze user data and automatically generate tags. For example, by analyzing a user's historical purchase records, it is determined that the user has high loyalty to the "Huawei" brand, generating the tag "Brand Preference: Huawei"; through real-time demand analysis, it is determined that the user currently has a strong purchase intention, generating the tag "Purchase Intention Strength: High".

[0043] Furthermore, through a multi-dimensional user profile tagging system, historical consumption record data is categorized by attribute tags to obtain user product interaction profile features.

[0044] Furthermore, through the real-time iteration submodule of user profiles, the user product interaction profile features are iteratively updated in real time based on the latest consumer needs and preferences under the new user interaction behaviors, and real-time user profile information is generated.

[0045] As a feasible implementation method, the real-time iteration submodule of user profiling, combined with a real-time iteration mechanism, updates user profile tags promptly when users generate new interactive behaviors (inquiring about new products, browsing product details, submitting purchase orders) or raise new needs. For example, if a user initially inquired about mobile phones but later asked about tablets, the "Recent Needs" tag is updated to "Mobile Phones, Tablets"; after a user purchases a laptop from a certain brand, the weight of the "Brand Loyalty" tag for that brand is increased. Simultaneously, based on a time decay algorithm, tags that haven't been updated for a long time are either downgraded or rendered invalid, ensuring that user profiles always reflect the latest consumer needs and preferences.

[0046] S104. Based on real-time demand context information, product knowledge graph, and real-time user profile information, personalized recommendations and information interaction processing are performed on consumers' product demand information to obtain personalized product recommendation results.

[0047] Specifically, the personalized product recommendation submodule filters candidate product sets from the product knowledge graph based on real-time demand context information and real-time user profile information to obtain the most matching candidate product set. Then, based on comprehensive multi-dimensional product features, the candidate products in the most matching candidate product set are intelligently sorted to obtain a list of recommended products.

[0048] As a feasible implementation method, the personalized product recommendation submodule can deeply integrate collaborative filtering, content matching, and knowledge graph reasoning algorithms to generate a highly personalized product recommendation list. The system first filters the most suitable candidate product set from the product knowledge graph based on the user's real-time needs and historical profile (preferred brands, price sensitivity, functional preferences, etc.). Then, leveraging the ranking learning capabilities of a large model, it intelligently ranks the candidate products based on multiple factors such as product popularity, user reviews, inventory status, and promotional information, generating the final recommendation result. The recommendation result is not only presented in list form but can also automatically generate recommendation reasons (such as "Based on your budget and photography needs, we recommend this high-performance mobile phone, whose camera score surpasses 90% of products in the same price range") through Natural Language Generation (NLG) technology, enhancing the persuasiveness of the recommendation.

[0049] Furthermore, the intelligent product comparison submodule compares the multi-dimensional correlation features of each product in the product recommendation list to obtain product correlation feature comparison information. These multi-dimensional correlation features include: core parameters, functional characteristics, price, user ratings, and after-sales service policies.

[0050] Furthermore, by using the product recommendation list information and product association feature comparison information, personalized recommendations are made based on the consumer's product needs, resulting in personalized product recommendation results.

[0051] As a feasible implementation method, the intelligent product comparison submodule allows users to initiate comparison requests for multiple products in the recommendation list. Based on a product knowledge graph, the system automatically extracts structured information for each product across dimensions such as core parameters, functional characteristics, price, user ratings, and after-sales service policies, generating clear and intuitive comparison tables (or natural language descriptions highlighting differences). Users can customize the comparison dimensions and product range via voice or text commands (such as "Compare the batteries and screens of A and B"). This module deeply integrates large model capabilities, not only listing data but also interpreting the actual experience impact behind data differences, assisting users in making more informed decisions and obtaining personalized product recommendations.

[0052] Furthermore, through the contextualized shopping cart sub-module, corresponding products from the personalized product recommendation results are added to the shopping cart under multiple instructions, resulting in shopping cart product information based on the user's e-commerce platform account.

[0053] As a feasible implementation method, a contextualized "add to cart" submodule is used to provide a seamless "add to cart" experience during the recommendation and comparison process. The system is deeply integrated with the e-commerce platform's shopping cart system via API. Users can add their desired items directly to their e-commerce platform account's shopping cart via voice commands, clicking buttons on the interactive interface, or directly replying "add to cart." After successful addition, the assistant will provide voice or text confirmation ("Item XX has been successfully added to your shopping cart") and can further intelligently ask the user whether they want to continue browsing similar products, whether they need to combine items for discounts, or whether they want to checkout immediately, proactively driving the transaction process.

[0054] Furthermore, through the integrated order query and management submodule, the order information corresponding to the shopping cart product demand information is processed under user intent parsing to obtain order status information. Then, the shopping cart product information and order status information are both associated with the personalized product recommendation results.

[0055] As a feasible implementation method, the integrated order query and management submodule can provide full-process services. This assistant integrates order query functionality. Users can query order status, logistics information, return and exchange progress, etc., through natural language (such as "Has my order from yesterday been shipped?" or "Check the logistics of the coat I bought"). The system is authorized to access the e-commerce platform's order system, uses a large model to analyze the user's query intent, accurately locates relevant orders, and returns structured information in a clear and user-friendly format (such as order number, product list, payment amount, logistics company, latest logistics trajectory, estimated delivery time, etc.). For abnormal statuses (such as logistics delays, out-of-stock items), the assistant will proactively explain and reassure users, and provide suggestions for subsequent operations (such as "Your order has been shipped, but the logistics information shows that the package is being sorted at the transit station, and the trajectory is expected to be updated tomorrow. If you need it urgently, I can contact customer service for priority processing").

[0056] Furthermore, through the dynamic interaction strategy submodule, the product demand information of the shopping guide is processed by progressive hierarchical demand parsing using a language analysis model to obtain real-time user demand intent information.

[0057] As a feasible implementation method, a dynamic interaction strategy submodule, leveraging the real-time intent recognition capabilities of a large model, dynamically adjusts communication strategies with consumers. In the initial stages of interaction, guided communication is employed, using concise and friendly questions to elicit user needs. Once the user's needs are clear, the approach shifts to a professional answer mode, detailing the recommended product's functions, advantages, and usage scenarios, comparing it with similar products, and providing objective purchase advice. When user hesitation is detected, a reassurance and persuasion mode is activated, emphasizing the product's core value, promotional activities, and after-sales guarantees to alleviate user concerns. Simultaneously, multi-turn dialogue context understanding is supported, allowing for coherent communication based on historical dialogue content, avoiding repetitive questioning, and improving interaction fluency.

[0058] Furthermore, the potential demand mining submodule processes the user's potential demand information from the product demand information, and generates potential product recommendations based on the mined potential demand intent information. Then, both the real-time user demand intent information and the potential product recommendation information are linked to the personalized product recommendation results.

[0059] As a feasible implementation method, a potential demand mining submodule is used, leveraging the deep semantic understanding and reasoning capabilities of a large model to proactively uncover consumers' latent needs. By analyzing implicit information in user statements, such as when a user buys baby formula, it can be inferred that they may need related products like baby food and bottles; when a user buys a laptop for design, it can be inferred that they may need supporting resources such as a high-performance graphics card and professional design software. At appropriate times, relevant products or services are proactively recommended to users, with explanations provided, such as, "After buying baby formula, would you also like to know about suitable complementary foods for your baby? Many mothers have reported that this complementary food works very well with the formula you chose," helping users discover unexpressed needs and improving the completeness and satisfaction of their shopping experience.

[0060] S105. Based on user feedback data, dynamically optimize the personalized product recommendation results to obtain an optimized personalized product recommendation strategy.

[0061] Specifically, the feedback data collection and analysis submodule collects multi-dimensional feedback data from consumers, identifies problem information within this data, and obtains the feedback data analysis results. This multi-dimensional feedback data includes: direct consumer feedback, behavioral feedback, and transaction result data.

[0062] As a feasible implementation method, the feedback data collection and analysis submodule can collect multi-dimensional feedback data, including direct consumer feedback (satisfaction ratings for recommended products, evaluations of shopping guide services, and problem feedback), behavioral feedback (clicks, favorites, purchases, and abandonments of recommended products), and transaction result data (sold products, transaction amount, conversion rate, repurchase rate, etc.). Multi-dimensional analysis of the feedback data identifies problems in the shopping guide process, such as low matching between recommended products and user needs, unpopular communication methods, and insufficient uncovering of potential needs. Simultaneously, it analyzes the common characteristics of successful shopping guide cases, such as the types of products recommended with high conversion rates and effective communication techniques.

[0063] Furthermore, through the model and strategy optimization submodule, the incremental feedback optimization features from the feedback data analysis results are incorporated into the personalized product recommendation results. The recommendation algorithm and matching rules of the personalized product recommendation results are then adaptively adjusted with weights to obtain the optimized personalized product recommendation strategy.

[0064] As a feasible implementation method, the model and strategy optimization submodule continuously optimizes the large model and recommendation strategy based on feedback data analysis results. For the large model, incremental learning technology is used to integrate new feedback data and excellent case data into model training, improving the model's ability to understand consumer needs, the accuracy of intent recognition, and the quality of communication script generation. For the recommendation strategy, algorithm weights and matching rules are adjusted. For example, to address situations where recommended products have low click-through rates but high conversion rates, the product display order is optimized; to address user feedback regarding "severe homogenization of recommended products," the weight of product diversity recommendations is increased. Simultaneously, an A / B testing mechanism is introduced to verify the effectiveness of the optimized model and strategy, and the optimal solution is selected for official deployment.

[0065] As a feasible implementation method, the industry data fusion and optimization submodule can be utilized to connect with external data such as e-commerce industry trend data, popular product lists, and consumer behavior reports, integrating industry dynamics into the optimization of shopping guide strategies. For example, when a certain type of product (such as summer sun protection products) enters its peak sales season, the recommendation priority can be adjusted to proactively recommend it to potential users; when new consumption trends emerge in the industry, the product knowledge graph tags and user profile analysis dimensions can be updated to ensure that the shopping guide assistant keeps up with industry development and provides recommendations and services that conform to market trends.

[0066] In addition, embodiments of this application also provide an online intelligent shopping guide device based on a large model, such as... Figure 2 As shown, the online intelligent shopping guide device 200 based on the large model specifically includes: At least one processor 201; and a memory 202 communicatively connected to the at least one processor 201; wherein the memory 202 stores instructions executable by the at least one processor 201 to enable the at least one processor 201 to execute: Multimodal demand perception and parsing processing is performed on the product demand information of the shopping guide to obtain real-time demand context information; By constructing and associating product information from e-commerce platforms into a knowledge graph, a product knowledge graph can be obtained. Based on consumers’ historical consumption records, we construct and iterate on a user profile tagging system to obtain real-time user profile information. Based on real-time demand context information, product knowledge graph, and real-time user profile information, personalized recommendations and information interaction processing are performed on consumers' shopping guide product demand information to obtain personalized product recommendation results. By using user feedback data, the personalized product recommendation results are dynamically optimized to obtain an optimized personalized product recommendation strategy.

[0067] This application utilizes a large language model for deep semantic analysis of consumers' multimodal needs, compressing the traditional process of repeatedly asking questions manually to the second level of demand insight. This significantly improves demand identification efficiency and dramatically increases the accuracy of demand capture. It also identifies potential related needs, enabling businesses to significantly reduce customer churn due to inaccurate recommendations and improve user decision-making efficiency. Furthermore, the intelligent interaction engine can generate reassuring messages, differentiated advantage interpretations, or incentive programs in real time, increasing user purchase intention conversion rates and significantly improving order conversion rates. By integrating knowledge graphs and user profiles, the system can identify the preferences of high-potential customers and optimal product combinations, providing data support for product selection and promotions. Intelligent add-to-cart and fulfillment tracking functions improve order completion rates and reduce sales losses due to abandoned purchases.

[0068] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device and medium embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the description of the method embodiments.

[0069] The devices and media provided in this application are one-to-one with the methods. Therefore, the devices and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0070] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0071] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0072] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0073] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0074] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0075] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0076] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0077] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0078] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of this specification.

Claims

1. An online intelligent shopping guide method based on a large model, characterized in that, The method includes: Multimodal demand perception and parsing processing is performed on the product demand information of the shopping guide to obtain real-time demand context information; By constructing and associating product information from e-commerce platforms into a knowledge graph, a product knowledge graph can be obtained. Based on consumers’ historical consumption records, we construct and iterate on a user profile tagging system to obtain real-time user profile information. Based on the real-time demand context information, the product knowledge graph, and the real-time user profile information, personalized recommendations and information interaction processing are performed on the consumer's shopping guide product demand information to obtain personalized product recommendation results. By using user feedback data, the personalized product recommendation results are dynamically optimized to obtain an optimized personalized product recommendation strategy.

2. The online intelligent shopping guide method based on a large model according to claim 1, characterized in that, The system performs multimodal demand perception and parsing on the product demand information of the shopping guide to obtain real-time demand context information, specifically including: The multimodal information acquisition submodule collects and processes the shopping guide product demand information in a multi-dimensional interactive manner to obtain multimodal information; wherein, the multi-dimensional interactive manner includes: text information, voice information and image information; The large-scale model semantic understanding submodule extracts key information about consumer demand types from the text information in the multimodal information; performs emotion recognition and analysis on the voice information in the multimodal information; identifies key product attributes from the image information in the multimodal information; and obtains multimodal demand element information based on the hierarchical processing of the multimodal information. Based on the multimodal demand element information, and through the demand element standardization sub-module, a demand element tagging system is constructed; wherein, the demand element tagging system includes at least: product category, core function, budget range, user group, usage scenario, preferred brand, and quality requirements; By using the aforementioned demand element tagging system, the contextual identification of key information in real-time shopping guide product demand information is performed to obtain the real-time demand context information.

3. The online intelligent shopping guide method based on a large model according to claim 1, characterized in that, By constructing and associating product information from e-commerce platforms into a knowledge graph, a product knowledge graph is obtained, which specifically includes: The product information collection and structuring submodule collects and processes multi-dimensional data from e-commerce platform product data sources to obtain multi-dimensional product data. The multi-dimensional product data includes: basic product information, functional attributes, relationships, and reputation data. The knowledge graph construction and association submodule performs hierarchical association processing on the attribute data in the multidimensional data of the products, and constructs the original associated product knowledge graph. The knowledge graph dynamic update submodule updates the original related product knowledge graph by updating the product data between corresponding nodes and relational attributes, thereby obtaining the product knowledge graph.

4. The online intelligent shopping guide method based on a large model according to claim 1, characterized in that, Based on consumers' historical consumption records, a user profile tagging system is constructed and iterated to obtain real-time user profile information, which specifically includes: The user data collection and integration submodule collects and integrates the consumer's historical multidimensional consumption data to obtain the historical consumption record data; wherein, the historical consumption record data includes: historical shopping records, interaction records, basic information, and real-time demand elements. A multi-dimensional user profile tagging system is established through a user profile tagging system construction sub-module; wherein, the multi-dimensional user profile tagging system includes: static tags, dynamic tags, behavioral tags, and evaluation tags; The historical consumption record data is classified by attribute tags through the multi-dimensional user profile tag system to obtain user product interaction profile features. The real-time user profile iteration submodule performs real-time iterative updates on the user product interaction profile features based on the latest consumer needs and preferences under new user interaction behaviors, and generates the real-time user profile information.

5. The online intelligent shopping guide method based on a large model according to claim 1, characterized in that, Based on the real-time demand context information, the product knowledge graph, and the real-time user profile information, personalized recommendation processing is performed on the consumer's product demand information to obtain personalized product recommendation results, specifically including: The personalized product recommendation submodule filters the candidate product set of the product knowledge graph based on the real-time demand context information and the real-time user profile information to obtain the most matching candidate product set. Based on the comprehensive multi-dimensional features of the products, the candidate products in the most matching candidate product set are intelligently sorted to obtain the product recommendation list information. The intelligent product comparison submodule compares the multi-dimensional correlation features of each product in the product recommendation list to obtain product correlation feature comparison information. The multi-dimensional correlation features include: core parameters, functional characteristics, price, user ratings, and after-sales service policies. By using the product recommendation list information and the product association feature comparison information, personalized recommendation processing is performed on the consumer's product demand information to obtain personalized product recommendation results.

6. The online intelligent shopping guide method based on a large model according to claim 5, characterized in that, The consumer's product recommendation information is processed through information interaction to obtain personalized product recommendation results, including: By using the contextualized shopping cart sub-module, the corresponding products in the personalized product recommendation results are added to the shopping cart under multiple instructions to obtain shopping cart product information based on the user's e-commerce platform account; Through the integrated order query and management submodule, the order integration information corresponding to the shopping guide product demand information is processed by querying under the user intent parsing to obtain the order status information; The shopping cart product information and the order status information are both associated with the personalized product recommendation results.

7. The online intelligent shopping guide method based on a large model according to claim 5, characterized in that, The process of interacting with the consumer's product recommendation information to obtain personalized product recommendation results also includes: Through the dynamic interaction strategy submodule, the product demand information of the shopping guide is processed by progressive hierarchical demand parsing using a language analysis model to obtain real-time user demand intent information. The potential demand mining submodule mines and processes users' potential demand information for shopping guide products, and generates potential product recommendation information based on the mined potential demand intent information. The real-time user demand intent information and the potential product recommendation information are both associated with the personalized product recommendation results.

8. The online intelligent shopping guide method based on a large model according to claim 1, characterized in that, By using user feedback data, the personalized product recommendation results are dynamically optimized to obtain an optimized personalized product recommendation strategy, which specifically includes: The feedback data collection and analysis submodule collects multidimensional feedback data from consumers and identifies problem information from the multidimensional feedback data to obtain feedback data analysis results. The multidimensional feedback data includes: direct consumer feedback, behavioral feedback, and transaction result data. The model and strategy optimization submodule incorporates the incremental feedback optimization features from the feedback data analysis results into the personalized product recommendation results, and adaptively adjusts the weights of the recommendation algorithm and matching rules for the personalized product recommendation results to obtain the optimized personalized product recommendation strategy.

9. An online intelligent shopping guide device based on a large model, characterized in that, The device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, enabling the at least one processor to execute a large-model-based online intelligent shopping guide method according to any one of claims 1-8.

10. A non-volatile computer storage medium, characterized in that, The storage medium is a non-volatile computer-readable storage medium that stores at least one program, each program including instructions that, when executed by a terminal, cause the terminal to perform an online intelligent shopping guide method based on a large model according to any one of claims 1-8.

Images