Intelligent commodity question and answer recommendation method and device, storage medium and computer equipment

Abstract

According to the intelligent commodity question and answer recommendation method and device, the storage medium and the computer equipment provided by the invention, after the question information input by the target user and the context information related to the question are obtained, the vectorized retrieval operation can be performed according to the obtained question information and the context information; and a preliminary result obtained by vector retrieval is filtered and screened by using a pre-constructed and continuously updated e-commerce knowledge graph, so that the accuracy and correlation of a final retrieval result are ensured. Then, the filtered final retrieval result, the original question information and the context information can be input into the large language model together, after the commodity keywords returned by the large language model are obtained, accurate search is conducted in a commodity library according to the commodity keywords, and according to the search result, the search efficiency is improved. In combination with the preference and historical behavior data of the user, comprehensive and personalized comprehensive recommendation is carried out on the target user, so that the shopping experience and satisfaction of the user are improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to an intelligent product question-and-answer recommendation method, device, storage medium, and computer equipment. Background Art

[0002] With the rapid development of mobile internet and the widespread adoption of e-commerce, many e-commerce platforms are facing the challenges of information overload and information disorientation. Consequently, recommendation systems have emerged. Recommendation systems play a crucial role on various internet platforms, enabling users to more efficiently and quickly discover products and information that they are interested in or potentially interested in.

[0003] However, existing recommendation systems mostly base their recommendations on users' historical behavior and interests. While this approach improves recommendation accuracy to a certain extent, it often overlooks the user's specific needs in the current context, leading to a mismatch between recommendation results and the user's actual needs, which in turn affects recommendation accuracy. Summary of the Invention

[0004] The purpose of this application is to solve at least one of the above-mentioned technical defects, especially the technical defect that the recommendation system in the existing technology often ignores the specific needs of users in the current context, resulting in a deviation between the recommendation results and the actual needs of the users, thereby affecting the accuracy of the recommendation.

[0005] This application provides an intelligent product question-answering recommendation method, which includes:

[0006] Obtain the question information and context information entered by the target user;

[0007] Performing a vector search based on the question information and the context information, and filtering the vector search results using a pre-built e-commerce knowledge graph to obtain a final search result;

[0008] Inputting the search results, the question information, and the context information into a large language model to obtain product keywords returned by the large language model;

[0009] Search the product library for the product keywords and make comprehensive recommendations to the target user based on the search results.

[0010] Optionally, obtaining context information includes:

[0011] Acquire multiple rounds of conversations before the target user inputs the question information, as well as short-term memory and long-term memory related to the target user;

[0012] The multiple rounds of dialogue, the short-term memory, and the long-term memory are used as context information.

[0013] Optionally, performing vector search based on the question information and the context information to obtain a vector search result includes:

[0014] Converting the question information and the context information into a text description vector;

[0015] Relevant information matching the text description vector is retrieved from a preset vector database, and a vector retrieval result is obtained.

[0016] Optionally, filtering the vector search results using a pre-built e-commerce knowledge graph to obtain final search results includes:

[0017] Extract entity IDs from vector search results;

[0018] Querying the pre-built e-commerce knowledge graph for product information associated with the entity ID;

[0019] The vector search results are filtered according to the product information to obtain a final search result.

[0020] Optionally, the process of constructing the e-commerce knowledge graph includes:

[0021] Obtain raw product data, predefined core product types in the e-commerce field, and attribute information corresponding to each core product type from multiple data sources;

[0022] After cleaning the original product data, intermediate product data is obtained;

[0023] Extracting data that matches the core product type from the intermediate product data as target product data, and extracting corresponding target attributes from the target product data;

[0024] According to the business logic and data characteristics of the e-commerce field, the relationship between different entities corresponding to the target product data and the target attributes is established, and after matching and fusing different data of the same entity, an e-commerce knowledge graph is obtained.

[0025] Optionally, after cleaning the original product data, obtaining the intermediate product data includes:

[0026] Deduplicating the original product data to obtain deduplicated original product data;

[0027] Convert the original product data after deduplication into a unified format and correct errors in the original product data after the unified format;

[0028] Fill missing values ​​in the corrected original product data, detect outliers in the filled original product data, and remove abnormal data;

[0029] The original e-commerce data after removing abnormal data is subjected to pattern mapping and alignment to obtain intermediate product data.

[0030] Optionally, the method further includes:

[0031] When changes are detected in the relevant data of the e-commerce platform, the e-commerce knowledge graph is updated according to the changed relevant data.

[0032] This application also provides an intelligent product question-answering recommendation device, comprising:

[0033] The information acquisition module is used to obtain the question information and context information input by the target user;

[0034] A hybrid retrieval module, configured to perform vector retrieval based on the question information and the context information, and filter the vector retrieval results using a pre-built e-commerce knowledge graph to obtain final retrieval results;

[0035] A keyword generation module, configured to input the search results, the question information, and the context information into a large language model to obtain product keywords returned by the large language model;

[0036] The comprehensive recommendation module is used to search for the product keywords in the product library and make comprehensive recommendations to the target user based on the search results.

[0037] The present application also provides a computer-readable storage medium, which stores computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the intelligent product question-and-answer recommendation method as described in any of the above embodiments.

[0038] The present application also provides a computer device, comprising: one or more processors, and a memory;

[0039] The memory stores computer-readable instructions, and when the computer-readable instructions are executed by the one or more processors, the steps of the intelligent product question-and-answer recommendation method as described in any one of the above embodiments are performed.

[0040] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0041] The intelligent product question-and-answer recommendation method, device, storage medium and computer equipment provided by the present application can first obtain the specific question information entered by the target user, and collect contextual information related to the question, so as to more comprehensively understand the user's actual needs. Then, the present application can perform a vectorized retrieval operation based on the obtained question information and its corresponding contextual information. In this process, the present application can use the pre-built and continuously updated e-commerce knowledge graph to perform fine filtering and screening on the preliminary results obtained by the vector retrieval, thereby ensuring the accuracy and relevance of the final retrieval results. Subsequently, the present application can input the filtered final retrieval results, together with the original question information and contextual information, into the large language model, and perform comprehensive analysis and processing of this information through the large language model to generate product keywords that are highly matched with user needs. Finally, the present application can perform accurate searches in the product library based on the obtained product keywords, and based on the search results, combined with the user's preferences and historical behavior data, make comprehensive and personalized recommendations to the target user to enhance the user's shopping experience and satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0043] Figure 1 A flowchart of an intelligent product question-and-answer recommendation method provided in an embodiment of the present application;

[0044] Figure 2 A schematic diagram of the process of filtering vector search results using a pre-built e-commerce knowledge graph provided in an embodiment of the present application;

[0045] Figure 3 A schematic diagram of the process of updating the e-commerce knowledge graph provided in an embodiment of the present application;

[0046] Figure 4 A schematic diagram of the structure of an intelligent question and answer recommendation device provided in an embodiment of the present application;

[0047] Figure 5 A schematic diagram of the internal structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0049] In one embodiment, Figure 1 As shown, Figure 1 This is a flow chart of a method for intelligent product question-and-answer recommendation provided in an embodiment of the present application. This application provides a method for intelligent product question-and-answer recommendation, which may include:

[0050] S110: Acquire question information and context information input by the target user.

[0051] In this step, when performing intelligent product question-and-answer recommendations, the question information and context information input by the target user can be obtained, so that accurate recommendations can be made to the target user based on the question information and context information.

[0052] Specifically, the question information input by the target user in this application includes text information and / or image information. Among them, the text information can be a query statement or question directly input by the target user, and the image information can be a product picture or related screenshot uploaded by the target user, so that the system can identify the target user's intention and make corresponding recommendations. The context information of this application can include multiple rounds of conversation records before the target user inputs the question information, as well as the target user's personal profile and historical behavior data, etc. These data can provide more comprehensive and in-depth background information for the target user's current needs, thereby improving the accuracy and personalization of recommendations.

[0053] S120: Perform vector search based on the question information and context information, and filter the vector search results through the pre-built e-commerce knowledge graph to obtain the final search results.

[0054] In this step, after obtaining the question information and context information input by the target user through S110, this application can also perform vector retrieval based on the question information and context information. After obtaining the vector retrieval results, this application can also filter the vector retrieval results through a pre-built e-commerce graph to obtain the final retrieval results.

[0055] When performing vector retrieval based on question information and context information, this application can convert the question information and context information into text description vectors through models such as the RAG (Retrieval Augmented Generation) information retrieval model or the Dense Passage Retriever (DPR) model, and then retrieve relevant information matching the text description vectors from a preset vector database to obtain vector retrieval results. The preset vector database can contain a large amount of text data such as product information, user reviews, and product descriptions, which are converted into vector form for efficient retrieval and matching.

[0056] It's understandable that RAG is a technology that combines external knowledge retrieval with a large language model. When generating text, RAG first retrieves relevant information from an external knowledge base and then feeds this information, along with the input prompt, into the large language model. This allows the model to generate more accurate and fact-based responses, effectively addressing the "hallucination" problem of the large language model and improving the quality and reliability of generated content.

[0057] Furthermore, after the present application performs vector retrieval based on the question information and context information and obtains the vector retrieval results, the present application can also filter the vector retrieval results through the pre-built e-commerce knowledge graph to obtain the final retrieval results. The e-commerce knowledge graph here refers to a knowledge base that contains various commodities and their related attributes and relationships in the e-commerce field. It extracts data that conforms to the core commodity types and the corresponding attribute information by cleaning, converting and integrating commodity data from different sources. At the same time, according to the business logic and data characteristics of the e-commerce field, the relationship between different entities corresponding to the extracted data and attribute information is established to form an e-commerce knowledge graph. This graph can help the system understand and filter the vector retrieval results more accurately, thereby effectively improving the accuracy and relevance of the final retrieval results.

[0058] S130: Input the search results, question information, and context information into the large language model to obtain product keywords returned by the large language model.

[0059] In this step, after obtaining the final search results through S120, this application can input the search results together with the question information and context information entered by the target user into the large language model, and obtain the product keywords returned by the large language model, so that accurate recommendations can be made for the target user based on the product keywords.

[0060] It is understandable that the large language model is a natural language processing model based on deep learning technology, which has powerful text generation and comprehension capabilities. Therefore, this application can use the large language model to give natural, coherent and accurate answers, and return the corresponding product keywords. Specifically, the large language model in this application can receive the search results after vector retrieval and e-commerce knowledge graph filtering, as well as the question information and context information entered by the target user. By conducting a comprehensive, detailed comprehensive analysis and efficient and accurate processing of this information, the large language model can generate highly matched and highly relevant product keywords based on the specific needs and preferences of the user, thereby effectively improving the accuracy of the search results and the satisfaction of the user experience. Moreover, since these product keywords reflect the specific needs and preferences of the target user in the current situation, they provide strong support for subsequent product recommendations.

[0061] For example, when a user asks about the performance of a certain mobile phone, this application can capture the user's specific concerns about the phone's performance, such as battery life, camera quality, or processor speed. By combining these specific needs with vector retrieval results and filtering using the e-commerce knowledge graph, this application can ensure that the information ultimately retrieved is closely related to the performance characteristics that the user cares about. This information is then input into the large language model, which further analyzes and extracts product keywords that best meet user needs, such as "long-lasting mobile phone", "high-pixel camera mobile phone", or "flagship processor mobile phone". These keywords not only accurately reflect the user's purchasing intentions, but also provide a clear direction for subsequent product searches and recommendations.

[0062] S140: Search for product keywords in the product library, and make comprehensive recommendations to target users based on the search results.

[0063] In this step, the search results, question information and context information are input into the large language model through S130. After obtaining the product keywords returned by the large language model, this application can also search for the product keywords in the product library and make comprehensive recommendations to the target users based on the search results.

[0064] In a specific implementation method, after obtaining the product keywords returned by the large language model, the present application can quickly locate the product entries containing these keywords in the index of the product library. In order to improve search efficiency and accuracy, the product library can pre-index and optimize to ensure that it can quickly respond to keyword search requests. Then, the present application can perform intelligent screening and sorting based on the searched product entries, combined with the target user's question information, contextual information, and historical behavior data. In this process, the present application can comprehensively consider multiple dimensions such as product matching, sales volume, evaluation, price, etc. to generate a product recommendation list that meets user needs and has personalized features. Finally, the system presents this recommendation list to the target user, providing the user with a convenient, efficient, and personalized shopping experience.

[0065] For example, when a user asks about a smartwatch suitable for outdoor sports, this application can capture the user's specific needs for the watch, such as waterproof performance, durability, and whether it has health monitoring functions. By combining these specific needs with the vector search results and filtering them using the e-commerce knowledge graph, this application can ensure that the retrieved information is closely related to the characteristics that the user cares about. Subsequently, this information will be input into the large language model, which will further analyze and extract product keywords that best meet the user's needs, such as "waterproof sports watch", "durable smartwatch" or "health monitoring outdoor watch". These keywords not only accurately reflect the user's purchasing intention, but also provide a clear direction for subsequent product searches and recommendations. After searching for relevant products in the product library, this application can also generate a personalized product recommendation list for the user based on product reviews, sales volume, price and other information to help users quickly find the most suitable product for them.

[0066] In addition, this application can also automatically attach data source descriptions to the product recommendation list, clearly indicating the information source of each recommended product, and increasing the transparency and credibility of the recommendation results. In this way, when users view the recommendation list, they can clearly understand what data and information each recommended product is based on, so that they can trust and accept these recommendations more. In addition, in order to improve the user's shopping experience, this application can also provide a detailed introduction and description of each product in the recommendation list, including the main functions, features, prices, user reviews, etc. of the product, so that users can fully understand the information of each product and make more informed purchasing decisions. At the same time, this application can also intelligently sort and optimize the recommendation list based on user preferences and historical behavior data, placing products that are more in line with user needs and more attractive at the front of the list, further improving the accuracy and personalization of recommendations.

[0067] In the above embodiment, the specific question information entered by the target user can be obtained first, and the context information related to the question can be collected at the same time, so as to more comprehensively understand the actual needs of the user. Then, the present application can perform a vectorized retrieval operation based on the obtained question information and its corresponding context information. In this process, the present application can use the pre-built and continuously updated e-commerce knowledge graph to perform fine filtering and screening on the preliminary results obtained by the vector retrieval, thereby ensuring the accuracy and relevance of the final retrieval results. Subsequently, the present application can input the filtered final retrieval results, together with the original question information and context information, into the large language model, and use the large language model to comprehensively analyze and process this information to generate product keywords that are highly matched with user needs. Finally, the present application can perform a precise search in the product library based on the obtained product keywords, and based on the search results, combined with the user's preferences and historical behavior data, make comprehensive and personalized recommendations to the target user to enhance the user's shopping experience and satisfaction.

[0068] In one embodiment, obtaining context information in S110 may include:

[0069] S111: Acquire multiple rounds of dialogues before the target user inputs the question information, as well as short-term memory and long-term memory related to the target user.

[0070] S112: Using the multiple rounds of dialogue, the short-term memory, and the long-term memory as context information.

[0071] In this embodiment, when obtaining context information, the present application can not only obtain multiple rounds of conversations between the question information input by the target user, but also obtain short-term memory and long-term memory related to the target user, and use the obtained multiple rounds of conversations, short-term memory and long-term memory as context information.

[0072] It is understandable that the context information of this application may include multiple rounds of conversations before the target user inputs the question information. These conversation contents reflect the user's gradual clarification of needs and consideration process for the product, which can not only maintain a coherent conversation process, but also solve the problem of reference resolution. At the same time, the context information of this application can also cover short-term memory and long-term memory related to the target user. Short-term memory includes but is not limited to product information that the user has recently browsed, purchased or searched, application filtering conditions, etc., while long-term memory involves the user's personal preferences, purchasing habits, and historical consumption records. Together, this information constitutes a comprehensive understanding of user needs, which helps to improve the accuracy and personalization of recommendations.

[0073] In one embodiment, performing vector search based on the question information and the context information in S120 to obtain vector search results may include:

[0074] S121: Convert the question information and the context information into a text description vector.

[0075] S122: Relevant information matching the text description vector is retrieved from a preset vector database, and a vector retrieval result is obtained.

[0076] In this embodiment, when converting question information and context information into text description vectors and performing vector retrieval, in order to ensure the accuracy and efficiency of the retrieval, the present application can take a series of optimization measures. First, the present application can use natural language processing technology to pre-process question information and context information, including steps such as removing stop words, stem extraction, and word form restoration to reduce noise and improve the purity of information. Secondly, when converting text description vectors, high-performance vector representation models such as BERT and GPT pre-trained language models can be selected. These models can capture semantic information and contextual relationships in the text, thereby generating more representative vector representations. In addition, the preset vector database also needs to be continuously updated and optimized to include more and more comprehensive text data such as product information and user reviews, and the data should be cleaned and integrated regularly to ensure that the information in the database is up to date and accurate.

[0077] After taking the above optimization measures, this application can retrieve relevant information that matches the text description vector in the vector database and obtain vector search results containing rich product information and related reviews. These results are not only highly relevant to the user's specific needs, but also further ensure the accuracy and pertinence of the information through filtering of the e-commerce knowledge graph. These carefully screened and processed vector search results will be input into the large language model, which can provide strong support for generating high-quality product keywords, thereby promoting the entire recommendation process to develop in a more intelligent and personalized direction.

[0078] In one embodiment, Figure 2 As shown, Figure 2 Schematic diagram of the process of filtering vector search results using a pre-built e-commerce knowledge graph provided in an embodiment of the present application; in S120, the vector search results are filtered using the pre-built e-commerce knowledge graph to obtain the final search results, which may include:

[0079] S123: Extract entity IDs from the vector search results.

[0080] S124: Query the product information associated with the entity ID in the pre-built e-commerce knowledge graph.

[0081] S125: Filter the vector search results according to the product information to obtain the final search results.

[0082] In this embodiment, when using the e-commerce knowledge graph to filter the vector search results, in order to ensure the accuracy and efficiency of the filtering, this application adopts a more detailed processing flow. First, this application can extract key entity IDs from the vector search results. These entity IDs usually represent key information such as products, brands, categories, etc., and are the bridge connecting the vector search results and the e-commerce knowledge graph. Then, this application can query the product information associated with these entity IDs in the pre-built e-commerce knowledge graph. This information may include detailed descriptions of the products, attributes, prices, user reviews, etc., which together constitute a comprehensive and in-depth understanding of the products.

[0083] Then, this application can carefully filter the vector search results based on the queried product information. In this process, this application can comprehensively consider multiple dimensions such as the matching degree, relevance, and timeliness of the product information to ensure that the final search results meet the needs of the user and are highly accurate and targeted. For example, if the vector search results contain product information that is not directly related to the user's question, or this information is outdated, this application will exclude it to avoid misleading or unnecessary trouble to the user. At the same time, for product information that is highly relevant to the user's question and has accurate and timely information, this application will retain it as an important part of the final search results.

[0084] In the above embodiment, the vector search results, filtered through the e-commerce knowledge graph, not only more accurately reflect user needs but also provide a solid foundation for subsequent product keyword generation and recommendation. This filtering process not only improves the accuracy of recommendations but also further enhances the user's shopping experience, allowing users to more easily find products that meet their needs.

[0085] In one embodiment, the process of constructing the e-commerce knowledge graph may include:

[0086] S210: Obtaining original product data, predefined core product types in the e-commerce field, and attribute information corresponding to each core product type from multiple data sources.

[0087] S211: After cleaning the original product data, intermediate product data is obtained.

[0088] S212: extracting data that matches the core product type from the intermediate product data as target product data, and extracting corresponding target attributes from the target product data.

[0089] S213: Based on the business logic and data characteristics of the e-commerce field, establish the relationship between different entities corresponding to the target product data and the target attributes, and obtain the e-commerce knowledge graph after matching and fusing different data of the same entity.

[0090] In this embodiment, when constructing the e-commerce knowledge graph, this application pays more attention to the diversity of data sources and the comprehensiveness of data. Therefore, this application obtains original product data from multiple data sources, which can ensure the richness and complementarity of information. These data sources can cover e-commerce platforms, business product libraries, operation material libraries, and professional evaluation websites. Among them, the business product library contains information such as SKU, description, specifications, price, inventory, etc.; the operation material library brings together a variety of rich media content such as marketing copy, promotional materials, product pictures, videos, etc.; the e-commerce platform contains order library sales information, which involves transaction data, product popularity, sales trends, combination purchases and customer preferences. In addition, big data statistics also cover relevant business indicators, such as user behavior, sales rankings, repurchase rates, and seasonal demand.

[0091] Furthermore, the core product types and their corresponding attribute information in the e-commerce field predefined in this application provide a clear direction and framework for the construction of the knowledge graph, helping to ensure the accuracy and practicality of the graph. Among them, this application can define the core entity types in the e-commerce field, including but not limited to products, categories, materials, manufacturers, etc. At the same time, attribute information is defined for each entity type, such as products have product names, brands, SKUs, descriptions, prices, and selling points; materials have titles and material content; manufacturers have names and introductions, etc. Finally, the relationships between entities are defined, such as materials - belong to products, products - belong to manufacturers, etc.

[0092] Next, this application can remove redundant, erroneous or irrelevant information by cleaning the original product data, and obtain more refined and reliable intermediate product data. Furthermore, this application can extract data that conforms to the core product type from the intermediate product data as target product data, and extract the corresponding target attributes. This step ensures that the information in the knowledge graph is closely related to the e-commerce field and has high quality and value. When establishing entity relationships, this application fully considers the business logic and data characteristics of the e-commerce field to ensure that the established relationships are consistent with the actual business scenario and can accurately reflect the intrinsic connection between products. At the same time, matching and fusing different data for the same entity can effectively integrate scattered information to form a more complete and accurate knowledge graph. The final e-commerce knowledge graph not only contains rich product information and attribute data, but also establishes a clear entity relationship network, which provides strong data support and knowledge guarantee for subsequent product recommendations.

[0093] In one embodiment, after the original product data is cleaned in S211, intermediate product data is obtained, which may include:

[0094] S2111: Deduplication is performed on the original product data to obtain deduplicated original product data.

[0095] S2112: Convert the original product data after deduplication into a unified format, and correct errors in the original product data in the unified format.

[0096] S2113: Fill missing values ​​in the corrected original product data, and perform outlier detection on the filled original product data and remove abnormal data.

[0097] S2114: Schema mapping and alignment are performed on the original e-commerce data after removing abnormal data to obtain intermediate product data.

[0098] In this embodiment, when cleaning the original commodity data, each link is particularly critical and indispensable. First of all, the deduplication operation is an initial step, and its importance is self-evident. Through this operation, the present application can accurately identify and effectively remove those redundant commodity entries or attribute information, thereby ensuring the uniqueness and accuracy of the data and avoiding analysis deviations caused by duplicate data. Then, after the deduplication operation is completed, the present application can convert the screened data into a unified format, such as a unified unit of measurement, color name, brand name, category classification system, etc. This step not only significantly improves the overall efficiency of data processing, but also ensures the consistency of the data, laying a solid foundation for subsequent error correction links. During this conversion process, the present application can use text matching technology and format conversion algorithms to automatically identify and convert commodity data in various formats to ensure that they can be processed seamlessly and efficiently on the same platform.

[0099] Afterwards, the error correction phase begins. This step mainly targets various errors in the data, such as spelling errors, incorrect prices, damaged image links, etc. This application uses the dual guarantees of automatic recognition by intelligent algorithms and manual review to accurately correct these errors, thereby greatly improving the accuracy and reliability of the data and providing more solid data support for subsequent analysis. Next, when filling missing values ​​in the error-corrected data, this application can adopt a variety of flexible filling strategies, including but not limited to mean filling, median filling, mode filling, and prediction filling based on machine learning models, etc., to ensure the integrity and consistency of the data and avoid analysis distortion caused by missing values.

[0100] At the same time, this application can also perform outlier detection on the filled data and remove those abnormal data that are obviously deviated from the normal range. This step can effectively identify and exclude outliers, prevent them from causing unnecessary interference to subsequent data analysis, and ensure the accuracy and reliability of the analysis results. Finally, when performing pattern mapping and alignment, this application can standardize data from different sources and formats, so that they can be fairly and effectively compared and analyzed on the same dimension, thereby obtaining more accurate and comprehensive intermediate commodity data.

[0101] After this series of meticulous and systematic data cleaning steps, this application provides a solid and reliable data foundation for subsequent product recommendations, significantly improving the accuracy and user satisfaction of the recommendation system, and ultimately achieving better user experience and commercial value.

[0102] In one embodiment, Figure 3 As shown, Figure 3 A schematic diagram of a process for updating an e-commerce knowledge graph provided in an embodiment of the present application; the method may further include:

[0103] S150: When changes are detected in the relevant data of the e-commerce platform, the e-commerce knowledge graph is updated according to the changed relevant data.

[0104] In this embodiment, the update process of the e-commerce knowledge graph can include multiple aspects. On the one hand, when product information on the e-commerce platform changes, such as new products being added or removed from shelves, price adjustments, inventory changes, etc., the system can automatically capture these changes and synchronize the changed product data to the e-commerce knowledge graph in real time or on a scheduled basis. On the other hand, information such as product attributes and relationships in the e-commerce knowledge graph can also be updated and optimized based on business logic or data characteristics to ensure the accuracy and completeness of the graph.

[0105] Furthermore, as the e-commerce sector evolves and changes, new product types, attributes, or relationships may continue to emerge. The system needs to be able to promptly incorporate this new information into the e-commerce knowledge graph to maintain its timeliness and advancement. Therefore, when updating the e-commerce knowledge graph, the system can also consider using incremental updates, where only the changed parts are updated. This reduces the consumption of computing and storage resources and improves update efficiency.

[0106] At the same time, in order to ensure the continued effectiveness and accuracy of the e-commerce knowledge graph, this application can also set up a regular verification and review mechanism. Under this mechanism, this application can regularly verify the data in the e-commerce knowledge graph, and by comparing it with the product data of the actual e-commerce platform, promptly discover and correct errors or outdated information in the graph. In addition, this application can also invite domain experts or senior users to manually review the data in the graph to ensure the accuracy and authority of the data. The implementation of these verification and review mechanisms will further enhance the reliability and practicality of the e-commerce knowledge graph, and provide more accurate and comprehensive data support for subsequent product recommendations.

[0107] The following describes an intelligent product question-and-answer recommendation device provided in an embodiment of the present application. The intelligent product question-and-answer recommendation device described below and the intelligent product question-and-answer recommendation method described above can refer to each other.

[0108] In one embodiment, Figure 4 As shown, Figure 4 This is a schematic diagram of the structure of an intelligent question-and-answer recommendation device provided in an embodiment of the present application. This application also provides an intelligent product question-and-answer recommendation device, which may include an information acquisition module 210, a hybrid search module 220, a keyword generation module 230, and a comprehensive recommendation module 240, specifically including the following:

[0109] The information acquisition module 210 is used to acquire question information and context information input by the target user.

[0110] The hybrid retrieval module 220 is configured to perform vector retrieval based on the question information and the context information, and filter the vector retrieval results using a pre-built e-commerce knowledge graph to obtain final retrieval results.

[0111] The keyword generation module 230 is configured to input the search results, the question information, and the context information into a large language model to obtain product keywords returned by the large language model.

[0112] The comprehensive recommendation module 240 is used to search for the product keywords in the product library and make comprehensive recommendations to the target user based on the search results.

[0113] In the above embodiment, the specific question information entered by the target user can be obtained first, and the context information related to the question can be collected at the same time, so as to more comprehensively understand the actual needs of the user. Then, the present application can perform a vectorized retrieval operation based on the obtained question information and its corresponding context information. In this process, the present application can use the pre-built and continuously updated e-commerce knowledge graph to perform fine filtering and screening on the preliminary results obtained by the vector retrieval, thereby ensuring the accuracy and relevance of the final retrieval results. Subsequently, the present application can input the filtered final retrieval results, together with the original question information and context information, into the large language model, and use the large language model to comprehensively analyze and process this information to generate product keywords that are highly matched with user needs. Finally, the present application can perform a precise search in the product library based on the obtained product keywords, and based on the search results, combined with the user's preferences and historical behavior data, make comprehensive and personalized recommendations to the target user to enhance the user's shopping experience and satisfaction.

[0114] In one embodiment, the present application also provides a computer-readable storage medium, which stores computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the intelligent product question-and-answer recommendation method as described in any of the above embodiments.

[0115] In one embodiment, the present application further provides a computer device, including: one or more processors, and a memory.

[0116] The memory stores computer-readable instructions, and when the computer-readable instructions are executed by the one or more processors, the steps of the intelligent product question-and-answer recommendation method as described in any one of the above embodiments are performed.

[0117] Schematically, as Figure 5 As shown, Figure 5 This is a schematic diagram of the internal structure of a computer device provided in an embodiment of the present application. The computer device 300 can be provided as a server. Figure 5 Computer device 300 includes a processing component 302, which further includes one or more processors, and memory resources represented by memory 301 for storing instructions executable by processing component 302, such as application programs. The application programs stored in memory 301 may include one or more modules, each corresponding to a set of instructions. Furthermore, processing component 302 is configured to execute the instructions to perform the intelligent product question-and-answer recommendation method of any of the above-described embodiments.

[0118] The computer device 300 may further include a power supply component 303 configured to perform power management of the computer device 300, a wired or wireless network interface 304 configured to connect the computer device 300 to a network, and an input / output (I / O) interface 305. The computer device 300 may operate based on an operating system stored in the memory 301, such as Windows Server™, Mac OS X™, Unix™, Linux™, Free BSD™, or the like.

[0119] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0120] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0121] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referenced to each other.

[0122] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An intelligent product question-answering recommendation method, characterized in that: The method comprises: Obtain the question information and context information entered by the target user; Performing a vector search based on the question information and the context information, and filtering the vector search results using a pre-built e-commerce knowledge graph to obtain a final search result; Inputting the search results, the question information, and the context information into a large language model to obtain product keywords returned by the large language model; Search the product library for the product keywords and make comprehensive recommendations to the target user based on the search results.

2. The intelligent product question-answering recommendation method according to claim 1, characterized in that: The acquiring of context information includes: Acquire multiple rounds of conversations before the target user inputs the question information, as well as short-term memory and long-term memory related to the target user; The multiple rounds of dialogue, the short-term memory, and the long-term memory are used as context information.

3. The intelligent product question-answering recommendation method according to claim 1, characterized in that: The performing vector search based on the question information and the context information to obtain a vector search result includes: Converting the question information and the context information into a text description vector; Relevant information matching the text description vector is retrieved from a preset vector database, and a vector retrieval result is obtained.

4. The intelligent product question-answering recommendation method according to claim 1, characterized in that: The vector search results are filtered using the pre-built e-commerce knowledge graph to obtain the final search results, including: Extract entity IDs from vector search results; Querying the pre-built e-commerce knowledge graph for product information associated with the entity ID; The vector search results are filtered according to the product information to obtain a final search result.

5. The intelligent product question-answering recommendation method according to claim 1 or 4, characterized in that: The construction process of the e-commerce knowledge graph includes: Obtain raw product data, predefined core product types in the e-commerce field, and attribute information corresponding to each core product type from multiple data sources; After cleaning the original product data, intermediate product data is obtained; Extracting data that matches the core product type from the intermediate product data as target product data, and extracting corresponding target attributes from the target product data; According to the business logic and data characteristics of the e-commerce field, the relationship between different entities corresponding to the target product data and the target attributes is established, and after matching and fusing different data of the same entity, an e-commerce knowledge graph is obtained.

6. The intelligent product question-answering recommendation method according to claim 5, characterized in that: After the raw product data is cleaned, intermediate product data is obtained, including: Deduplicating the original product data to obtain deduplicated original product data; Convert the original product data after deduplication into a unified format and correct errors in the original product data after the unified format; Fill missing values ​​in the corrected original product data, detect outliers in the filled original product data, and remove abnormal data; The original e-commerce data after removing abnormal data is subjected to pattern mapping and alignment to obtain intermediate product data.

7. The intelligent product question-answering recommendation method according to claim 1, characterized in that: The method further comprises: When changes are detected in the relevant data of the e-commerce platform, the e-commerce knowledge graph is updated according to the changed relevant data.

8. An intelligent product question-answering recommendation device, characterized in that: include: The information acquisition module is used to obtain the question information and context information input by the target user; A hybrid retrieval module, configured to perform vector retrieval based on the question information and the context information, and filter the vector retrieval results using a pre-built e-commerce knowledge graph to obtain final retrieval results; A keyword generation module, configured to input the search results, the question information, and the context information into a large language model to obtain product keywords returned by the large language model; The comprehensive recommendation module is used to search for the product keywords in the product library and make comprehensive recommendations to the target user based on the search results.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the steps of the intelligent commodity question-and-answer recommendation method according to any one of claims 1 to 7.

10. A computer device, characterized in that: include: one or more processors, and memory; The memory stores computer-readable instructions, which, when executed by the one or more processors, perform the steps of the intelligent commodity question-and-answer recommendation method according to any one of claims 1 to 7.

Images