A query method and device for purchasing commodity information

Abstract

The application discloses a kind of query method and device of purchasing commodity information, it is related to wisdom supply chain technical field.The specific embodiment of the method includes: the received purchasing commodity query information is converted into purchasing commodity query vector;According to purchasing commodity query vector, query pre-configured commodity attribute vector database, obtain candidate commodity set and each candidate commodity corresponding commodity attribute vector;Based on purchasing commodity query vector and each candidate commodity corresponding commodity attribute vector, the multi-modal sorting processing is carried out to candidate commodity set, and commodity query result is determined according to sorting result.This embodiment receives purchasing commodity query information, directly according to the above information in pre-configured commodity attribute vector database is automatically searched and matched, obtains candidate commodity set, and then the multi-modal screening is carried out to candidate commodity set, to satisfy dynamic business requirement, improve the accuracy of query efficiency and query result.

Description

Technical Field

[0001] This invention relates to the field of smart supply chain technology, and in particular to a method and apparatus for querying procurement commodity information. Background Technology

[0002] In industrial goods retail scenarios, suppliers need to quote prices and fulfill orders based on customer orders. Currently, most suppliers use spreadsheets (such as Excel) to manage their inventory. However, when faced with a massive and diverse range of products, this method is inefficient, inaccurate, and unable to adapt to changing business needs. Summary of the Invention

[0003] In view of this, embodiments of the present invention provide a method and apparatus for querying procurement product information. When procurement product query information is received, the method automatically searches and matches in a pre-configured product attribute vector database based on the above information to obtain a candidate product set. The candidate product set is then subjected to multimodal filtering to meet dynamic business needs, thereby improving query efficiency and the accuracy of query results.

[0004] To achieve the above objectives, according to one aspect of the present invention, a method for querying purchase commodity information is provided, comprising: The received purchase product query information is transformed into a purchase product query vector; The system queries the pre-configured product attribute vector database based on the product query vector to obtain a set of candidate products and the product attribute vector corresponding to each candidate product. Based on the purchase product query vector and the product attribute vector corresponding to each candidate product, the candidate product set is sorted in a multimodal manner, and the product query result is determined according to the sorting result.

[0005] Optionally, the pre-configured product attribute vector database is obtained through the following steps: Acquire product attribute data from multiple sources; the product attribute data includes product name information, product brand information, product model information, and product specification information; The product attribute data is standardized based on a predefined unified text template. The standardized product attribute data is then input into a pre-trained semantic vector model to obtain product attribute vectors. A vector index structure is constructed based on the product attribute vector, and the vector index structure is stored to obtain a product attribute vector database.

[0006] Optionally, a pre-configured product attribute vector database is queried based on the product query vector to obtain a set of candidate products and the product attribute vector corresponding to each candidate product, including: Calculate the similarity between the purchase product query vector and each product attribute vector in the product attribute vector database, and obtain the similarity calculation result; Based on the similarity calculation results, a preset number of candidate product attribute vectors are obtained from the product attribute vector database, and the products corresponding to the candidate product attribute vectors are determined, resulting in a set of candidate products and a product attribute vector corresponding to each candidate product.

[0007] Optionally, based on the purchase product query vector and the product attribute vector corresponding to each candidate product, the candidate product set is subjected to multimodal sorting processing, and the product query result is determined according to the sorting result, including: Based on the purchase product query vector and the product attribute vector corresponding to each candidate product, the multi-dimensional matching result of each candidate product is determined; Based on the multi-dimensional matching results of each candidate product, sort each candidate product in the candidate product set to obtain the sorting result; Based on the preset query quantity, the sorted results are used to obtain the product query results.

[0008] Optionally, based on the purchase product query vector and the product attribute vector corresponding to each candidate product, the multi-dimensional matching result of each candidate product is determined, including: The system concatenates the product query information and the product attribute text corresponding to each candidate product to generate model prompts. The model prompts are input into a pre-trained first language model to obtain the text modality matching results; The purchase product query vector and the product attribute vector corresponding to each candidate product are input into a pre-trained second language model to obtain the vector modality matching result. The pre-configured business weight parameters, text modality matching results, and vector modality matching results are weighted and summed to obtain multi-dimensional matching results.

[0009] Optionally, after determining the product query results based on the sorting results, the following steps are also included: Display product search results to users and receive user feedback data on product search results; The business weight parameters are adjusted based on feedback behavior data, and a weighted sum is performed based on the adjusted business weight parameters.

[0010] According to a second aspect of the present invention, a device for querying purchase information is provided, comprising: The conversion module is used to convert the received purchase product query information into purchase product query vectors; The query module is used to query a pre-configured product attribute vector database based on the product query vector to obtain a set of candidate products and the product attribute vector corresponding to each candidate product. The determination module is used to perform multimodal sorting on the candidate product set based on the purchase product query vector and the product attribute vector corresponding to each candidate product, and determine the product query result based on the sorting result.

[0011] According to a third aspect of the present invention, an electronic device is provided, comprising: One or more processors; Memory, used to store one or more programs. When one or more programs are executed by one or more processors, the one or more processors implement the methods of any of the above embodiments.

[0012] According to a fourth aspect of the present invention, a computer-readable medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method of any of the above embodiments.

[0013] According to a fifth aspect of the present invention, a computer program product is provided, including a computer program that, when executed by a processor, implements the method of any of the above embodiments.

[0014] One embodiment of the above invention has the following advantages or beneficial effects: It transforms received purchase query information into a purchase query vector; it queries a pre-configured product attribute vector database based on the purchase query vector to obtain a candidate product set and a corresponding product attribute vector for each candidate product; based on the purchase query vector and the corresponding product attribute vector for each candidate product, it performs multimodal sorting on the candidate product set and determines the product query result based on the sorting result. In this embodiment, upon receiving purchase query information, it automatically searches and matches in the pre-configured product attribute vector database directly based on the above information to obtain a candidate product set, and then performs multimodal filtering on the candidate product set to meet dynamic business needs, thereby improving query efficiency and the accuracy of query results.

[0015] The further effects of the aforementioned unconventional alternative methods will be explained below in conjunction with specific implementation methods. Attached Figure Description

[0016] The accompanying drawings are provided to better understand the invention and are not intended to unduly limit the scope of the invention. Wherein: Figure 1 This is a schematic diagram of the main flow of the method for querying purchase commodity information according to an embodiment of the present invention; Figure 2This is a schematic diagram of the main flow of a method for querying purchase commodity information according to a preferred embodiment of the present invention; Figure 3 This is a schematic diagram of the main modules of the purchase commodity information query device according to an embodiment of the present invention; Figure 4 This is an exemplary system architecture diagram in which embodiments of the present invention can be applied; Figure 5 This is a schematic diagram of the structure of a computer system suitable for implementing terminal devices or servers of the present invention. Detailed Implementation

[0017] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of the present invention, including various details to aid understanding. These details should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the invention. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0018] It should be noted that the acquisition, storage, and application of personal information involved in the embodiments of the present invention comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0019] In industrial goods retail scenarios, suppliers need to quote prices and fulfill orders based on customer orders. Currently, most suppliers use Excel spreadsheets to manage inventory. However, when faced with a massive and diverse range of products, this method is inefficient, inaccurate, and unable to adapt to changing business needs.

[0020] In view of this, according to one aspect of the present invention, a method for querying purchase commodity information is provided.

[0021] Figure 1 This is a schematic diagram illustrating the main flow of a method for querying purchased goods information according to an embodiment of the present invention. Figure 1 As shown, the method for querying purchase commodity information according to an embodiment of the present invention includes the following steps S101 to S103.

[0022] Step S101: Convert the received purchase commodity query information into a purchase commodity query vector.

[0023] Purchase product query information refers to the various query contents entered by users when searching for purchased products. This content can include natural language text descriptions (such as "304 stainless steel M6 bolts" or "black waterproof sports backpack"), structured fields containing brand, model, or functional characteristics (such as brand fields, category fields, and specification fields), and contextual information from the search scenario (such as the category the user is currently browsing, historical click behavior, or query trigger words pushed in recommendation scenarios). The aforementioned purchase product query information is typically semantically complex and diverse in expression, requiring vectorization technology to transform it into a high-dimensional semantic representation that can be directly processed by computers. A purchase product query vector is a dense semantic vector obtained by encoding the purchase product query information. This vector can have a fixed dimension (such as 768 or 1024 dimensions) and can express the semantic relationships between query terms in a vector space.

[0024] When converting purchase query information into purchase query vectors, one approach is to encode the query information using a pre-trained semantic vector model (e.g., a Transformer-based text embedding model). The complete purchase query text is directly input into this model to obtain the corresponding purchase query vector. This method captures the contextual semantic relationships within the purchase query text and is suitable for natural language queries. Another approach, for structured purchase query information, firstly, fields such as brand, category, and attribute values ​​are structurally concatenated according to a pre-defined format to construct a unified text expression. This expression is then input into a semantic vector model for encoding, generating a structured purchase query vector. This method fully utilizes the structured attributes explicitly input by the user, improving the vector representation's ability to identify purchase entity entities. These methods can be used independently or in combination to ultimately obtain purchase query vectors that represent the query intent.

[0025] Step S102: Query the pre-configured product attribute vector database according to the product query vector to obtain the candidate product set and the product attribute vector corresponding to each candidate product.

[0026] The product attribute vector database stores the product attribute vector for each product. Each product corresponds to a product attribute vector, representing its position in the semantic space. When querying the product attribute vector database based on a purchase product query vector, the query vector is semantically matched with all product attribute vectors in the database. Products with higher similarity scores are selected to form a candidate product set. This candidate product set refers to the set of all products whose semantic expression is most similar to the purchase product query vector and meets a certain similarity threshold or ranking requirement; each candidate product in the set is associated with its corresponding product attribute vector.

[0027] When querying a product attribute vector database based on a product query vector, a similarity retrieval method using a vector index structure can be employed. This involves using inverted vector indexes, quantized vector indexes, or graph index structures (such as those based on hierarchical navigable small-world graphs) to perform efficient vector retrieval on the product attribute vector database. By calculating the similarity between the product query vector and the product attribute vectors in the database, the products with the highest similarity are selected as a candidate product set. Alternatively, a multimodal extended retrieval method can be used. Before performing vector retrieval, the query vector is first enhanced using a large language model to form multiple semantic variant vectors or multimodal fusion vectors. Then, multiple searches are performed on the product attribute vector database, and the product sets obtained from these searches are merged and deduplicated to obtain a candidate product set that covers a more complete semantic range. Simultaneously, the product attribute vector of each candidate product is obtained.

[0028] Step S103: Based on the purchase product query vector and the product attribute vector corresponding to each candidate product, perform multimodal sorting on the candidate product set, and determine the product query result based on the sorting result.

[0029] In this embodiment, the product query result is a set of target products or target products that match the user's product query information. This product query result is ultimately returned to the user to meet their product query needs. Specifically, multimodal ranking processing can be performed based on a retrieval enhancement generation mechanism. The product query information is concatenated with the original attribute text of candidate products to construct structured model prompts. These model prompts are then input into a pre-trained first language model, which outputs the textual semantic matching degree result between the candidate products and the user query using generative reasoning. Simultaneously, the product query vector and the product attribute vectors of candidate products are input into a pre-trained second language model, which calculates another vector semantic matching degree result based on similarity in the vector space. The two matching degree results are then fused to generate the final ranking score. Alternatively, a multimodal ranking process can be performed based on a multi-model integration mechanism. Multiple homogeneous or heterogeneous semantic models model the semantic similarity between the purchase product query vector and the product attribute vector. Multiple sets of matching results are obtained through independent calculation, and then weighted and summarized in combination with pre-configured business indicators (such as inventory priority factor and gross profit margin factor) to form a comprehensive ranking score. The candidate product set is ranked according to the comprehensive ranking score, and the products ranked higher are selected as the final product query results.

[0030] This embodiment transforms received purchase query information into purchase query vectors; it then queries a pre-configured product attribute vector database based on these vectors to obtain a candidate product set and a corresponding product attribute vector for each candidate product; finally, it performs multimodal sorting on the candidate product set based on the query vectors and the corresponding product attribute vectors, and determines the product query result based on the sorting result. Upon receiving purchase query information, this embodiment automatically searches and matches in the pre-configured product attribute vector database to obtain a candidate product set, and then performs multimodal filtering on the candidate product set to meet dynamic business needs, thereby improving query efficiency and the accuracy of query results.

[0031] Optionally, the pre-configured product attribute vector database is obtained through the following steps: acquiring multi-source product attribute data; wherein, the product attribute data includes product name information, product brand information, product model information, and product specification information; standardizing the product attribute data based on a predefined unified text template, inputting the standardized product attribute data into a pre-trained semantic vector model to obtain product attribute vectors; constructing a vector index structure based on the product attribute vectors, and storing the vector index structure to obtain the product attribute vector database.

[0032] In this embodiment, product attribute data can originate from various business systems such as product management systems, supply chain systems, image recognition systems, and operation and maintenance systems. It covers product names, titles, descriptions, brand information, specifications, attribute tags, category tags, image content information, and structured indicator information. Since multi-source data differ in format, field naming, text style, and semantic granularity, this embodiment first standardizes the product attribute data based on a predefined unified text template. This unified text template is used to uniformly organize and rearrange attribute fields from different sources. For example, it converts all product attribute descriptions into a unified text format according to a fixed structure of "basic information—specification attributes—functional description—image summary," thereby reducing the expression differences between data from different sources.

[0033] After standardization, the standardized product attribute data is input into a pre-trained semantic vector model to obtain product attribute vectors that represent the semantic features of product attributes. The semantic vector model can employ a text semantic matching model based on a dual-tower architecture or a higher-performance large-scale language model embedding model. This step maps complex, lengthy, and structurally inconsistent product attribute text into vectors with fixed dimensions and computable similarity. To improve vector retrieval performance and support online queries of large-scale product data, a vector index structure is further constructed based on the product attribute vectors. The vector index structure can employ a vector compression structure based on product quantization or a high-dimensional approximate nearest neighbor index structure based on a graph structure to achieve fast approximate search of product attribute vectors. The constructed vector index structure is stored in a vector database as a pre-configured product attribute vector database for subsequent efficient retrieval of candidate products based on the product query vector.

[0034] This embodiment, by pre-configuring a product attribute vector database, enables efficient, stable, and high-recall candidate product retrieval during the query phase. It effectively reduces the impact of differences in format and semantic expression between different data sources on matching accuracy, while improving the system's real-time response capability in large-scale data scenarios and enhancing the accuracy and reliability of product query results.

[0035] Optionally, the process involves querying a pre-configured product attribute vector database based on the product query vector to obtain a set of candidate products and a product attribute vector corresponding to each candidate product. This includes: calculating the similarity between the product query vector and each product attribute vector in the product attribute vector database to obtain the similarity calculation result; obtaining a preset number of candidate product attribute vectors from the product attribute vector database based on the similarity calculation result, and determining the products corresponding to the candidate product attribute vectors to obtain a set of candidate products and a product attribute vector corresponding to each candidate product.

[0036] In this embodiment, similarity can be obtained through cosine similarity, dot product similarity, or other vector space measures, forming a set of similarity calculation results for all products. Based on these similarity calculation results, product attribute vectors that are closest to the purchase product query vector are selected from the product attribute vector database according to the similarity ranking from high to low or low to high. For example, the 100 product attribute vectors with the highest similarity are selected. The product identification information corresponding to each product attribute vector is further determined, thereby constructing a candidate product set and the corresponding product attribute vector for each candidate product. In addition to the above implementation based on direct similarity calculation, approximate nearest neighbor retrieval can also be performed based on tree structure indexes (such as ball trees, covering trees, or hierarchical navigation small world graph structures) to achieve efficient retrieval of massive product attribute vectors.

[0037] By combining similarity calculation with recall strategies, this embodiment achieves efficient retrieval of a large-scale product database. This enables the system to filter out candidate products that are most relevant to the user's query semantics from a massive number of products within milliseconds, improving the accuracy of the recall stage, reducing computational resource consumption, and increasing the response speed and throughput of the overall retrieval chain, thereby significantly improving the user query experience and system stability.

[0038] Optionally, based on the purchase product query vector and the product attribute vector corresponding to each candidate product, the candidate product set is subjected to multimodal sorting processing, and the product query result is determined according to the sorting result, including: determining the multi-dimensional matching result of each candidate product based on the purchase product query vector and the product attribute vector corresponding to each candidate product; sorting each candidate product in the candidate product set according to the multi-dimensional matching result of each candidate product to obtain the sorting result; and filtering the sorting result based on a preset query quantity to obtain the product query result.

[0039] In this embodiment, the multi-dimensional matching results of candidate products at the semantic, vector, content, and business levels are determined by fusing and analyzing the purchase product query vector with the product attribute vector corresponding to each candidate product. Specifically, a deep semantic matching method based on a language model can be used, constructing model prompts in a unified format from the purchase product query information and the original attribute text of the candidate products, and inputting them into a pre-trained language model. The language model then combines the query context and attribute description to generate text modality matching results. Alternatively, a vector space-based similarity calculation method can be used, inputting the purchase product query vector and product attribute vector into a pre-trained semantic vector model, and generating vector modality matching results through cosine similarity, dot product similarity, or other deep matching structures. Both methods together constitute the multi-dimensional matching results. After determining the multi-dimensional matching results, each candidate product in the candidate product set can be sorted according to the combination of multi-dimensional features. For example, text modality matching results and vector modality matching results can be fused according to preset weights, or key business indicators such as inventory factors and gross profit margin factors can be introduced to combine semantic relevance and business importance into a ranking score after unified normalization. Alternatively, a learned ranking model (such as a gradient boosting ranking model or a deep learning-based ranking model) can be used to score multi-dimensional features to obtain a stable ranking order. After ranking, the top-ranked products can be filtered from the ranking results based on a preset number of queries to obtain the final product query results.

[0040] Alternatively, text features, vector features, and structured business features can be input into a unified multimodal fusion model, generating a fusion score through a multi-layer attention mechanism; or a reinforcement learning framework can be used to continuously update the ranking strategy based on user click behavior. Adopting any of these methods can improve ranking stability, increase the matching accuracy between candidate products and purchased product query information, and effectively enhance retrieval quality and user experience.

[0041] Optionally, based on the purchase product query vector and the product attribute vector corresponding to each candidate product, the multi-dimensional matching result of each candidate product is determined, including: concatenating the purchase product query information and the obtained product attribute text corresponding to each candidate product to generate model prompt words; inputting the model prompt words into a pre-trained first language model to obtain the text modality matching result; inputting the purchase product query vector and the product attribute vector corresponding to each candidate product into a pre-trained second language model to obtain the vector modality matching result; and weighting and summing the pre-configured business weight parameters, the text modality matching result, and the vector modality matching result to obtain the multi-dimensional matching result.

[0042] In this embodiment, the model prompts include the user's query requirements, descriptions of the core attributes of candidate products, optional contextual content, and instructions to guide the language model in outputting a matching score. This enables the language model to perform inference calculations based on a unified and standardized text structure. The first language model can be a large-scale language model with strong understanding and reasoning capabilities, such as a deep neural network model capable of inferring relevance, judging semantic consistency, and disambiguation. After understanding the query semantics and candidate product semantics in the prompts, the first language model outputs a text modality matching result between the candidate product and the query based on its own language reasoning ability. This text modality matching result can be represented as a continuous numerical value, used to measure the strength of the match as judged by the language model from dimensions such as semantics, logic, and text consistency.

[0043] To further quantify the similarity between the semantics of user queries and candidate products from a vector space perspective, the query vector for purchasing products and the corresponding product attribute vector for each candidate product are input into a pre-trained second language model. The second language model can be a semantic vector model with efficient vector similarity calculation capabilities. It can perform similarity calculations on the input vectors based on deep encoding networks, such as through cosine similarity, vector dot product, or other multi-dimensional vector metrics, and output vector modality matching results. These vector modality matching results reflect the degree of similarity between candidate products and user queries at the vector space level, and together with the text modality matching results, constitute cross-modal, multi-perspective matching information.

[0044] After obtaining matching results from the first language model and the second language model respectively, pre-configured business weight parameters are introduced to weight and summarize the matching results across different dimensions. These business weight parameters can be set according to actual business needs, product sales strategies, and resource management strategies, such as inventory priority parameters, gross profit margin priority parameters, or other adjustment factors related to business performance. Based on preset weighting rules, the text modality matching results, vector modality matching results, and business weight parameters are weighted and summed to calculate the final multi-dimensional matching result for each candidate product. This multi-dimensional matching result comprehensively reflects factors such as semantic reasoning, vector similarity, and business priority, making the ranking results more closely aligned with actual business scenarios. Furthermore, a collaborative filtering model based on user historical behavior can be introduced to supplement the evaluation of the potential click probability of candidate products.

[0045] This embodiment improves the relevance of query results to the user's true intent, enhances the accuracy of query recall and the business value of ranking, enabling users to find products that meet their needs faster and more accurately, thereby improving the overall intelligence level of the retrieval system and the user experience.

[0046] Optionally, after determining the product query results based on the sorting results, the method further includes: displaying the product query results to the user and receiving user feedback behavior data on the product query results; adjusting the business weight parameters based on the feedback behavior data, and performing a weighted summation based on the adjusted business weight parameters.

[0047] Specifically, product search results are displayed to users in the form of lists or images. Feedback data is collected in real time when users browse, click, add items to their cart, or place orders, forming user feedback records related to the current query. This feedback data is then structured and transformed into incremental feedback information that influences business weight parameters. A pre-set adaptive weight update strategy dynamically adjusts the business weight parameters used for weighted summation. Weight updates can employ an exponential smoothing adaptive update algorithm, such as iteratively correcting inventory priority factor weights, gross profit margin factor weights, or other business-related weights through weighted accumulation, shifting them in a direction that better reflects real user preferences. After weight adjustment, the adjusted business weight parameters are used to perform a new weighted summation on the multi-dimensional matching results of future queries, thus automatically adapting the final ranking results to the user's long-term behavioral trends.

[0048] In addition to updating weights based on actual user clicks and purchase behavior, external weight control methods based on business operation strategies can also be adopted. For example, the weight of high-margin categories can be dynamically increased during specific promotional periods, the weight of inventory factors can be automatically increased during clearance cycles, or manual weight injection can be carried out according to seasonal demand fluctuations, thereby achieving flexible sorting that is controllable by the business.

[0049] This embodiment introduces a feedback-driven weight adaptive mechanism into the product purchase query process, enabling the system to continuously absorb implicit expressions of user behavior data regarding ranking preferences. This makes the product purchase query more closely aligned with real needs and improves the accuracy of the ranking results.

[0050] Figure 2 This is a schematic diagram illustrating the main flow of a method for querying purchased goods information according to a preferred embodiment of the present invention. Before a user queries for purchased goods, this embodiment first prepares the data. Specifically, it integrates all the goods attribute data from the supplier's enterprise resource planning system, business database, or spreadsheet files. This goods attribute data includes key text attributes such as goods name, brand, model, and specifications, as well as optional image information. The goods name, brand, model, specifications, and optional image descriptions are standardized using a predefined unified text template, transforming the originally scattered attribute content into a unified text description with a complete semantic structure. Subsequently, the standardized goods attribute data is input into a pre-trained semantic vector model to extract deep semantic features of the goods, resulting in fixed-dimensional goods attribute vectors. Based on the generated goods attribute vectors, an efficient vector index structure is further constructed and stored in a goods attribute vector database to achieve low-latency, high-recall retrieval of massive amounts of goods.

[0051] like Figure 2 As shown, the system receives user-inputted product query information, constructs standardized product query text using a unified text template, and inputs it into the same semantic vector model used to build the vector database for vectorization processing, obtaining a product query vector representing the semantics of the user query. Based on this product query vector, a similarity retrieval is performed in a pre-configured product attribute vector database. The similarity calculation result is obtained by calculating the similarity between the product query vector and each product attribute vector, and the candidate product attribute vector with the highest similarity is selected according to a preset recall quantity to determine the corresponding candidate product set, thus achieving the initial recall of candidate products. For the recalled initial candidate products, the system further filters them in conjunction with business rules, including but not limited to brand whitelist filtering, price range filtering, and gross profit margin threshold filtering, to ensure the accuracy of candidate products in terms of business satisfaction.

[0052] Based on the determined set of candidate products, the system performs multimodal ranking processing based on the purchase product query vector and the corresponding product attribute vector of each candidate product to achieve multi-dimensional deep matching and comprehensive scoring. First, a first language model evaluation based on retrieval enhancement generation technology is performed to obtain text modality matching results. This involves concatenating the purchase product query information with the original product attribute text of each candidate product to generate model prompts, which are then input into a pre-trained first language model to obtain text modality matching results representing semantic consistency and professional scenario understanding. Next, a homogeneous or heterogeneous second language model evaluation is performed to obtain vector modality matching results. This involves inputting the purchase product query vector and the candidate product attribute vectors into another homogeneous or heterogeneous semantic model to obtain vector modality matching results representing fine-grained similarity of the vector space. Finally, the pre-configured business weight parameters, text modality matching results, and vector modality matching results are weighted and summed. By fusing inventory priority factors, gross profit margin factors, and multi-model matching scores, a multi-dimensional matching result is obtained. The candidate product set is sorted based on the multi-dimensional matching results of each candidate product, and the sorted results are filtered according to the preset number of queries. Finally, the product query results are obtained and displayed to the user.

[0053] This embodiment, after obtaining and displaying product query results to the user, also constructs a closed-loop feedback mechanism based on user behavior. This involves collecting and analyzing user feedback behavior data regarding product query results (such as implicit behaviors like clicks, adding to cart, and placing orders), and dynamically adjusting business weight parameters based on the analysis results. A weight update formula based on exponential smoothing is used to ensure that the business weight parameters continuously converge to values ​​that best match the actual business situation. Furthermore, the semantic vector model is periodically incrementally trained based on newly added product data and updated query logs, enabling the model to continuously adapt to changes in industrial product materials and improve its understanding of new products and new query patterns.

[0054] Through the above implementation methods, this embodiment realizes a full-link intelligent recommendation process from data standardization, semantic vector construction, vector database retrieval, RAG-enhanced language model matching, multi-model fusion ranking to user behavior-driven adaptive optimization. It can effectively improve the accuracy, real-time performance, and business adaptability of industrial product material recommendations, and improve the efficiency of industrial product procurement and the level of enterprise intelligence.

[0055] According to a second aspect of the present invention, a device for querying purchase commodity information is provided.

[0056] Figure 3 This is a schematic diagram of the main modules of a product purchase information query device according to an embodiment of the present invention. Figure 3 As shown, the product purchase information query device 300 includes: The conversion module 301 is used to convert the received purchase product query information into purchase product query vectors; The query module 302 is used to query a pre-configured product attribute vector database based on the product query vector to obtain a set of candidate products and the product attribute vector corresponding to each candidate product. The determination module 303 is used to perform multimodal sorting on the candidate product set based on the purchase product query vector and the product attribute vector corresponding to each candidate product, and determine the product query result based on the sorting result.

[0057] Optionally, the purchase commodity information query device 300 also includes a configuration module, which is used for: Acquire product attribute data from multiple sources; product attribute data includes product name information, product brand information, product model information, and product specification information; The product attribute data is standardized based on a predefined unified text template. The standardized product attribute data is then input into a pre-trained semantic vector model to obtain product attribute vectors. A vector index structure is constructed based on the product attribute vector, and the vector index structure is stored to obtain a product attribute vector database.

[0058] Optionally, the query module 302 is also used for: Calculate the similarity between the purchase product query vector and each product attribute vector in the product attribute vector database, and obtain the similarity calculation result; Based on the similarity calculation results, a preset number of candidate product attribute vectors are obtained from the product attribute vector database, and the products corresponding to the candidate product attribute vectors are determined, resulting in a set of candidate products and a product attribute vector corresponding to each candidate product.

[0059] Optionally, the determining module 303 is also used for: Based on the purchase product query vector and the product attribute vector corresponding to each candidate product, the multi-dimensional matching result of each candidate product is determined; Based on the multi-dimensional matching results of each candidate product, sort each candidate product in the candidate product set to obtain the sorting result; Based on the preset query quantity, the sorted results are used to obtain the product query results.

[0060] Optionally, the determining module 303 is also used for: The system concatenates the product query information and the product attribute text corresponding to each candidate product to generate model prompts. The model prompts are input into a pre-trained first language model to obtain the text modality matching results; The purchase product query vector and the product attribute vector corresponding to each candidate product are input into a pre-trained second language model to obtain the vector modality matching result. The pre-configured business weight parameters, text modality matching results, and vector modality matching results are weighted and summed to obtain multi-dimensional matching results.

[0061] Optionally, the purchase commodity information query device 300 also includes an adjustment module, which is used for: Display product search results to users and receive user feedback data on product search results; The business weight parameters are adjusted based on feedback behavior data, and a weighted sum is performed based on the adjusted business weight parameters.

[0062] It should be noted that the specific implementation details of the procurement information query device of the present invention have been described in detail in the above procurement information query method, so the content will not be repeated here.

[0063] According to a third aspect of the present invention, an electronic device is provided, comprising: one or more processors; and a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method provided in the first aspect of the present invention.

[0064] According to a fourth aspect of the present invention, a computer-readable medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method provided in the first aspect of the present invention.

[0065] According to a fifth aspect of the present invention, a computer program product is provided, including a computer program that, when executed by a processor, implements the method provided in the first aspect of the present invention.

[0066] Figure 4 An exemplary system architecture 400 is shown, which can be applied to the method or apparatus for querying purchase information of goods according to embodiments of the present invention.

[0067] like Figure 4 As shown, system architecture 400 may include terminal devices 401, 402, and 403, a network 404, and a server 405. Network 404 serves as the medium for providing communication links between terminal devices 401, 402, and 403 and server 405. Network 404 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0068] Users can use terminal devices 401, 402, and 403 to interact with server 405 via network 404 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 401, 402, and 403, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0069] Terminal devices 401, 402, and 403 can be various electronic devices with displays that support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0070] Server 405 can be a server that provides various services, such as a backend management server that supports shopping websites browsed by users using terminal devices 401, 402, and 403 (for example only). The backend management server can analyze and process data such as received purchase query requests, and feed back the processing results (such as product query results - for example only) to the terminal device.

[0071] It should be noted that the method for querying purchased goods information provided in this embodiment of the invention is generally run by server 405, and correspondingly, the device for querying purchased goods information is generally set in server 405.

[0072] It should be understood that Figure 4 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0073] The following is for reference. Figure 5 It shows a schematic diagram of the structure of a computer system 500 suitable for implementing a terminal device of the present invention. Figure 5 The terminal device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0074] like Figure 5 As shown, the computer system 500 includes a central processing unit (CPU) 501, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 502 or programs loaded from storage section 508 into random access memory (RAM) 503. The RAM 503 also stores various programs and data required for the operation of the system 500. The CPU 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0075] The following components are connected to I / O interface 505: an input section 506 including a keyboard, mouse, etc.; an output section 507 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 508 including a hard disk, etc.; and a communication section 509 including a network interface card such as a LAN card, modem, etc. The communication section 509 performs communication processing via a network such as the Internet. Drive 510 is also connected to I / O interface 505 as needed. Removable media 511, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 510 as needed so that computer programs read from them can be installed into storage section 508 as needed.

[0076] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 509, and / or installed from removable medium 511. When the computer program is run by the central processing unit (CPU) 501, it performs the functions defined above in the system of this invention.

[0077] It should be noted that the computer-readable medium shown in this invention can be a computer-readable signal medium, a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit programs for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0078] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more operable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually operate substantially in parallel, and they may sometimes operate in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0079] The modules described in the embodiments of the present invention can be implemented in software or hardware. The described modules can also be housed in a processor; for example, a processor may be described as including a conversion module, a query module, and a determination module. The names of these modules do not necessarily limit the module itself; for example, the conversion module may also be described as "a module for converting received purchase commodity query information into purchase commodity query vectors."

[0080] In another aspect, the present invention also provides a computer-readable medium, which may be included in the device described in the above embodiments; or it may exist independently and not assembled into the device. The computer-readable medium carries one or more programs, which, when executed by the device, cause the device to: convert received purchase commodity query information into a purchase commodity query vector; query a pre-configured commodity attribute vector database according to the purchase commodity query vector to obtain a set of candidate commodities and a commodity attribute vector corresponding to each candidate commodity; perform multimodal sorting processing on the candidate commodity set based on the purchase commodity query vector and the commodity attribute vector corresponding to each candidate commodity; and determine the commodity query result based on the sorting result.

[0081] The computer program product provided in this embodiment of the invention includes a computer program that, when executed by a processor, implements the method for querying purchased goods information in this embodiment of the invention.

[0082] The technical solution of this embodiment of the invention has the following advantages or beneficial effects: It converts received purchase commodity query information into purchase commodity query vectors; it queries a pre-configured commodity attribute vector database based on the purchase commodity query vectors to obtain a set of candidate commodities and a commodity attribute vector corresponding to each candidate commodity; based on the purchase commodity query vectors and the commodity attribute vectors corresponding to each candidate commodity, it performs multimodal sorting processing on the candidate commodity set, and determines the commodity query result based on the sorting result. In this embodiment, when receiving purchase commodity query information, it directly and automatically searches and matches in the pre-configured commodity attribute vector database based on the above information to obtain a set of candidate commodities, and then performs multimodal filtering on the candidate commodity set to meet dynamic business needs, thereby improving query efficiency and the accuracy of query results.

[0083] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can occur depending on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

[0084] It should be noted that the acquisition, storage, and application of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

Claims

1. A method for querying procurement commodity information, characterized in that, include: The received purchase product query information is transformed into a purchase product query vector; The pre-configured product attribute vector database is queried based on the product query vector to obtain a set of candidate products and the product attribute vector corresponding to each candidate product; Based on the procurement product query vector and the product attribute vector corresponding to each candidate product, the candidate product set is subjected to multimodal sorting processing, and the product query result is determined according to the sorting result.

2. The method according to claim 1, characterized in that, The pre-configured product attribute vector database is obtained through the following steps: Acquire product attribute data from multiple sources; wherein, the product attribute data includes product name information, product brand information, product model information, and product specification information; The product attribute data is standardized based on a predefined unified text template, and the standardized product attribute data is input into a pre-trained semantic vector model to obtain product attribute vectors. A vector index structure is constructed based on the product attribute vector, and the vector index structure is stored to obtain the product attribute vector database.

3. The method according to claim 1, characterized in that, The system queries a pre-configured product attribute vector database based on the product query vector to obtain a set of candidate products and a product attribute vector corresponding to each candidate product, including: Calculate the similarity between the purchased product query vector and each product attribute vector in the product attribute vector database to obtain the similarity calculation result; Based on the similarity calculation results, a preset number of candidate product attribute vectors are obtained from the product attribute vector database, and the products corresponding to the candidate product attribute vectors are determined to obtain a set of candidate products and a product attribute vector corresponding to each candidate product.

4. The method according to claim 1, characterized in that, Based on the procurement product query vector and the product attribute vector corresponding to each candidate product, the candidate product set is subjected to multimodal sorting processing, and the product query result is determined according to the sorting result, including: Based on the procurement product query vector and the product attribute vector corresponding to each candidate product, the multi-dimensional matching result of each candidate product is determined; Based on the multi-dimensional matching results of each candidate product, each candidate product in the candidate product set is sorted to obtain a sorting result; The sorting results are filtered based on a preset number of queries to obtain the product query results.

5. The method according to claim 4, characterized in that, Based on the procurement product query vector and the product attribute vector corresponding to each candidate product, the multi-dimensional matching result of each candidate product is determined, including: The purchased goods query information and the obtained product attribute text corresponding to each candidate product are concatenated to generate model prompt words; The model prompts are input into a pre-trained first language model to obtain the text modality matching result; The purchase product query vector and the product attribute vector corresponding to each candidate product are input into a pre-trained second language model to obtain the vector modality matching result. The pre-configured business weight parameters, the text matching result, and the vector matching result are weighted and summed to obtain the multi-dimensional matching result.

6. The method according to claim 5, characterized in that, After determining the product search results based on the sorting results, the following is also included: The product search results are displayed to the user, and the user's feedback behavior data on the product search results is received. The business weight parameters are adjusted based on the feedback behavior data, and a weighted sum is performed based on the adjusted business weight parameters.

7. A device for querying purchase commodity information, characterized in that, include: The conversion module is used to convert the received purchase product query information into purchase product query vectors; The query module is used to query a pre-configured product attribute vector database based on the purchased product query vector to obtain a set of candidate products and the product attribute vector corresponding to each candidate product; The determination module is used to perform multimodal sorting on the candidate product set based on the purchase product query vector and the product attribute vector corresponding to each candidate product, and determine the product query result based on the sorting result.

8. An electronic device, characterized in that, include: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-6.

9. A computer-readable medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-6.

Images