CN-122022939-A - Intelligent recommendation method and system for commodity search

CN122022939ACN 122022939 ACN122022939 ACN 122022939ACN-122022939-A

Abstract

The invention provides an intelligent recommendation method and system for commodity searching in the technical field of information retrieval and intelligent recommendation, and the method comprises the steps of S1, receiving search query input by a user, carrying out natural language processing on the search query to extract key entities and query intentions, S2, carrying out commodity matching through an AI model comprising a classification module, a fuzzy matching module and a normalization module based on the key entities and the query intentions to obtain a matching result, S3, carrying out multi-feature fusion sequencing comprising coarse sequencing and fine sequencing on the matching result by combining user behavior data comprising at least one of browsing behavior, clicking behavior and collecting behavior to obtain a sequenced commodity list, and S4, displaying the commodity list through a visual interface. The commodity recommendation method has the advantages that accuracy, adaptability and user experience of commodity recommendation are greatly improved.

Inventors

LIN FUQIN
LIN LONG
YANG FENG
DENG RONGJIE
ZHANG SHENGBIN
WANG LINJIE
ZHENG XINGANG

Assignees

数采小博科技发展有限公司

Dates

Publication Date: 20260512
Application Date: 20251215

Claims (10)

1. The intelligent recommending method for commodity searching is characterized by comprising the following steps: Step S1, receiving search query input by a user through a visual interface, and carrying out natural language processing at least comprising text standardization, chinese word segmentation and entity recognition on the search query so as to extract key entities and query intentions in the search query, wherein the entity recognition comprises universal entity recognition and custom entity recognition, the universal entity recognition is used for extracting brands, types and models, and the custom entity recognition is configured based on purchasing scenes so as to recognize professional purchasing intentions; step S2, carrying out commodity matching through an AI model comprising a classification module, a fuzzy matching module and a normalization module based on the key entity and the query intention to obtain a matching result; step S3, combining user behavior data comprising at least one of browsing behavior, clicking behavior and collecting behavior, and performing multi-feature fusion sequencing comprising coarse sequencing and fine sequencing on the matching result to obtain a sequenced commodity list; And S4, displaying the commodity list through the visual interface, wherein the visual interface provides a search effect evaluation interface for dynamically intervening the ordering rule and the ordering weight of the commodity list.
2. The intelligent recommendation method for commodity searching according to claim 1, wherein in step S2, said classification module is configured to classify commodity names and brands; the fuzzy matching module is used for processing the hyponyms, the noise descriptions and the redundant information, supporting the expansion of the hyponyms and the noise filtering; The normalization module is used for carrying out standardized comparison on characteristic elements of different commodities.
3. The intelligent recommendation method for commodity searching according to claim 1, wherein in step S2, the AI model achieves semantic matching through a vector index technology, and the vector index is constructed based on Milvus architecture and is used for calculating semantic similarity between search query and commodity.
4. The intelligent recommendation method for commodity searching according to claim 1, wherein in step S3, said coarse ranking is based on text relevance and semantic relevance; the fine ranking is based on business rules and user individuation characteristics; The searching multi-feature fusion ordering further comprises weighting calculation based on commodity features and scene features, wherein the commodity features at least comprise prices, sales and time to shelf, and the scene features at least comprise industry scenes and proprietary purchasing scenes.
5. The intelligent recommendation method for commodity searching according to claim 1, wherein in step S4, the search effect evaluation interface integrates a user behavior analysis function, collects search logs through an ELK log system, and generates hot word recommendation and search result related recommendation.
6. The intelligent recommendation system for commodity searching is characterized by comprising the following modules: The system comprises a search query input module, a user input module and a display module, wherein the search query input module is used for receiving search queries input by a user through a visual interface, and performing natural language processing at least comprising text standardization, chinese word segmentation and entity recognition on the search queries to extract key entities and query intentions in the search queries; the commodity matching module is used for carrying out commodity matching through an AI model comprising a classification module, a fuzzy matching module and a normalization module based on the key entity and the query intention to obtain a matching result; The commodity ordering module is used for combining user behavior data comprising at least one of browsing behaviors, clicking behaviors and collecting behaviors, and performing multi-feature fusion ordering comprising coarse ordering and fine ordering on the matching result to obtain an ordered commodity list; And the commodity recommendation module is used for displaying the commodity list through the visual interface, and the visual interface provides a search effect evaluation interface for dynamically intervening the ordering rule and the ordering weight of the commodity list.
7. The intelligent recommendation system for commodity searching according to claim 6, wherein said commodity matching module is configured to classify a commodity name and a brand; the fuzzy matching module is used for processing the hyponyms, the noise descriptions and the redundant information, supporting the expansion of the hyponyms and the noise filtering; The normalization module is used for carrying out standardized comparison on characteristic elements of different commodities.
8. The intelligent recommendation system for commodity searching according to claim 6, wherein in the commodity matching module, the AI model achieves semantic matching through a vector index technology, and the vector index is constructed based on Milvus architecture and is used for calculating semantic similarity between search queries and commodities.
9. The intelligent recommendation system for merchandise search of claim 6, wherein in said merchandise ordering module, said coarse ranking is based on text relevance and semantic relevance; the fine ranking is based on business rules and user individuation characteristics; The searching multi-feature fusion ordering further comprises weighting calculation based on commodity features and scene features, wherein the commodity features at least comprise prices, sales and time to shelf, and the scene features at least comprise industry scenes and proprietary purchasing scenes.
10. The intelligent recommendation system for commodity searching according to claim 6, wherein in said commodity recommendation module, said search effect evaluation interface integrates a user behavior analysis function, collects search logs through an ELK log system, and generates hotword recommendations and search result related recommendations.

Description

Intelligent recommendation method and system for commodity search Technical Field The invention relates to the technical field of information retrieval and intelligent recommendation, in particular to an intelligent recommendation method and system for commodity searching. Background With the rapid development of electronic commerce, commodity search systems have become an important tool for enterprise purchasing and consumer daily shopping. At present, a common commodity searching method mainly depends on text keyword matching and a rule engine, for example, an open source search engine (such as OpenSearch) is adopted to realize word segmentation processing, synonym expansion and weight sorting of query. The method has a certain effect in a general consumption scene (C end), but is difficult to adapt to professional requirements such as bulk commodity purchase, cross-provider price comparison and the like when facing enterprise purchase (B end). Because the existing engine has limitation on expansion capability, customized ordering rules and weight adjustment are difficult to flexibly support, so that search results are often inaccurate, and the practicability is insufficient. Specifically, the current commodity searching method mainly has the following problems: 1. The search results are incomplete and the ranking is unreasonable, namely, the traditional search algorithm has insufficient coverage on related commodities, and the actual relevance between the ranking logic and the search word is weak. For example, when searching for "oil", the system may return to the oil can and other edge commodities, while the core oil such as lubricating oil, edible oil and the like cannot be displayed in front, and when searching for synonyms such as "neutral pen" and "sign pen", the number difference of the results is remarkable (such as 716 and 307) and the defects of synonym recognition and expansion capability are reflected. 2. The intelligent recommendation has weak function, namely, under the professional purchasing scene, the intelligent recommendation lacks deep understanding of the intention of the user. For example, when searching for "scratch board", the first few pages of results are mostly scratch board conveyor accessories, but the scratch board commodity actually needed does not appear until page 8, and when searching for "battery", the results are mixed with irrelevant commodity such as printer, etc., which indicates that the existing algorithm can not effectively filter noise information. 3. The expandability of the technical architecture is insufficient, the existing system (such as OpenSearch) is difficult to flexibly integrate an artificial intelligent module, still relies on rule management, and lacks of system acquisition and analysis of user behavior data, so that personalized recommendation cannot be realized. In addition, the common double-tower semantic model is more focused on global features in commodity retrieval, has low sensitivity to short texts, and lacks information interaction between commodities to influence recall effects. 4. And the evaluation and intervention mechanism is lacking, and operators lack effective search effect evaluation tools, so that dynamic adjustment and intervention on the ordering rules and weights are difficult. Therefore, how to provide an intelligent recommendation method and system for commodity searching, so as to improve accuracy, adaptability and user experience of commodity recommendation, is a technical problem to be solved urgently. Disclosure of Invention The invention aims to solve the technical problem of providing an intelligent recommendation method and system for commodity searching, which are used for improving the accuracy, adaptability and user experience of commodity recommendation. In a first aspect, the present invention provides an intelligent recommendation method for searching for goods, including the steps of: Step S1, receiving search query input by a user through a visual interface, and carrying out natural language processing at least comprising text standardization, chinese word segmentation and entity recognition on the search query so as to extract key entities and query intentions in the search query, wherein the entity recognition comprises universal entity recognition and custom entity recognition, the universal entity recognition is used for extracting brands, types and models, and the custom entity recognition is configured based on purchasing scenes so as to recognize professional purchasing intentions; step S2, carrying out commodity matching through an AI model comprising a classification module, a fuzzy matching module and a normalization module based on the key entity and the query intention to obtain a matching result; step S3, combining user behavior data comprising at least one of browsing behavior, clicking behavior and collecting behavior, and performing multi-feature fusion sequencing comprising coarse sequencing an