CN-121329562-B - Personalized recommendation method and system based on knowledge graph

CN121329562BCN 121329562 BCN121329562 BCN 121329562BCN-121329562-B

Abstract

The invention provides a personalized recommendation method and a personalized recommendation system based on a knowledge graph, and relates to the technical field of knowledge graphs. The method comprises the steps of firstly collecting multi-source heterogeneous data such as basic information of a product source, a delivery period of the product, financial constraints, policy constraints and user behaviors, writing the multi-source heterogeneous data into a shopping knowledge graph through intelligent cleaning and entity-relation extraction, then constructing a user portrait by combining user budget, preference and purchase qualification, reasoning and screening a compliance product source candidate set in the knowledge graph by taking the portrait as a retrieval condition, generating comprehensive scores and sequencing the product sources in the candidate set according to preference matching degree, price adaptation degree, matching integrity and value-increasing potential, and finally outputting a plurality of product sources with the highest comprehensive scores and corresponding three-section recommendation interpretation, so that deep fusion of the multi-source data is realized, and the recommendation result is ensured to be accurate, compliance and interpretability.

Inventors

LI HUI
CHEN PANFENG
MA DAN
XU HUARONG
Min Shengtian
XIA SHENGJIE
TAN XINGZHOU

Assignees

贵州大学
贵州优联博睿科技有限公司

Dates

Publication Date: 20260512
Application Date: 20251124

Claims (9)

1. The personalized recommendation method based on the knowledge graph is characterized by comprising the following steps executed by a computer system: the method comprises the steps of collecting multi-source heterogeneous data of a product source, wherein the multi-source heterogeneous data comprises basic information data, market supply and demand data, policy constraint data, financial constraint data and user behavior data of the product source; Performing cleaning and entity-relation extraction on the multi-source heterogeneous data to obtain extraction information, and writing the extraction information into a shopping knowledge graph; constructing a user demand portrait according to user budget, preference and purchase qualification information; Taking the user demand portraits as search conditions, and reasoning in the shopping knowledge graph to obtain a product source candidate set meeting the policy and financial constraint; Comparing the attribute information of each product source in the product source candidate set with the user demand portrait, calculating the matching degree score of each product source relative to the user demand, and generating a comprehensive score by combining the object factors of price level, matched service and value-increasing potential; Sorting the product source candidate sets according to the comprehensive scores, and outputting a plurality of product sources with highest scores and corresponding recommendation interpretation information; And taking the user demand portrait as a retrieval condition, reasoning and obtaining a product source candidate set meeting policy and financial constraint in the shopping knowledge graph, wherein the product source candidate set comprises: Taking a budget interval, an applicable scene, special requirements and type preferences in the user requirement portrait as retrieval starting points, executing semantic subgraph construction in the shopping knowledge graph, and screening a preliminary product source set meeting space position, product quality and price conditions; removing product source nodes which are not provided with purchasing qualification or payment capability by a user from the preliminary product source set based on the purchasing qualification state and the financial constraint in the user demand portrait, and forming a compliant product source set; Calculating the map association confidence of each compliant product source entity in the compliant product source set based on the multi-jump path relation between the user entity and the product source entity in the shopping knowledge map, wherein the map association confidence reflects the fitting degree of the product source and the user requirement on the semantic structure; screening the product source entities with the atlas association confidence coefficient not lower than a preset threshold value in the compliant product source set as the product source candidate set; the expression of the map association confidence coefficient is as follows: Wherein, the The confidence is related to the atlas of the user entity and the product source entity; For the different semantic path types connecting the user entity and the product source entity, automatically traversing and counting by a shopping knowledge graph; Reflecting the semantic distance for the shortest path length of the user entity and the product source entity in the shopping knowledge graph, and calculating through a standard graph searching algorithm; a vector is embedded for the graph of the user entity, A vector is embedded for a graph of product source entities, And Generated by the same unsupervised embedding model; the Euclidean distance is used for measuring the difference of semantic similarity.
2. The knowledge-based personalized recommendation method according to claim 1, wherein the basic information data of the product source comprises supply amount, grade, type, total price, unit price, date of production and geographic location of the product source; The market supply and demand data comprises market total supply, product quality guarantee period, total amount of replaceable products and market total demand; the policy constraint data comprises a limited purchase policy, a limited credit policy, a purchase qualification requirement and tax standard; The financial constraint data comprises interest rate, credit payment deadline, staged payment proportion and user credit score; The user behavior data comprise browsing records, collection records, consultation records, trial records and active feedback contents of the user.
3. The knowledge-based personalized recommendation method according to claim 1, wherein performing cleansing and entity-relation extraction on the multi-source heterogeneous data to obtain extraction information, and writing the extraction information into a shopping knowledge graph, comprises: respectively performing format standardization, missing value complementation and abnormal value detection on the multi-source heterogeneous data to obtain a cleaned structured data set; Based on natural language processing technology, performing named entity recognition on text information in the structured dataset, and extracting core entities of product sources, positions, matched services, policy terms, financial products and user behaviors; Digging attribute relations and association relations from the core entities by combining a preset domain ontology dictionary and semantic rules, and generating extraction information taking an entity-relation-entity triplet as a unit; and writing the extraction information into a graph database to construct a node set and an edge set of the shopping knowledge graph.
4. The knowledge-based personalized recommendation method according to claim 1, wherein constructing a user demand portrayal according to user budget, preference and purchase qualification information comprises: acquiring basic information provided by a user, estimating the shopping budget capacity of the user according to the basic information, and extracting population characteristic items, wherein the basic information comprises age, marital status, occupation, annual income, family structure and resident urban area; analyzing historical behavior data of a user, and generating shopping behavior preference distribution of the user by adopting a time weighted modeling method, wherein the historical behavior data comprises a browsing record, a collection record, a consultation record and a trial record; Performing feature fusion on the population characteristic items, the shopping budget capacity and the shopping behavior preference distribution to obtain preliminary portrait features of the user; Based on the basic information and the current policy constraint data, a preset purchasing qualification checking module is called to evaluate shopping area limitation, loan maximum rating and qualification state of the user, and the evaluation result is used as portrait constraint attribute; and constructing a user demand portrait vector according to the combination of the preliminary portrait features and the constraint attributes.
5. The knowledge-based personalized recommendation method according to claim 4, wherein the expression of the shopping behavior preference distribution is: Wherein, the For users A vector of shopping behavior preference distributions; Recording the total number of the accumulated historical behaviors of the user; is the first The strip behavior records the attribute vector of the associated product source, and is generated after unified characterization by a semantic embedding model; is the first The interaction depth factor of the behavior type corresponding to the behavior record is naturally obtained according to the behavior category mapping, and is browsed as 1, collected as 2, consulted as 3 and tried as 4; as the current time stamp is to be used, Is the first The occurrence timestamp of the bar behavior; an effective response weight function representing behavior; Is normalized by a factor defined as the sum of all behavior response weights, i.e For connecting Normalized to a probabilistic preference vector.
6. The knowledge-graph-based personalized recommendation method according to claim 4, wherein based on the basic information and the current policy constraint data, invoking a preset purchase qualification checking module to evaluate shopping area limitation, highest loan amount and qualification state of the user, and taking the evaluation result as portrait constraint attributes, comprising: The method comprises the steps of obtaining basic identity information of a user, wherein the basic identity information comprises a household registration type, a marital state, the number of family members, tax or social security records and a fixed asset holding condition; Searching a limited purchase rule, a limited credit rule and a limited purchase identification standard of the current region based on a market regulation policy corresponding to the current region; matching the basic identity information with the market regulation policy, judging whether the user has purchasing qualification in each target area, and outputting a corresponding area restriction list; calling a preset purchasing qualification checking module, and automatically calculating the highest loan amount obtained by the user according to the income level, liability condition, credit record and current interest rate environment of the user; And generating a purchase qualification state identification in the preliminary portrait characteristic according to the region restriction list, the loan maximum and whether qualification conditions are met, and taking the purchase qualification state identification as the portrait constraint attribute.
7. A knowledge-based shopping personalized recommendation system for implementing the method of any one of claims 1-6, comprising: The multi-source data acquisition unit is used for acquiring multi-source heterogeneous data of a product source, wherein the multi-source heterogeneous data comprises basic information data, market supply and demand data, policy constraint data, financial constraint data and user behavior data of the product source; The map construction and updating unit is used for executing cleaning and entity-relation extraction on the multi-source heterogeneous data to obtain extraction information, and writing the extraction information into a shopping knowledge map; the user portrait generation unit is used for constructing a user demand portrait according to user budget, preference and purchase qualification information; the candidate product source reasoning and screening unit is used for reasoning and obtaining a product source candidate set meeting the policy and financial constraint in the shopping knowledge graph by taking the user demand portrait as a retrieval condition; the matching degree calculating and scoring generating unit is used for comparing the attribute information of each product source in the product source candidate set with the user demand portrait, calculating the matching degree score of each product source relative to the user demand, and generating a comprehensive score by combining the object price level, the matched service and the target factors of the value-increasing potential; And the sequencing and recommendation interpretation output unit is used for sequencing the product source candidate sets according to the comprehensive scores and outputting a plurality of product sources with highest scores and corresponding recommendation interpretation information.
8. The electronic equipment is characterized by comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory are communicated with each other through the communication bus; the memory has stored therein a computer program which, when executed by the processor, causes the processor to perform the knowledge-graph based personalized recommendation method of any one of claims 1-6.
9. A computer-readable storage medium, on which a computer program is stored, characterized in that the program, when executed, implements the knowledge-graph-based personalized recommendation method according to any one of claims 1-6.

Description

Personalized recommendation method and system based on knowledge graph Technical Field The invention relates to the technical field of knowledge graphs in artificial intelligence general technology, in particular to a personalized recommendation method and system based on knowledge graphs. Background With the opening of various commodity information platforms, big data government interfaces and financial science and technology, shoppers can acquire mass product sources, interest rates and policy information on the internet. Knowledge maps are widely used in the recommendation fields of commodities, travel, movies and the like due to good semantic expression, collaborative filtering is still a mainstream interest modeling method, and the knowledge maps are combined to form an important trend for improving recommendation accuracy. At present, the industry is transitioning from "interest-driven" to "constraint-interest-driven" where, on the one hand, shopper aesthetic preferences, usage needs are captured, and, on the other hand, hard constraints such as related financial policies, government consumption coupon dispensing policies, limited purchase qualifications, regional product release periods, etc. must be met. Multimodal data (images, text, geographic information) is increasingly abundant, and graphic neural network-large language model collaborative reasoning and "interpretable recommendation" are becoming research hotspots. The existing scheme often uses a consumer product recommendation framework to ignore the value decision characteristic of a shopping scene, and mainly has the technical problems that ① lacks of coverage policy and cross-domain knowledge fusion of financial and market supply and demand data, ② is used for relieving insufficient cold start of a new product source and a new user, ③ has weak decoupling and individuation weighing capability on multi-objective conflict (price-quality-rise value), and ④ is difficult to respond to policy change in real time and output compliance explanation. Disclosure of Invention In order to overcome the defects of the prior art, the invention aims to provide a personalized recommendation method and a personalized recommendation system based on a knowledge graph, which are used for realizing accurate, compliance and interpretable shopping personalized recommendation by combining a user portrait and a multidimensional constraint condition through fusing multisource data to construct the knowledge graph. Based on a first main aspect of the present invention, there is provided a personalized recommendation method based on a knowledge graph, comprising the following steps performed by a computer system: acquiring basic information of a product source, a release period of the product, financial constraints, policy constraints, user behaviors and other multi-source heterogeneous data; Performing cleaning and entity-relation extraction on the multi-source heterogeneous data to obtain extraction information, and writing the extraction information into a shopping knowledge graph; constructing a user demand portrait according to user budget, preference and purchase qualification information; Taking the user demand portraits as search conditions, and reasoning in the shopping knowledge graph to obtain a product source candidate set meeting the policy and financial constraint; Comparing the attribute information of each product source in the product source candidate set with the user demand portrait, calculating the matching degree score of each product source relative to the user demand, and generating a comprehensive score by combining the object factors of price level, matched service and value-increasing potential; and sequencing the product source candidate sets according to the comprehensive scores, and outputting a plurality of product sources with highest scores and corresponding recommendation interpretation information. Preferably, the multi-source heterogeneous data comprises basic information data of a product source, market supply and demand data, financial constraint data, policy constraint data and user behavior data, wherein the product source information data comprises supply quantity, grade, type, total price, unit price, production date and geographic position (producing place) of the product source, the market supply and demand data comprises market total supply, product quality guarantee period, replaceable product total amount and market total demand, the financial constraint data comprises interest rate, credit payment term, stage payment proportion and user credit score, the policy constraint data comprises purchase limiting policy, credit limiting policy, purchase qualification requirement and tax standard, and the user behavior data comprises browsing records, collecting records, consultation records, trial records and active feedback content of a user. As a further preferred aspect, in the foregoing method, performing cleansing and entity-re