Search

CN-122022947-A - Resource exchange recommendation method, equipment, medium and product

CN122022947ACN 122022947 ACN122022947 ACN 122022947ACN-122022947-A

Abstract

The invention discloses a resource exchange recommendation method, equipment, medium and product, which are applied to the field of financial science and technology and comprise the steps of receiving natural language input by a target user, analyzing the natural language to obtain resource exchange intention, extracting multi-dimensional constraint characteristics of the target user, evaluating the resource exchange intention according to the multi-dimensional constraint characteristics through an intention confidence evaluation model to obtain confidence scores, and carrying out resource exchange recommendation on the target user according to the multi-dimensional constraint characteristics when the resource exchange intention is determined to be fuzzy intention according to the confidence scores. The accuracy of intention recognition is improved by extracting multidimensional constraint features and obtaining confidence scores through an intention confidence evaluation model. When the resource exchange intention is determined to be the fuzzy intention according to the confidence score, drawing labels are extracted and recommended, real preference behind the fuzzy requirement can be accurately mined, the recommendation is attached to the user reality, invalid recommendation is reduced, and user exchange intention and experience are improved.

Inventors

  • WANG XIAOHONG

Assignees

  • 中国工商银行股份有限公司

Dates

Publication Date
20260512
Application Date
20260126

Claims (10)

  1. 1. A resource redemption recommendation method, comprising: Receiving natural language input by a target user, and analyzing the natural language to obtain a resource exchange intention; Extracting multi-dimensional constraint features of a target user, and evaluating the resource exchange intention according to the multi-dimensional constraint features through an intention confidence evaluation model to obtain a confidence score, wherein the multi-dimensional constraint features comprise intention significance features, user attribute features, user behavior features and resource exchange constraint features; And when the resource exchange intention is determined to be the fuzzy intention according to the confidence score, carrying out resource exchange recommendation on the target user according to the multi-dimensional constraint characteristics.
  2. 2. The method of claim 1, wherein parsing the natural language to obtain a resource redemption intent comprises: Generating a first prompt word, carrying out semantic analysis on the natural language based on the first prompt word calling big model, filtering redundant information irrelevant to resource exchange, extracting intention elements relevant to the resource exchange, integrating the intention elements, and generating a resource exchange intention, wherein the intention elements comprise action verb elements, specific pointing elements and constraint description elements.
  3. 3. The method of claim 2, wherein the evaluating the resource redemption intent by the intent confidence assessment model based on the multi-dimensional constraint features to obtain a confidence score comprises: normalizing all intention elements in the multidimensional constraint features and the resource exchange intention to generate all standard input features; Generating a second prompting word, calling a large model based on the second prompting word, and determining feature weights corresponding to all standard input features through a dynamic weight adjustment mechanism based on preset association rules, wherein the second prompting word comprises preset rules; and inputting each standard input feature and corresponding feature weight into a pre-trained intention confidence assessment model, and calculating a confidence score.
  4. 4. The method of claim 1, wherein the making a resource redemption recommendation for a target user based on the multi-dimensional constraint features comprises: generating a third prompting word, calling a large model based on the third prompting word, and generating a portrait tag according to the multidimensional constraint features; Obtaining a pool of resource exchange marks, wherein the pool of resource exchange marks comprises attributes of each resource exchange mark and corresponding resource exchange marks, required resources, stock states and labels of the resource exchange marks; screening the resource exchange labels matched with the target user in the resource exchange label pool based on the pre-established association mapping of the image labels and the resource exchange labels, and generating a candidate label set; And filtering the candidate target set based on the resource exchange constraint features to generate a resource recommendation result.
  5. 5. The method of claim 4, wherein generating the resource recommendation after filtering the candidate target set based on the resource redemption constraint feature comprises: determining a conversion constraint condition through the resource conversion constraint characteristics, filtering the resource conversion labels of which the resources required by the candidate label set exceed the conversion constraint condition, and generating a subset of the candidate labels; and generating a resource recommendation result based on the subset of the candidate targets.
  6. 6. The method of claim 5, wherein generating a resource recommendation based on the subset of candidate targets comprises: determining the historical exchange times of each resource exchange mark in the subset of the candidate marks, and sorting the subset of the candidate marks according to the order of the historical exchange times from high to low to generate a sorted subset of the candidate marks; selecting a specified number of resource exchange labels from the subset of candidate labels as final recommendation labels; generating a fourth prompting word, and calling a large model based on the fourth prompting word to generate an explanatory text based on multidimensional constraint features and a final recommendation target; And combining the interpretable text with a final recommendation target to generate a resource recommendation result.
  7. 7. The method according to claim 1, characterized in that the method further comprises: And when the resource exchange intention is determined to be an explicit intention according to the confidence score, calling a query interface to query a target object matched with the resource exchange intention in a pool of the resource exchange object, and taking the target object as a resource recommendation result.
  8. 8. An electronic device, the electronic device comprising: At least one processor; and a memory communicatively coupled to the at least one processor; Wherein the memory stores a computer program executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.
  9. 9. A computer storage medium storing computer instructions for causing a processor to perform the method of any one of claims 1-7 when executed.
  10. 10. A computer program product, characterized in that it comprises a computer program which, when executed by a processor, implements the method according to any of claims 1-7.

Description

Resource exchange recommendation method, equipment, medium and product Technical Field The invention relates to the field of financial science and technology, in particular to a resource exchange recommendation method, equipment, medium and product. Background With the rapid development of mobile financial services, mobile banking has become a core carrier for users to transact financial services and enjoy value-added services, and resource exchange is used as an important value-added service for improving user viscosity and clawing user assets, and the service quality of the mobile banking directly influences user experience and platform liveness. The current mobile phone bank resource exchange scene mainly depends on two implementation modes, namely a passive response mode based on keyword search, a user needs to actively input explicit information such as names, categories or brands of resource exchange marks, a system returns a list of the resource exchange marks according to keyword matching, and a recommendation mechanism based on simple rules, such as displaying preset hot resource exchange marks according to a point gear, or randomly screening and sorting the display of the resource exchange marks to the user or displaying the hot resource exchange marks with a good list. The existing resource exchange technology completely depends on the user to provide clear query keywords, and when the user only expresses fuzzy requirements, effective intention recognition and guidance cannot be performed. In addition, the prior art relies on manual preset rules, cannot adapt to personalized scenes, residual resource differences and consumption habit changes of different users, and the traditional recommendation model needs a large amount of historical behavior data support, so that the effect is rapidly reduced in the scenes of new users or low-frequency exchange users, and finally, the user experience is poor and the resource exchange rate is low. Disclosure of Invention The invention provides a resource exchange recommendation method, equipment, medium and product, which are used for analyzing natural language of a user, evaluating the definition of exchange intention by combining multidimensional constraint features, and realizing labeled personalized recommendation aiming at fuzzy intention, thereby solving the technical problems of inaccurate user intention identification, low recommendation correlation, opaque rule and low exchange efficiency in traditional resource exchange. According to one aspect of the invention, there is provided a resource redemption recommendation method, the method comprising: receiving natural language input by a target user, and analyzing the natural language to obtain a resource exchange intention; Extracting multi-dimensional constraint features of a target user, and evaluating the resource exchange intention according to the multi-dimensional constraint features through an intention confidence evaluation model to obtain confidence scores, wherein the multi-dimensional constraint features comprise intention significance features, user attribute features, user behavior features and resource exchange constraint features; and when the resource exchange intention is determined to be the fuzzy intention according to the confidence score, carrying out resource exchange recommendation on the target user according to the multidimensional constraint characteristics. Optionally, analyzing the natural language to obtain the resource exchange intention, which comprises the steps of generating a first prompt word, invoking a large model based on the first prompt word to perform semantic analysis on the natural language, filtering redundant information irrelevant to resource exchange, extracting intention elements relevant to the resource exchange, integrating the intention elements, and generating the resource exchange intention, wherein the intention elements comprise action verb elements, specific pointing elements and constraint description elements. The method has the advantages that relevant effective information of resource exchange is accurately screened, redundant interference is eliminated, and pertinence and integrity of extraction of the intention elements are ensured. Optionally, evaluating the resource exchange intention according to the multi-dimensional constraint features through an intention confidence evaluation model to obtain a confidence score, wherein the method comprises the steps of carrying out normalization processing on the multi-dimensional constraint features and intention elements in the resource exchange intention to generate standard input features, generating a second prompting word, calling a large model based on the second prompting word, determining feature weights corresponding to the standard input features through a dynamic weight adjustment mechanism based on a preset association rule, wherein the second prompting word comprises the preset rule, input