Personalized recommendation cold start method and system based on prompt learning

Through the method based on prompt learning, using verb and pre-trained language model to process user and project features, the problem of low recommendation accuracy when new users are cold-started is solved, and more efficient personalized recommendation effect is achieved, especially in the BERT model Significant improvements have been made under application.

CN118964747AInactive Publication Date: 2024-11-15HUNAN NORMAL UNIVERSITY
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202411445468.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-16
Publication Date
2024-11-15
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing recommendation system lacks user-project interaction data when new users are cold-started, resulting in low accuracy and poor results in personalized recommendations, which cannot effectively solve the cold-start recommendation problem.

Method used

Using a prompt learning method, by obtaining the attribute information of users and items from the input recommendation set, normalizing the verb to user characteristics and item characteristics, and filling them into the prompt word template containing the mask, input pre- The trained language model predicts recommendation results and improves the accuracy of cold-start recommendations.

Benefits of technology

The performance and accuracy of cold start recommendations are improved, and the personalized recommendations are enhanced through information from large models and prompt templates, especially when using the BERT model, the optimal GAUC score is obtained.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118964747A_ABST
    Figure CN118964747A_ABST
Patent Text Reader

Abstract

The invention discloses a personalized recommendation cold start method and system based on prompt learning. The method comprises the following steps: S1, acquiring attribute information of a user and an item from an input recommendation set; s2, normalizing the input configuration files of the user and the project into user features and project features by using a verb device; s3, the user features and the item features are filled with cue word templates containing masks to supplement contexts of users and items, and cue words containing user feature information, item feature information and masks are obtained; and S4, inputting the cue word containing the mask into a pre-trained language model, and outputting a final recommendation result according to the probability of predicting each word in a given target vocabulary set by the pre-trained language model. According to the method, the performance of cold start recommendation is improved, and the accuracy of personalized cold start recommendation is improved according to the information of the large model and the prompt template.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Recommendation reason generation method and device, equipment and storage medium

    CN116720004A

  • Setting item processing method and device, electronic equipment and storage medium

    CN117424956A

  • Multi-task recommendation method and device based on user evaluation information, terminal and medium

    CN118260411A

  • Commodity information processing method and commodity searching method

    CN118537091A

  • Method of training information generation model, method of generating information, and device

    US20230075339A1