User Tag Generation via Multi-Level Feature Merging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for generating user tags based on statistical user profiles struggle to accurately determine sparse behaviors and interests, resulting in low accuracy of user tags.
Innovation Solution
A method and apparatus that obtain discrete user data from multiple feature fields, merge intra-field features to form inter-field features, perform feature crossing on sub-features, and select user tags based on these features to enhance tag accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If statistical user profiles are used to generate user tags, then the generation process is simple, but the accuracy of user tags is low
Solution Approach 1:
The patent segments the feature processing into multiple distinct stages: obtaining discrete user data from multiple feature fields, generating intra-field features for each feature field separately, merging these intra-field features to create inter-field features, and finally performing feature crossing to generate user tags. This segmentation allows each stage to be optimized independently, improving overall accuracy while managing complexity through structured organization.
Solution Approach 2:
The patent merges features at multiple levels: first merging discrete user data within each feature field to create intra-field features, then merging these intra-field features across different feature fields to create inter-field features, and finally merging these inter-field features through feature crossing to generate comprehensive user tags. This multi-level merging approach ensures that sparse behaviors from different sources are adequately combined to improve tag accuracy.
2Measurement precision
If discrete user data from multiple feature fields is merged and feature crossing is performed, then the accuracy of user tag generation is improved, but the computational complexity increases
Solution Approach 1:
The patent performs preliminary actions by first generating intra-field features from discrete user data within each feature field before merging them. This preliminary feature generation organizes and pre-processes the data in a structured manner, reducing the complexity of subsequent merging and feature crossing operations. By preparing features in advance at the intra-field level, the system manages computational complexity while maintaining accuracy improvements.
3Measurement precision
If intra-field features are merged to obtain inter-field features, then sparse behaviors are adequately merged, but the data processing time increases
Solution Approach 1:
The patent segments the data processing into distinct feature fields, processing and merging data within each field separately to generate intra-field features before combining them into inter-field features. This segmentation allows for more efficient processing by organizing sparse behaviors into manageable groups, reducing overall processing time while maintaining accuracy in determining user characteristics.
Data Source
AI summary
This application relates to a user tag generation method performed by a computer device, relating to the field of neural networks. The method includes: obtaining discrete user data corresponding to a target user identifier in multiple feature fields respectively; for each feature field, obtaining an intra-field feature corresponding to the target user identifier according to the discrete user data in the feature field; merging the intra-field features to obtain an inter-field feature corresponding to the target user identifier; performing feature crossing on sub-features in the inter-field feature to obtain a cross feature corresponding to the target user identifier; and selecting, from candidate user tags, a target user tag corresponding to the target user identifier according to the inter-field feature and the cross feature. The solutions provided by this application can improve the accuracy of generating user tags.