Recommendation Model Training for Cold-Start User-Object Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing recommendation methods fail to provide accurate recommendations for users or products without historical interaction behaviors, such as newly released products or newly registered users.
Innovation Solution
A model training method involving the establishment of association relationships between sample users, objects, and their features, using graph convolutional networks to extract collaborative and content features, and training networks with training labels to determine accurate recommendations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If collaborative filtering algorithm is used based on historical interaction behaviors, then recommendation accuracy for users with history is improved, but recommendation capability for users or products without historical interaction behaviors deteriorates
Solution Approach 1:
The patent pre-establishs three types of association relationships (user-object interaction relationships, object similarity relationships, and user preferences relationships) during the model training phase. This preliminary structuring of data relationships enables the system to handle both cold-start scenarios and traditional recommendation scenarios without requiring separate handling logic, thereby resolving the contradiction between accuracy for historical users and adaptability for new users/products.
Solution Approach 2:
The patent introduces association relationships as intermediary structures that mediate between users and objects. These association relationships (including direct interactions, similarity-based connections, and preference-based connections) serve as bridges that enable recommendation for both users with and without interaction history, thus resolving the contradiction between maintaining accuracy for historical users and providing capability for new users/products.
2Measurement precision
If multiple types of association relationships and feature networks are established, then recommendation accuracy is improved, but system complexity increases
Solution Approach 1:
The patent segments the recommendation system into distinct functional modules: an extraction network for feature extraction, a first feature network for processing user-object interaction features, and a second feature network for processing object similarity features. Each network has a specific function and processes specific types of data, which reduces overall system complexity by making each component simpler and more specialized while maintaining high recommendation accuracy through their coordinated operation.
Solution Approach 2:
The patent merges multiple types of association relationships (user-object interactions, object similarities, user preferences) into a unified training framework that uses a single loss function to optimize all relationships simultaneously. This merging approach reduces complexity by consolidating what could be separate systems into one integrated model, while still capturing the nuanced differences through distinct feature networks.
Data Source
AI summary
A model training method, a recommendation method, and a search method are provided. A first feature network and a second feature network are trained based on first association relationships between sample users and second sample objects with which the sample users have interactive behaviors, second association relationships between first sample objects and second sample objects that satisfy a similarity condition with the first sample objects, third association relationships between the first sample objects and the sample users, and training labels of whether the sample users have interactive behaviors with the second sample objects and the first sample objects respectively. The first feature network extracts collaborative features of a target user, and the second feature network extracts content features of a target object. A matching result of the content features and the collaborative features is used to determine whether to recommend the target object to the target user.