Disentangled Commodity Recommendation via Multi-Feedback Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing recommendation systems fail to accurately capture user interests from multi-feedback scenarios due to homogeneity in learned user interests, lack of interpretability, and low accuracy in recommending commodities, especially when dealing with noise in multiple feedbacks.
Innovation Solution
A disentangled commodity recommendation method that filters and aggregates user representations from clicked, unclicked, and disliked commodity sequences using a multi-head attention mechanism, weight assignment based on behavioral time information, and clustering to determine interested commodities, improving recommendation accuracy and click-through rates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If user interests are obtained only from positive feedback (clicked commodities), then the recommendation system is simple to implement, but the learned user interests become homogeneous and simplified, affecting click-through rate
Solution Approach 1:
The patent segments user feedback into multiple categories: clicked commodities, unclicked commodities, and disliked commodities. This segmentation allows the system to process different types of user interactions separately, capturing diverse user interests rather than relying solely on positive feedback, thereby improving user interest accuracy while maintaining implementation feasibility
Solution Approach 2:
The patent introduces an interest filter mechanism as an intermediary that processes multiple feedback types. This filter acts as a mediator to identify user interests from various feedback sources (clicked, unclicked, disliked commodities) and weights them appropriately, improving the accuracy of user interest representation without significantly complicating the system architecture
2Quantity of substance
If existing multi-feedback recommendation technology is used, then more user feedback is utilized, but the system cannot disentangle user interests, lacks interpretability, and has low recommendation accuracy
Solution Approach 1:
The patent extracts disentangled user interests from multiple feedback types by separating and analyzing different feedback sources (clicked, unclicked, disliked commodities) independently. The interest filter mechanism extracts meaningful user interests from each feedback type and combines them with appropriate weighting, enabling the system to utilize diverse feedback data while maintaining high recommendation accuracy and interpretability
Solution Approach 2:
The patent applies different processing methods and weighting strategies to different feedback types. Specifically, the system treats clicked, unclicked, and disliked commodities differently through the interest filter mechanism, assigning different weights and processing methods to each feedback type based on its informational value, thereby improving overall recommendation accuracy while maintaining system interpretability
3Ease of operation
If all clicked and unclicked commodities are treated equally, then the processing is simple, but noise in multiple feedbacks cannot be filtered, reducing recommendation quality
Solution Approach 1:
The patent changes the parameter of feedback weighting by introducing an interest filter mechanism that dynamically weights different feedback types. The system adjusts the weight of clicked, unclicked, and disliked commodities based on their informational value and relevance to user interests, effectively filtering noise while maintaining reasonable processing complexity and improving recommendation quality
Data Source
AI summary
A disentangled commodity recommendation method includes: receiving information of commodities to be recommended and historical behavioral information of a user, the historical behavioral information including a clicked commodity sequence, an unclicked commodity sequence, a disliked commodity sequence and behavioral time information of the user; filtering the clicked commodity sequence and the unclicked commodity sequence according to the disliked commodity sequence to obtain representations of interested commodities of the user; filtering the representations of the interested commodities according to the behavioral time information of the user and the information of the commodities to be recommended to obtain representations of historical interested commodities of the user; clustering and aggregating the representations of the historical interested commodities to obtain a plurality of disentangled representations of the user; and determining whether the commodities to be recommended are the interested commodities of the user according to the plurality of disentangled representations.