Adaptive Negative Sample Generation for Sequential Recommendation Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Sequential recommendation models face challenges in accurately predicting user preferences due to changes in negative interests over time, as existing methods often use uninformative random sampling of negative items, leading to inaccurate learned user preferences.
Innovation Solution
The Generating Negative items (GenNi) approach adaptively generates negative samples based on the current next item prediction of the model, using similarity between user interests and item embeddings, without requiring additional generative modules, thus improving training accuracy and reducing computational costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If random sampling of negative items is used, then the method is simple to implement, but the quality of negative samples is poor leading to inaccurate user preference prediction
Solution Approach 1:
The patent changes the sampling distribution parameter from uniform random sampling to non-uniform sampling based on user interest similarity. Items are sampled according to their similarity score to the user's current interest representation, where items with lower similarity (more dissimilar items) have higher sampling probability. This parameter change transforms the sampling process from random to interest-based, resolving the contradiction between implementation simplicity and prediction accuracy.
Solution Approach 2:
The patent replaces the mechanical random sampling process with an interest-based sampling mechanism. Instead of using pure randomness, the system uses user interest representations and item embeddings to compute similarity scores, which then drive the sampling process. This substitution maintains computational efficiency while significantly improving negative sample quality for accurate preference prediction.
2Measurement precision
If additional generative modules are added to generate negative samples, then negative sample quality improves, but device complexity and computational cost increase
Solution Approach 1:
The patent makes the existing sequential recommendation model multi-functional by enabling it to perform both positive item prediction and negative sample generation using the same model components. The encoder and sampler modules that originally handled positive predictions are reused to generate negative samples by sampling items with lowest similarity scores. This eliminates the need for separate generative modules while maintaining high negative sample quality.
Solution Approach 2:
The sequential recommendation model serves itself by using its own encoder and sampler components to generate negative samples. The model's existing interest representation and item embedding mechanisms are leveraged to create informative negative samples without requiring external generative systems. This self-service approach reduces overall system complexity while improving negative sample quality.
Data Source
AI summary
Embodiments described herein provide methods and systems for training a sequential recommendation model. A system receives a plurality of user behavior sequences, and encodes those sequences into a plurality of user interest representations. The system predicts a next item using a sequential recommendation model, producing a probability distribution over a set of items. The next interacted item in a sequence is selected as a positive sample, and a negative sample is selected based on the generated probability distribution. The positive and negative samples are used to compute a contrastive loss and update the sequential recommendation model.