327 / 2024-04-24 14:45:11

Mitigating Exposure Bias in Recommendation Systems Using a Neural Collaborative Filtering Model Based on Reward Feature Integration

Neural collaborative filtering,Linear upper confidence bound,Exposure bias,Personalized recommendation

To address the issue of strong exposure bias caused by the sparsity of interaction data and uneven exposure in recommendation systems, this paper proposes a de-biasing model that combines neural collaborative filtering with the Linear Upper Confidence Bound (LinUCB) algorithm. Firstly, by analyzing the interaction data between users and items, the neural collaborative filtering algorithm is used to learn the characteristics of users and items and to capture their latent preferences. Secondly, the LinUCB algorithm is introduced, and its generated reward features are embedded into the neural collaborative filtering model to enhance the model's exploration capabilities for items with low exposure, thereby effectively mitigating exposure bias. Finally, experiments conducted on the MovieLens-100K and MovieLens-1M datasets demonstrate that the model not only improves recommendation accuracy but also significantly reduces exposure bias.