Generating Explanations for Explainable Recommendations Using Filter-Enhanced Time-Series Information

Generating explanations for recommended items is crucial in recommender systems, as it helps users understand how the recommendations align with their preferences, thereby enhancing user satisfaction. Typically, these explanations are produced using natural language generation. However, existing methods often rely solely on item reviews and IDs, ignoring critical historical user behaviors such as previous purchases and sequences, which are essential for improving the effectiveness of recommendations and user satisfaction. To address this issue, we propose a Transformer-based method designed to generate explanations by leveraging time-series information extracted through Transformer-based sequential recommendation. This approach not only captures the temporal dynamics of user interactions but also assigns linguistic meaning to the relationships between time-series information and recommended items, thereby enriching the explanations for recommended items. Additionally, we designed a filter layer that attenuates the noise in the frequency domain of the time-series information, to maximize the benefits. Extensive experiments on three datasets demonstrated that, in most cases, the proposed method generates explanations that are both reasonable and effective compared to state-of-the-art explanation generation methods. Further experiments and analyses have verified the effectiveness of this approach.