An explainable recommendation algorithm based on content summarization and linear attention

Jun Ai, Haolin Li, Zhan Su, Fengyu Zhao

Abstract: Recommendation algorithms can alleviate the problem of information explosion and cater to the needs of users to quickly lock in preferred items, promote business development, and have important theoretical significance and broad theoretical value. Explainable recommendation algorithms can not only complete recommendation tasks, but also generate recommendation explanations, so that users can more easily accept preferences. Research related to natural language text generation has promoted the progress of explainable text generation technology for recommendation systems. This paper proposes an explainable recommendation algorithm based on content summarization and linear attention mechanism. The model uses the keyword extraction algorithm to extract key information from user comment text as an important feature of subsequent text generation tasks, and further introduces linear Transformer to improve the training speed of the model and enhance its scalability. In addition, the model also uses the Vivaldi synthetic coordinate algorithm to deeply mine user and item features and uses the Kolmogorov–Arnold neural network model to reduce the error of predicted ratings. Compared with existing leading algorithms, the algorithm in this paper has achieved significant improvements in text generation and recommendation rating prediction. This paper reveals that applying the linear attention mechanism to the explainable recommendation algorithm can greatly reduce the training cost and improve scalability, and the fusion of synthetic coordinates and attention can further mine the hidden information of the recommendation system, effectively improving the performance of the recommendation algorithm.