Improved Neural Collaborative Filtering Recommendation Method based on Multi-Head Attention and Feature Fusion
Yan Tang, Xin Zhang, Xiu Zhang · 2023
Recommender systems play an important role in personalized recommendations, and neural collaborative filtering models have achieved some success as an effective recommendation method. However, neural collaborative filtering has limitations in feature representation capabilities and interaction relationship modeling, which limit their recommendation accuracy and personalization capability. In order to address this problem, this paper proposes an improved neural collaborative filtering method, called MHAN-MF. In this method, we propose two improvements: (1) by concatenating the user embedding vector, item embedding vector, and the element-wise product of the user and item embedding vector, the concatenated vector is used as the input feature of the Multi-Layer Perceptron(MLP), thereby enriching the input features of the model and improving its expression ability; (2) the introduction of multi-head attention enables the model to focus on different parts of the input features from different perspectives, to better capture relevant features in the data, and to improve the ability to model complex interactions. We experimentally validated the MHANMF method on publicly accessible datasets. The experimental results show that the MHAN-MF method significantly outperforms other methods in terms of recommendation effectiveness, which verifies the effectiveness of our proposed method.