LSTM-Based Recommendation Approach for Interaction Records

Interactive platforms such as Spotify and Steam currently play an increasingly important role on the Internet. Users continuously use the content on these platforms. Therefore, the most important data in interactive platforms are interaction records, which contain an enormous amount of information regarding user interests at any given time. However, previous recommendation approaches have been unable to process such records satisfactorily. Therefore, we propose an LSTM-based recommendation approach for interaction records. In our approach, we used a recurrent neural network (RNN) based on LSTM to make recommendations by learning user interests and their changing trend. We propose a pretreatment called serial filling at equal ratio to apply LSTM. Further, we used a dimensionality reduction technique based on matrix factorization to improve the system efficiency. Finally, we evaluated our approach using Steam datasets. As indicated by the results, our approach performs better than other conventional approaches in three aspects: Accuracy, efficiency, and diversity.