Leveraging natural language processing techniques for enhanced recommender systems

Recommender systems often use NLP methods primarily for content processing. In this study, we propose a new approach to building recommender systems, in which user interaction data with content is considered within the framework of a natural language model. Thus, the user preference vectorization model Pref2Vec is proposed as the basis for a hybrid recommender system. Moreover, a concept of a user embedding space (UES) is introduced, which represents a set of extended embeddings that capture end-user preferences. A new method of applying clustering analysis to the recommendation process is also proposed. The Pref2Vec model and the UES class were implemented in the Python programming language as an extension of the functionality of the Gensim library. The model was evaluated using Recall@k and NDCG@k metrics. The comparative analysis showed that the results obtained are comparable with the performance of the BPRMF, GRU4Rec and NextItRec models, which indicates the potential of the proposed model.