Construction and Analysis of Digital Library User Interest Model based on Word2vec and LDA

Digital library methods often rely on traditional content-based recommendation models or collaborative filtering-based recommendation algorithms, which have problems such as inaccurate capture of user interests and difficulty in deeply understanding semantic information. Therefore, this paper proposes a method that combines Word2Vec and LDA (latent Dirichlet allocation model). First, this paper uses the Word2Vec model to learn word vectors for texts such as search queries and book titles in user behavior data, thereby capturing the semantic similarity between words; then, topic modeling is performed based on the LDA model, and the topic distribution is optimized with word vectors as input, thereby accurately identifying users' interest areas and preferences. Experimental results show that the Word2Vec+LDA model performs well in various indicators, with Top-1 accuracy reaching 0.65, Top-3 accuracy reaching 0.72, Top-5 accuracy reaching 0.78, Top-10 accuracy reaching 0.83, and average recommendation accuracy reaching 0.75.