Dynamic Hybrid Recommendation Model Construction and User Discovery Research

This study tackles keyword dependency and latency in book recommendations via a hybrid model fusing collaborative filtering with matrix factorization. The results were used to build an intelligent book recommendation system (recommendation center) with a web interface. Traditional library book recommendation systems rely primarily on users actively searching for the titles they need. Their limitation lies in the large number of matching titles that appear when keywords are entered. In our study, to further improve recommendation accuracy, the recommendation system addresses the problem of professional cognitive limitations users face when making choices through similarity calculations. Furthermore, a user search module is added to the recommendation system to ensure accurate recommendations. When generating a recommendation item for a user, the recommendation system first searches related sentences, then calculates the similarity between the target user and the related user and uses the similarity value as a weight. Finally, based on the previously calculated similarity value, it performs a weighted average of the differences between all ratings. A time-decay clustering algorithm (λ = 0.85) using multi-source data achieves 41 % increase in user similarity and 35 % in new book discovery for a user. The tests demonstrate 27 % increase in accuracy with 3300 concurrent requests (5s/300ms response).