Abstract: The library’s personalized recommendation system emphasizes recommendation accuracy,
which cannot meet readers’ need for diversity. In this paper, we introduce the deep learning algorithm
into the library recommendation system, and discuss the problem of the recommendation diversity. First,
according to the historical checkout data, we form the reader’s borrowing behavior co-occurrence matrix
combined with time series; then, the co-occurrence matrix is regarded as context, and we identify the
reader’s potential interest by using the potential semantic analysis of Word2vec; finally, we identify the
reader’s interest and provide varied recommendation results. In this paper, we use more than 5 410 000 data
from the Pudong Library in Shanghai to experiment. Comparing to association rule algorithm, we find that
our method has a good impact on the recommendation diversity.

Key words: Word2vec, Library recommender system,, Diversity