Massive online reviews express users' opinions and attitudes towards books and their aspects (e.g. content, printing etc.), thus reflecting the impacts of books. This paper conducted multi-level fine-grained mining of online book reviews to evaluate the impacts of books. This paper first collected books' online reviews. Then, fine-grained book review mining was conducted to get evaluation metrics, and multiple text representation methods were compared and domain sentiment lexicons were constructed to improve the performance of metric mining. Finally, evaluation metrics were integrated to identify book impacts. The experimental results show that although there are disciplines differences, online book reviews can be used as an effective resource for book impact assessment. In addition, in this corpus, the performance of one-hot text representation is superior to that of topic model based methods and the deep learning based method. Meanwhile, sentiment lexicon with fuzzy sentiment words can effectively improve the performance of sentiment classification.