| With the quick development of the Internet and technology, e-book and author resources become rich and colorful. But it becomes more and more difficult to find personal interests in the vast amounts of resources. Hence, recommendation system can help readers find valuable information more quickly. In our daily life, readers tend to read the same books and the similar books close to the writing style or contents of their favorite author. Personal book recommendation and author recommendation have become hot concern in online e-book market. Therefore, the research of content-based e-book and author recommending has important practical value. The main work is as follows.As to the research of content-based e-book recommendation, traditional text processing algorithms deal more with survey of short text than long text. Because compared with short text such as news, long text such as e-book has higher dimension, more complex processing and more difficult measurement of word semantic relationships. This thesis uses the crawlers to get the experimental dataset from the authoritative website. As for the characteristic of high dimension and complex processing, the thesis adopts the idea of partition and divides e-book text into several parts. By building words relationship mapping matrix, the thesis puts forward the model of multi-dimensional latent semantic algorithm to express semantic relevance between words. According to the difficult measurement of word semantic relationships, the thesis applies the fusion distance of global semantic and local semantic to measure the similarity of different e-book contents. Then, the thesis studies the experiment of the involved parameters to achieve better recommendation results. The experimental results show that under the measure of five quantitative indicators, the multi-dimensional latent semantic analysis model is better than other traditional text processing algorithms.As to the research of content-based author recommendation, most survey focuses on the academic expert’s recommendation and uses the single characteristics. As for the matter above, this thesis uses the crawlers to gain the three related heterogeneous characteristics, namely author biography, book introduction and user comment. After data preprocessing, the thesis tackles the tree-structured representation for three characteristics of an author. Then this thesis applies multilayer self-organizing mapping algorithm that server as a clustering technique to handle author recommendation. The thesis designs two experiments and discussion the involved experimental parameters. The experimental results show that under the measure of five quantitative indicators, the model of multilayer self-organizing mapping based on author tree is superior to the traditional text processing algorithms. |
