| With the continuous development of the mobile reading industry,more and more users are used to read books through mobile reading.Facing the tremendous amount of e-books online, how to help users quickly find what they really want and overcome the difficulty in selections has become a new demand in mobile reading industry.This thesis originates from the actual requirements of CMCC (China Mobile Communications Corporation) mobile reading platform, aims to reform the previous recommendation system, reduce request delay, and solve the"cold-start" problem of new items in the recommendation system.From the perspective of book’s own properties,such as title, abstract,introduction and book’s marketing parameters, such as order rate, transfer order rate, this thesis splits the original book-similarity score into the following four parts: editor score, base score, similarity score and collaboration score. Because different scores have distinct time complexity, data quantity and demand of time delay, this thesis uses different algorithms to calculate different scores. Then the weighted sum of the four scores will be the final book-similarity score. This method improves the personality and accuracy of the recommendation results and solves the "cold-start" problem of new items in the recommendation system. At the same time, in order to guarantee the recommendation system can return the results in time and process the data reliably. This thesis uses different open-source big-data frameworks to deal with different data modules which improves the efficiency of the processing and the real-time of the recommendation.Finally, this thesis validates the recommendation system through A/B test by using many common measures, such as accuracy and diversity to guarantee that the recommendation algorithm works efficiently. Now the recommendation algorithm in the thesis has been deployed in app MIGU mobile reading platform and brings good economy benefit. |
PDF Full Text Request