Bookle ： 書籍推薦系統 - 基於讀者的評論與閱讀行為

Abstract

推薦系統就好比是一種獨特且專門的資訊搜尋引擎，在本篇論文中，我們嚐試從新的角度來開發書籍推薦系統，傳統的書籍推薦系統，大多透過記錄過去使用者的購買行為來做推薦，或是要求使用者以填充特定資料的方式來過濾搜尋結果，而我們的做法是藉由使用者描述書籍內容的方式做為搜尋依據，企圖找出更能貼近讀者需求的書籍。 在實作上，我們收集眾多的書籍評論作為基礎資料，藉由資料探勘與資訊擷取技術，找出評論中的關鍵資訊，統整這些資訊後，進而推測出各書籍的主要內容，更進一步的，為了評斷每篇評論的可信度與價值，系統利用閱讀行為衡量評論品質，讓各篇評論獲得應有的價值，讓推薦的結果更具說服力。

Recommender system is like a search engine for specific and specialized information. In this paper, we attempt to develop a book recommender system from a new perspective. Most of the traditional book recommender systems are developed through the records of purchasing behaviors of users, or specific words user are required to key in for filtering search results. However, what our system needs is the description about the book content that users are interested in. The purpose of our work is attempting to find the best books that meet the needs of readers. In the implementation of the project, we collected a large amount of book reviews as source data. With data mining and information retrieval technology, we planed to find the critical information out from reviews and integrated the information, and then the main contents of books can be speculated. Furthermore, in order to determine the credibility and value of each review, the system would measure review quality by reading behaviors. So that the reviews would be correctly scored which makes the result of recommendation become more convincing.