基於使用者影響力之群體書籍推薦

Abstract

隨著網路的發展，人們不再是被動的接受資訊，而是能在社群網站上分享自己的意見與想法，使用者對事物的喜好也可能因其他使用者的意見而被影響。另外，有著相似喜好的使用者們時常會集結成團體並共同分項喜愛的事物，而使用者偏好也可能被群體中的成員影響。但現有的推薦系統研究多著重於分析個人喜好，且對群體推薦的方法也很少考慮成員之間彼此的影響力。 因此，本研究以外國網站Goodreads為資料集來源，該網站為以書籍閱讀並分享喜好(評分)為目的的社群網站，本研究提出一個新穎的分析群體喜好的方式，基於分析成員的三種影響力，分別是群體成員影響力、評論影響力及推薦影響力。另外，我們採用LDA來分析書籍的主題特徵。我們所提出的方法整合了群體成員的三種影響力及上述的書籍主題特徵，來加以建立群體主題特徵檔。最後，我們提出的方法整合了成員們對書籍的評分，以及書籍與群體間主題特徵計算而來的相似度，加以預測群體對於書籍的喜好分數來做推薦。由最後的實驗結果顯示我們提出的方式確實能改善推薦成效。

With the development of internet, users not only receive information passively but also share their own opinion and thinking in the social network. Accordingly, users’ preferences for items may be affected by other users through opinion sharing and social interactions in the network. Moreover, users with similar preferences usually form a group to share items with each other, and thus users’ preferences may be affected by group members. Existing researches often focus on analyzing personal preferences, and group recommendation approaches considering the influences of group members are relatively few. In this work, we investigate group recommendation approaches for recommending books of the website-Goodreads, which is a social network website for sharing interests (ratings) and opinions to books. We propose novel group recommendation approach by analyzing groups’ preferences based on three types of group member influences - Group Member Influence, Review Influence and Recommendation Influence. LDA (Latent Dirichlet Allocation) is adopted to derive the latent topic features of book contents. The proposed approach then integrates member influences to derive a group’s topic profiles from the topic features of books collected by the group. Finally, the proposed approach integrates members’ ratings on books and similarity measures between book topic features and group topic profile to predict the group’s preference scores on books. The experimental results show that our proposed approach can effectively improve the quality of recommendations.