以多社群為基礎之部落格文件推薦

Abstract

以協同過濾與社群技術為基礎之知識文件推薦可有效達到知識分享。本研究主要以多社群的概念，提出一個適用於網路上的部落格知識文件推薦機制，根據社群內成員所感興趣的不同知識領域，推薦使用者各知識領域的部落格文件，以符合使用者多元興趣的需求，並達到Web 2.0網路概念之知識分享。 所提機制主要分成兩大階段，第一階段是多社群分析，考量使用者多元的興趣，建立不同領域的興趣社群，每個使用者可以參與一個以上的興趣社群；第二階段是結合多社群與協同過濾之部落格知識文件推薦。本研究提出四個以多社群為基礎之部落格文件推薦方法，分別為參與社群最大分數方法(PCMM)、主要社群方法(PCM)、使用者權重方法(UWCM)以及社群權重方法(CWM)。推薦方法中的使用者相似度計算是根據使用者特徵檔(使用者所感興趣的文件內容)，本研究提出主題性使用者特徵檔的概念，根據興趣社群中使用者撰寫過的部落格知識文件，產生屬於不同社群領域的動態使用者特徵檔，以計算出在各不同社群領域中的使用者相似度。本研究探討整體性的使用者特徵檔以及主題性的使用者特徵檔是否會產生不同的推薦結果。實驗結果顯示，以多社群為基礎之部落格文件推薦方法優於傳統之協同過濾推薦方法，且以主題性使用者特徵檔進行推薦可以提升推薦成效。

Document recommendations based on collaborative filtering and community techniques have been applied to promote knowledge sharing effectively. This study proposes two stages to implement a knowledge sharing platform. First stage is to create communities automatically based on user interests on documents. A community represents a knowledge interest domain. Generally, a user can join more than one community to form a multi-community environment. The second stage is to provide blog document recommendation based on collaborative filtering and multi-communities. This study proposes Blog document recommendation methods based on multi-communities, including Participated Community Max Score Method (PCMM), Primary Community Method (PCM), User Weight in Community Method (UWCM) and Community Weight Method (CWM). User profiles, which represent the interests of users and are created from documents accessed by users, are used to compute the similarity among users. Two users may have different similarity measures in different communities, since they have different subject-based user profiles in different communities. This research considers the subject-based user profiles in the four methods to discuss the effect on recommendation quality. The experimental results show that the proposed methods based on multi-communities outperform traditional methods. Moreover, recommendations considering subject-based user profiles can improve the effectiveness of recommendations.