以社群理論為基礎建構於點對點網路中之音樂推薦系統

Abstract

近幾年來點對點系統日益普及以及應用廣泛，使得多媒體檔案的分享越來越有效率。但在這個資訊爆炸的時代，讓使用者逐一搜尋自己有興趣的檔案將會花費越來越多的時間，而推薦系統能夠在大量的選擇中，根據使用者的興趣挑選出符合品味的檔案，幫助使用者更有效率的獲取所需。 大部分現存的推薦系統是集中式的架構，然而這並不適用於分散式的點對點系統。因此在這篇論文中，我們提出了一個分散式音樂搜尋推薦系統。為了讓使用者在只有部分資訊的情況下能有效率的搜尋和推薦，我們建立了以社群理論為基礎的疊代式網路，將系統中對音樂有相同喜好的使用者聚集在一起。 此外，我們利用一個特徵向量來代表系統中的一首歌，並利用此向量建立一個內容式的過濾機制。再者，我們在所有的特徵中選擇一個主要的屬性，代表使用者聆聽音樂的興趣，藉此提出一個協同式的過濾方法。最後，我們觀察使用者的下載行為建立合作式的過濾機制。我們利用名為AudioScrobbler網站上的使用者資料做模擬實驗，這個網站記錄了使用者聆聽音樂的喜好。實驗結果證實我們的系統能夠讓使用者快速的搜尋到想要的物件，同時提供也比現存的推薦系統更佳的推薦。

The pervasive use of Peer-to-Peer (P2P) systems and the growing demand for personalization from the consumers has made future business focus on the niche market instead of the mass market. The recommender system, which is able to timely select interested data to the individual user, has become the key to any successful business. Currently most recommendation systems are based on a centralized architecture, none the less, this is not suitable for P2P networks. The focus of this paper is to propose a distributed system for music search and recommendation on unstructured P2P networks. The idea of our work is to construct a social-based overlay network that can cluster a small set of peers, which have similar tastes for music, from thousands to millions of peers. That is, peers with similar interests can be connected by shorter paths so that they can exchange multimedia content more efficiently. In addition, we choose a set of proper meta-data (a characteristic vector) to represent a music object and use them to construct the characteristic-vector-based content filter. Next, a dominant attribute, which is one of the attributes in the characteristic vector of a music object, is used to build the profile of a peer. With the idea of the social network, a P2P profile-based collaborative filter is proposed. Finally, we explore the item-to-item relationship to construct a history-based cooperative filter. We use simulations and a real database called AudioScrobbler which tracks users' listening habits, to evaluate the performance of the recommendation system. The results show that our system is able to offer efficient query and significant improvement for recommendation services compared with existing recommendation systems.