本論文提出一套在非結構同儕網路上以特徵相似度為基準之搜尋方法，此方法設計上以構成物件名稱的關鍵字為該物件之特徵，同儕之間利用自身所擁有物件的特徵，計算和其它同儕之間語義上的距離，建構一個語義性重疊網路(Semantic Overlay Network；SON)，縮短具有共同興趣的同儕在語義網路上的距離，使得詢問訊息能有目標地在一個局部範圍內傳遞，減少詢問訊息傳送所產生的網路流量負載(traffic overhead)。 此方法也同時考量同儕系統內物件具有不同的冷門或熱門的特性，提出以特徵相似性來決定詢問訊息傳送的對象的同儕評比機制，並且搭配快取檢索機制來促進熱門請求物件的傳送，最後，本研究整合上述之方法提出一套可適性並行搜尋演算法，模擬結果顯示在靜態語義重疊網路上，並行搜尋演算法能達到與Flooding方法相近的搜尋命中率，並大幅減少詢問訊息數量。 關鍵字：同儕網路, 搜尋相似度, 語義重疊網路, 同儕評比This paper proposes a keyword-based search approach based on search similarity in unstructured peer-to-peer networks. Peers extract the keyword terms from object names as the objects’ features. With the feature similarity among peers’ objects, they compute semantic distance between any two peers, and then collectively construct a semantic overlay network (SON). Peers with common access/request interests can be relatively close with shorter distances. This property can used to guide the search requests to be routed in semantic vicinity where target objects could be found, and thus reduce traffic overhead to some extent. Consider the property of object popularity, i.e., popular vs. unpopular objects. This study designs a peer ranking mechanism based on the feature-based similarity which can be used to route queries efficiently and also proposes a cache index mechanism which can improve the forwarding of queries for popular objects. With all above, this paper therefore designs an adaptive parallel search approach. Simulation results show that the proposed search mechanism can not only improve the hit rate close to that by the flooding approach, but also significantly reduce the amount of query messages. Keywords: Peer-to-peer networks, Search similarity, Semantic overlay network, Peer ranking
