支援語義搜尋之自組式同儕網路

Abstract

在最近幾年裡，同儕網路已成為網路上檔案管理與分享平台的重要技術之一。而隨著網路資訊的急遽累積，如何在網路上快速有效的搜尋相關文件以符合使用者的要求變的更加重要。然而，在同儕網路中，資訊是由大量的電腦主機共同管理，如何以關鍵字搜尋文件且避免浪費過多電腦資源已是一個很大的挑戰，要進一步搜尋語義相關的文件將更為困難。針對此問題，我們設計一種自組式同儕網路，讓網路中每個節點觀察所有經過的信息，不斷學習與替換所管理的檔案以及其他同儕的資訊，藉此讓自己所管理的檔案能逐漸偏向某一類相關的資料。 除此之外，整個同儕網路也能逐漸演進並具有small world的特性，使搜尋更加快速準確。有別於其他以DHT為基礎的同儕網路像是pSearch及SSW，我們的方法是採用Freenet的概念讓網路能自我組織。我們進行了大量的實驗，並且與pSearch做比較。實驗結果顯示我們的系統能有效的自我的組織，讓搜尋能以更快的速度得到比pSearch更精確的資料。

P2P network has become a popular platform for distributed file sharing and management in recent years. As the information accumulated in the network growing exponentially, targeted content-based search is becoming a necessary feature for P2P networks to be viable. However, because information in a P2P system is usually dispersed among a large number of peers, providing efficient keyword-based search is already challenging, not to mention supporting semantic search which allows users to query information using high-level concepts. To address this challenge, we propose a self-organizing P2P network, by letting each peer continuously observe query and response information passed by and update information about related documents and other peers gradually. Because each peer also keeps several references to other peers that hold completely unrelated documents, the resulting network can gradually evolve into a small-world network where a query can be routed to the set of related peers quickly following very simple greedy strategy. Our method adopts some of the key principles from Freenet, and is architecturally different from other DHT-based P2P networks such as pSearch and SSW. We conduct extensive simulation and compare our approach with pSearch. The simulation results show that our network keeps improving itself, and can achieve better search quality even with a smaller number of search steps when compared to pSearch.