合作式網站推薦系統

Abstract

網路開啟了另一個資訊交流的大門，但網路資料量的龐大，讓資料尋找與過濾變得困難。事實上在許多領域上也有類似問題，例如新聞、書籍等等；所以最近幾年發展出一種新的系統，其目的為篩選出特定領域上我們喜愛的事物，這類型的機制被稱作為過濾系統或是推薦系統。 本研究嘗試將推薦系統應用在網站上，欲藉由此機制幫助使用者尋找喜愛的網站，同時盡力解決一般推薦系統所遇到的問題。其中取得評比方面，採用網路代理伺服器來記錄使用者瀏覽資訊；同好度計算方面以類別代替網站作為評比單位，並佐以網站分類模組；最後以模擬實驗來驗證本模型的預測準確度高於傳統模型。本研究的貢獻在於：創造新型態的網站推薦方式，解決傳統推薦系統所遇到的問題，並且達成了提高預測準確度的目標。

The Internet has opened up a new way for knowledge exchange, but it is difficult to find and filter knowledge because the quantity is very large. In fact, this problem exists in many domains, for example, news articles and books. So a system been developed in recent years and its purpose is to point out what we want in a particular filed. This kind of mechanism is named Filtering System or Recommender System. In this thesis, we try to apply Recommender System in web sites to help users finding their favorite web sites, and to solve the problems of traditional model. In the rating aspect, we use proxy sever as the recorder to collect user browsing behavior. In the fraternity finding aspect, we employ classes in place of web sites as the rating units and we also utilize web classification modules. Finally, we show that the predictive inaccuracy of our model is lower than traditional ones based on simulation results. In summary, the contributions of this thesis include creating a new way of web sites recommendation, solving the problems of traditional cases, and achieving the goal of raising predictive accuracy.