以可重複序列挖掘網路瀏覽規則之研究

Abstract

網頁瀏覽過程中的許多資料，如能適當的整理，應可幫忙了解使用者的瀏覽行為。而了解瀏覽者的行為不僅可幫系統預取所需網頁，並可幫系統業者決定在那些網頁上刊登廣告，也可替proxy server制定較佳的資料更新策略，減少使用者在瀏覽網頁的等待時間。在本篇論文中我們利用資料挖礦技術來建立使用者瀏覽關聯規則，以了解使用者的瀏覽行徑，與先前研究不同的是為能了解使用者瀏覽順序，此關聯規則是建立在序列的資料結構上並且將發現的關聯規則分成前溯型及後推型兩種；為了解決序列獨特的重複問題，此論文設計了特別的門檻值計算方法；為了解決門檻值與序列長度成反向成長的問題，此論文設計了Next Pass Large Threshold及Next Pass Large Sequence。

Web traversal patterns and rules are valuable to both Electronic Commerce and System Designers. If business owners know users` traversal behaviors, they can put advertisement banners in proper web pages with proper order. The same information can help systems to pre-fetch web pages and reduce response time. In this article, we propose a new data mining method to find the traversal patterns and associated rules. Traversal patterns are recorded in sequences, which have total orders among their elements. Sequences may have duplicated elements, and hence requires a new threshold computing method. The new method results in thresholds decreasing when sequences expanding. To resolve the issue, we design Next Pass Large Threshold and Next Pass Large Sequences to forecast needed sequences and thresholds. To expand sequences properly, sequence join, instead of traditional set join is employed. Since sequences contain orders, the rules established include forward reasoning and backward reasoning. Forward reasoning asserts rules in the order of event happening. Backward reasoning, on the other hand, asserts the rules in the reversed order. Both rules are valuable to EC and system designers.