HyperFacet: 多維度瀏覽之加速中介系統

Abstract

在網路越來越被普遍使用的情況下，透過線上找尋資訊和物品已經成為越來越重要的議題。當使用者是要對於那些物品有大略的瀏覽的時候，browsing 通常是比較適合的方法。然而傳統的browsing 介面有時候會沒有效率且讓人困惑。因此一種新的介面叫做multifaceted browsing interface 被發明了出來。 這種介面非常方便並且充滿彈性，讓許多使用者給予良好的評價。不過他有一些缺點，導致於該系統在實用上不是被普遍的使用。這些缺點之中，最令人在意的就是反應時間很久。由於在製造這種介面的時候需要複雜的query，因此這類的系統如Flamenco 平均都需要超過五秒的時間來顯示每一個網頁。 我們提出了叫做HyperFacet 的中介軟體，讓它架在原有的系統上，利用binary的操作來取代複雜的SQL。這個系統會將資料的特性轉換成binary arrays，並且透過hash 來提升他的性能。同時我們利用binary 的特性，設計一個非常有效率的方法來搜尋相似的物品。 使用我們中介軟體的系統只要做小部分的修改，並且記憶體跟事前處理的時間都不需要花費太多。結果使用我們的方法的系統，在三種的資料分配狀況中都明顯的減少了反應的時間，同時在不同的情況下都有穩定的表現。

With the massive amounts of data on the web, finding information and items on the net is becoming an increasingly critical issue. When the user is generally exploring the collection, browsing is usually more suitable for the task. However, traditional browsing can sometimes be inefficient and confusing. Thus, multifaceted browsing interface was invented. Multifaceted browsing systems are flexible and convenient therefore has a high credit from users. Yet its shortcomings prevent itself from becoming commonly adopted in practice. Among them, the most concerning issue is the long response time of the system. Due to its high complexity queries when generating the interface, multifaceted systems such as Flamenco often take more than 5 seconds to show its webpages. We propose our HyperFacet middleware to apply to existing multifaceted systems that uses binary operations to replace complex SQL queries. It transfers the attributes of the items into binary arrays and furthermore hash them for better efficiency. With the characteristic of binary arrays, we also designed a fast recommendation utility that retrieve items with similar attributes in a very efficient method. The systems applying our middleware need only small modification. Moreover, it costs little in memory and preprocessing time. As a result, our solution obviously reduces the response time of the system in three different type of data distributions, and performs more steady in various situations.