Clustering e-commerce search engines based on their search interface pages using WISE-Cluster

Abstract

In this paper, we propose a new approach to clustering e-commerce search engines (ESEs) on the Web. Our approach utilizes the features available on the interface page of each ESE, including the label terms and value terms appearing in the search form, the number of images, normalized price terms as well as other terms. The experimental results based on more than 400 ESEs indicate that the proposed approach has good clustering accuracy. The importance of different types of features is analyzed and the terms in the search form are the most important feature in obtaining quality clusters.

Classifying Structured Web Sources Using Aggressive Feature Selection

“…Second, as mentioned in Section 1, the works of He et al (2004), Lu et al (2006) and Barbosa et al (2007), together with this paper, are on the same topic of SWS categorization. While the prior works show that it is feasible to determine the domain of a source by using discriminative features extracted from source's search interfaces, we take a step further.…”

“…Similar to other studies Lu et al, 2006;Barbosa et al, 2007), we assume that one web database, together with its search interfaces, belongs to one category. Since a document representing a web source belongs to one category, and there are multiple categories, our problem is equivalent to a single-label multiclass text categorization problem (Sebastiani, 2002).…”

“…However, it is not difficult to see that in a search interface, words that help in distinguishing categories (e.g., 'author', 'textbooks') mingle with many more other words. Indiscriminative or noisy terms (e.g., 'sort by', 'state') also occur inside forms, as observed in (Lu et al, 2006). Thus, this paper investigates on how to identify features suitable for categorizing structured web sources, i.e., the feature selection (FS) problem.…”

“…in an Amazon's search form) are the right "representatives" for structured sources, and used only them. Subsequently, in addition to form labels as the most important feature, Lu et al (2006) identified and utilized other features such as form values (e.g., 'hardcover', 'paperback') and other regular text terms. In these two works, features inputed to their clustering algorithms must be extracted from HTML pages by another technique (see Lu et al, 2006).…”

“…Subsequently, in addition to form labels as the most important feature, Lu et al (2006) identified and utilized other features such as form values (e.g., 'hardcover', 'paperback') and other regular text terms. In these two works, features inputed to their clustering algorithms must be extracted from HTML pages by another technique (see Lu et al, 2006). In contrast, Barbosa et al (2007) argued that such an extraction task is hard to automate, so they used all the text (bagof-words) of a search interface, which is partitioned into text of the form and text of the page, together with backlinks pointing to the interface.…”