Serving Comparative Shopping Links Non-invasively

Marketing and commerce are two activities that have primarily funded the economic lifecycle of the Internet. From seller’s perspective, the objective is to sway user traffic onto its Web-site which offers merchandise and/or service. Strategies to fulfill this goal typically range from push and pull ads to spam. Ads are presented at publishers’ sites which offer content popular among the general public. From buyer’s perspective though, shopping has become an overly distractive and deceiving process as search engines paired with recommendation systems often spread their results over many Web-pages so to present as many as possible ads. We propose a simple, user-friendly tool which aims to offer comparative shopping to the consumer with minimal distraction. The key idea is to detect whether a specific Web-page is commercial, i.e., whether it sells an individual product or service. The detection is performed in real-time at the client with focus on exceptionally low false positives. For each commercial page, we aim to identify the product name P from its hypertext and send P to a knowledge server which would respond with a list mathbbL of URLs at which P is sold in increasing order of pricing. The browser would then present mathbbL in an non-invasive fashion to the user. The user could browse K sites in mathbbL with only K clicks resulting in a simple and effective shopping experience. In this paper, we present certain statistical properties of collected commercial Web-pages, introduce a novel classifier of Boolean spaces used in the detector, and compare its performance to an SVM with quadratic programming.