Modelling User Behaviour on Page Content and Layout in Recommender Systems

Abstract

Listen

Translate article

AbstractWith the exponential growth of information on the Web, recommender systems play an important role in many service applications such as e-commerce and e-learning. Recommender systems are used to assist users in navigating the Web or propose items that the users are likely interested in. Most of the currently prevalent approaches use collaborative filtering based on the preference of a group of similar users. In the past decade, there has been some but rather limited research in personalized recommender systems incorporating an individual user’s explicit and implicit feedbacks. In our previous work, a personalized recommender system that extracts an individual user’s preference and the associated Web browsing behaviour such as print and bookmark, has been designed and implemented. In this chapter, Web browsing behaviour reflecting a user’s preference on layout and design is investigated. We postulate that when a user browses a page, her actions on the content and links could be associated with personal preference on an object’s location, icon shape, colour scheme, etc. Furthermore, tags and labels of selected objects contain valuable information to facilitate the recommendation process. Consequently, systematic and automatic analysis of the relationship between information preference and Web browsing behaviour based on structure and schema learning could be exploited to complement recommendation utilizing content similarity. Survey and related work on personal recommender systems that model Web browsing behaviour are presented. A proof-of-concept system is designed with the objective to study whether there is a correlation between browsing behaviour, both in the content and visual aspects of a Web page, and user preference.KeywordsRecommender SystemCollaborative FilterUser InterestVisual AspectPage ContentThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.