Semantic Usage Navigation Patterns for Predicting Users' Navigation Requests

Abstract

The explosive growth of the World Wide Web (WWW) has resulted in intricate Web sites,demanding for tools and methods to complement user skills in the task of searching for thedesired information. In this context Web usage mining techniques have been developed for thediscovery and analysis of frequent navigation patterns from Web server logs, which can be used asinput for recommendation engines. Web usage mining techniques have been associated with Webcontent mining approaches in order to increase the accuracy of recommendation mechanisms.Existing approaches represent Web pages content essentially by means of keywords, N-gramsor ontologies of concepts, being, therefore, incapable of capturing the semantic information andthe relationships among pages at the semantic level. Herein, we propose a method that combinesusage patterns extracted from server logs with detailed semantic data that characterizes thecontent of the corresponding pages. Thus, a method to extract and analyze frequent semanticnavigation patterns which are fed into a recommendation engine is proposed. We argue that byintegrating usage and Web pages detailed semantic information in the personalization processwe will be able to increase the recommendation accuracy. The proposed method is an example ofsemantic Web mining that combines two fast developing research areas; Semantic Web and WebUsage Mining. We conducted an extensive experimental evaluation that provides strong evidencethat the recommendation accuracy increases with the integration of semantic and usage data.The results show that the proposed method is able to achieve 15-17% better accuracy than a usagebased model, 5-7% better than a N-gram based model and 4-6% better than a ontology based model.Also the proposed method is able to address the new item problem of solely usage based techniquesby augmenting navigation patterns with newly added pages in a Web site.