Use of Implicit User Feedback to Support Semantics-Based Personalized Document Recommendation

The development of the Internet and digitized documents has made it possible for data and information to be easily transferred, exchanged and shared online. However, for Internet users, this easy access to information also carries the risk of information overload. Document recommender systems are becoming an indispensable tool, helping Internet users effectively retrieve the information they need from the millions of documents available online. In this study, we design and evaluate an Implicit-feedback-based Concept-Expansion (ICE) document recommendation technique to address the difficulties inherent in acquiring relevant feedback. The ICE technique determines a focal user's preferred documents by implicitly observing and analyzing his or her browsing behavior in order to make appropriate document recommendations. Using a domain concept heterarchy (e.g., domain ontology) and employing the Spreading Activation Model (SAM), the ICE technique expands the concepts existing in the preferred documents. Documents with a greater number of related and/or expanded concepts are then considered to be . potentially appealing to the focal user and are recommended as such. A laboratory experiment was conducted to compare the system performance of the ICE technique with that of three benchmark document recommendation techniques: Explicit-feedback-based Concept-Expansion (ECE), keyword-based, and random. The results of the experiment show that the ICE approach proposed by this study is more effective than random or keyword-based document recommender systems. Although there is no significant performance difference between ICE and ECE, the ICE technique is expected to cost less in terms of user effort. Overall, the findings of this study provide some interesting implications for improving the quality of document recommender systems.