臺大管理論叢 NTU Management Review VOL.28 NO.3

85 NTU Management Review Vol. 28 No. 3 Dec. 2018 documents that are relevant to a topic (Mooney and Roy, 2000), it does not take the semantic meanings of documents into consideration when making recommendations (Lops, de Gemmis, and Semeraro, 2011). To address the need for effective document recommendations, Du, Li, and King (2009) proposed a semantic-based approach, which does take such meanings into consideration and recommends semantically relevant documents on the basis of a semantic network. Liang et al. (2008) also proposed a semantic-expansion approach to document recommendation, using a semantic network and employing the spreading activation model in order to expand users’ preference profiles. Though the semantic-based and semantic-expansion approaches have proven effective in recommending documents, the time cost and a heavy dependency on user- provided relevancy feedback may put semantic-based document recommender systems at a practical disadvantage. Privacy concerns and the extra effort required to fill out a questionnaire increase the difficulty of acquiring user preference feedback. (Lee and Cranage, 2011; Li, 2014; Wu, Huang, Yen, and Popova, 2012). Besides, users’ preferences may change over time. Users may be constantly asked to interact with the recommender systems in order to keep their preference profiles up to date. Therefore, the means by which to seamlessly and effortlessly update users’ preference profiles has become a critical issue for content providers attempting to provide appropriate personalized document recommendations. In this study, we propose an Implicit-feedback-based Concept-Expansion document recommendation (hereinafter abbreviated as ICE) technique to address the difficulties inherent in acquiring relevance feedback. Our proposed ICE technique is intended to estimate a focal user’s document preferences via observing and analyzing his or her browsing behavior implicitly in order to recommend appropriate documents. 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 that involve a greater number of related and/or expanded concepts are considered to have potential appeal and will be recommended to the focal user. The remainder of this paper is organized as follows. We briefly review some related works in Section 2. In Section 3, we describe the design of our proposed ICE technique. This is followed by Section 4, in which we discuss the experimental evaluation results and analyses. Section 5 presents our conclusion.

RkJQdWJsaXNoZXIy MTYzMDc=