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

Use of Implicit User Feedback to Support Semantics-Based Personalized Document Recommendation 86 2. Literature Review 2.1 Semantic-Based Document Recommender System In response to the fierce competition in B-to-C (Business-to-Consumer) e-commerce, many online stores have devoted themselves to providing personalized services such as effective product recommendations. Joining retailers of physical products, many content providers have also equipped their websites to provide personalized content recommendation services to their readers, recommending relevant books, news, websites, or academic articles. For example, udn.com 3 recommends related news and hot news while the user reads the currently displayed news article. ScienceDirect recommends articles related to the specific search result selected by the user. A recommender system makes recommendations by sifting through a vast collection of products or services to identify those that appear relevant or interesting to focal customers. Content-based filtering and collaborative filtering are two promising recommendation approaches (Lee et al., 2012; Liang et al., 2008). The content-based filtering approach has been widely used in textual document recommendations (Cheng and Hu, 2007), because it assumes that there are important associations among documents which can be analyzed, measured and compared according to their respective content attribute values (Alspector, Kolcz, and Karunanithi, 1998; Liang et al., 2006). Traditional content-based document recommendation approach views keywords as important features, and recommend documents that share similar keyword features with the focal user’s preferred documents. However, this approach is unable to capture the semantics of user interests, and suffers from synonymy and polysemy problems (Lops et al., 2011). Based on keyword matching, a relevant document can be overlooked because of a synonym, while polysemy may cause an irrelevant document to be recommended. The semantic-based approach was developed to address the limitations of the content- based recommendation approach (Middleton, Roure, and Shadbolt, 2009). The semantic- based approach expands the semantic meanings of keywords based on a semantic network. Instead of matching keywords, this approach measures the semantic similarity between documents, and recommends documents that share similar semantical meanings with documents preferred by the focal user. Liang et al. (2008) also proposed a semantic- expanding approach to personalized document recommendations and showed its 3 http://udn.com