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 90 3.1 User Preference Assessment ICE technique identifies the documents that may interest a focal user by analyzing the user’s browsing behavior; that is, the time spent browsing through documents may reveal the user’s interests. Previous research has indicated that users can be roughly classified into two groups according to the length of time they spend on documents of interest (Parsons, Ralph, and Gallagher, 2004; White, Jose, and Ruthven, 2006). One group spends more time on documents which interest them, while the other group spends less time on the same documents (perhaps because they are familiar with a certain kind of document). Therefore, the focal user’s browsing style must be determined before the ICE- based system can make the first recommendation. Specifically, the ICE requires both the preliminary preference feedback and the time spent on browsing documents to determine the user’s browsing style. The algorithm first selects six documents from among the documents browsed by the user: the three that garnered the longest average browsing time, and the three that garnered the shortest average browsing time. Average browsing time is defined as the time the user spent browsing a document (in seconds) divided by the length of the document (in words). Subsequently, the average interest ratings of the three documents with the longest average browsing time and the three with the shortest average browsing time are calculated and compared. If the former is larger than the latter, the target user is classified as a “slow reader”; otherwise, the target user is classified as a “quick reader.” Table 1 shows an example of a user’s preference feedback and the time spent browsing the documents. Apparently, the average interest rating of the three documents with the shortest average browsing time is (7+4+5)/3 = 5.33, larger than the rating achieved by the three documents with the longest average browsing time ((3+5+5)/3 = 4.33). As a result, the system classifies the user as a quick reader. Once the browsing style is determined, the ICE technique can identify the documents of interest from the focal user’s browsed documents so as to expand the relevant concepts. The top- k documents which obtained the longest average browsing time are used to determine the preferences of slow readers. Conversely, the top- k documents that garnered the shortest average browsing time were used to determine the preferences of quick readers.