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

97 NTU Management Review Vol. 28 No. 3 Dec. 2018 experimental system interface was controlled successfully. We also compared three subject characteristics among the four treatment groups: English ability, computer self- efficacy, and average usage time of online periodical databases per month. The results of ANOVA analyses indicate no significant differences in measurements of these three characteristics among the four treatment groups. This implies that the four subject groups are homogeneous. In addition, document order had no significant effect on either users’ interest evaluations or users’ reading time. Hence, the order of the recommended documents is not dependent on user preferences. The associations between interest score order and users’ rating among the four treatments are illustrated in Figure 3. As the interest score decreases, users’ average ratings of the recommended documents also decrease. We argue that the limited number of documents in our experimental system is the cause of the dramatic decrease in recommendation precision. Hence, we adopted users’ ratings of the top five recommended documents in Stages 1 and 2 to evaluate performance differences among the four treatments. The average rating in each group in Stages 1 and 2 is shown in Table 5. Figure 3 Users’ Average Evaluation of Recommended Documents Note: 1. Red line: average scores of ten recommended documents in Stage 2. 2. Blue line: average scores of ten recommended documents in Stage 1. 3. Horizontal axis: the sequence of recommended documents in descending order by interest score. 4. Vertical axis: the user evaluation score (from 1 to 7).