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 96 the final stage (Stage 2), subjects were required to log into the system again. Each subject rated the abstracts of another ten recommended papers and indicated the perceived degree of recommendation quality and system interface quality, ranging from 1 (low quality) to 5 (high quality). In the end, 93 subjects successfully completed the experiment. The demographics of the 93 subjects and the number and percentage of subjects assigned to each treatment are listed in Table 4. Males made up 75.27% of the total sample (Females = 24.73%), and most subjects were graduate students. All subjects had experience using academic electronic databases. Table 4 Demographics of Subjects Variable Frequency (%) Random- based Keyword- based ECE ICE Sub-total Gender Female 11 (47.83%) 3 (13.64%) 4 (15.38%) 5 (22.73%) 23 (24.73%) Male 12 (52.17%) 19 (86.36%) 22 (84.62%) 17 (77.27%) 70 (75.27%) Age 22-23 5 (21.74%) 6 (27.27%) 4 (15.38%) 10 (45.45%) 25 (26.88%) 24-25 13 (56.52%) 13 (59.09%) 16 (61.54%) 9 (40.91%) 51 (54.84%) 26-27 4 (17.39%) 3 (13.64%) 4 (15.38%) 2 (9.09%) 13 (13.98%) 28-29 1 (4.35%) 0 (0.00%) 1 (3.85%) 1 (4.55%) 3 (3.23%) 30-31 0 (0.00%) 0 (0.00%) 1 (3.85%) 0 (0.00%) 1 (1.08%) Average usage time of online periodical database per month (hours) < 1 5 (21.74%) 1 (4.55%) 5 (19.23%) 5 (22.73%) 16 (17.20%) 1-2 3 (13.04%) 3 (13.64%) 1 (3.85%) 3 (13.64%) 10 (10.75%) 2-3 7 (30.43%) 8 (36.36%) 10 (38.46%) 7 (31.82%) 32 (34.41%) 3-4 2 (8.7%) 1 (4.55%) 3 (11.54%) 4 (18.18%) 10 (10.75%) 4-5 3 (13.04%) 6 (27.27%) 3 (11.54%) 3 (13.64%) 15 (16.13%) > 5 3 (13.04%) 3 (13.64%) 4 (15.38%) 0 (0.00%) 10 (10.75%) Total 23 (100%) 22 (100%) 26 (100%) 22 (100%) 93 (100%) 4.3 Analysis Results Since all values of Cronbach’s α are higher than 0.8 (computer self-efficacy: 0.839; recommendation quality: 0.802; system interface quality: 0.819), the reliability of the measurements is acceptable. In order to test whether the system interface quality was consistent across all four treatments, the four related items were aggregated into one overall system interface quality score. Then an Analysis of Variance (ANOVA) was conducted to compare system quality among the four treatments. The result shows no significant differences between these four treatments. Hence, we conclude that the