next up previous
Next: Performance vs. TFIDF Up: Evaluation Previous: Experiment

Consistency of Results

Because WordSieve is a probabilistic algorithm, it does not produce exactly the same results every time it is run. To test its performance variation, all simulated user sessions were run through WordSieve four times and the results were compared. Although there was variation among individual rankings of documents, the variation was not statistically significant according to a repeated measures ANOVA test (F(3, 1146)=2.26, p<.05). This suggests that although WordSieve works probabilistically, it is consistent across even a relatively small set of data (in our experiment, 60 minutes worth of browsing). This also suggests that the results of the experiment are not the results of ``lucky'' runs of the simulation, but that our data set is large enough to faithfully represent the abilities of the algorithm.



Travis Bauer
2002-01-25