Monitoring Human Website Interactions for Online Stores - New Contributions in Information Systems and Technologies - page 361

We have also selected ten most active participants and built individual decision
trees for each of them. For interest expressed in five-point nominal scale resulting
trees misclassification errors varied significantly between 25% and 65,7%. The most
important indicators with positive correlation where mouse_clicks , vertical-scroll and
prod_desc_time . Trees obtained for individual users show that the behaviour patterns
in relation to user interest vary significantly among them. It may be a promising idea
for websites to implement individual preference models for each user (or user seg-
ment), especially for frequent and committed ones.

5 Conclusion

In this paper, an ECPM tool for monitoring human-website interactions was pre-
sented. The main advantage of ECPM is that it allows for detailed tracking of user
behaviours on any website. This opens great opportunity for carrying out wide-
ranging studies in real world, not just on sites prepared only for the purpose of a par-
ticular study. The first attempt to determine which implicit factors are a good sign of
user interest with regard to e-commerce websites. Based on our experiments and ana-
lyzes, it can be concluded that monitoring scrolling activities, mouse usage and time
spent on a website (including active tab time) and its particular sections are a good
starting point. Especially significant correlation was shown by indicators such as ver-
tical_scroll , prod_other_time , page_height , mouse_distance , prod_desc_time , us-
er_active_time , tab_active_time and mouse_clicks. We plan to conduct further studies
to analyze in detail the importance of those indicators and the underlying causes
behind them in order to adequately interpret the relationships discovered.

6 Future Work

Although the results are promising, the experiments should be repeated over an ex-
tended period of time with larger number of participants to ensure that the study is
fully representative. We also consider to broaden the implicit factors spectrum and try
modelling methods other than decision tree learning. Another goal is to utilize the
ECPM tool for website usability evaluation.

References

1. Nichols, D.M.: Implicit Ratings and Filtering. In: Proceedings of the 5th DELOS Work-
shop on Filtering and Collaborative Filtering, pp. 31-36, Hungary (1997)
2. Middleton, S.E., Shadbolt, N.R., De Roure, D.C.: Capturing Interest through Inference and
Visualization: Ontological User Profiling in Recommender Systems. In: Proceedings of the
Second Annual Conference on Knowledge Capture (2003)
3. Avery, C., Zeckhauser, R.: Recommender systems for evaluating Computer Messages.
Communications of the ACM 40(3), 40-88 (1997)