21th AIAI 2025, 26 - 29 June 2025, Limassol, Cyprus

Evaluation of Validity of Users’ Behavioral Models Detected from Session Data

Erika Nazaruka, Jurijs Kornienko, Vitaly Zabiniako, Toms Rožkalns

Abstract:

  Determination of users’ behavioral models from datasets of users’ session da-ta relates to the clustering task within the application-level web data mining process. The well-known problem of clustering methods is instability in re-sults for different situations. Therefore, evaluation of clustering validity is an essential step. Internal and external cluster validity indices exist, but they cannot replace the evaluation of a human expert in a certain business do-main. This paper represents the results of a human expert’s evaluation of two clustering methods – the clickstream method with the Louvain algo-rithm and the agglomerative sequence alignment method with the hierar-chical agglomerative clustering algorithm – applied to a real-world dataset. The evaluation process is done in three iterations. The findings prove the suitability of the clickstream method for this task and discover additional steps to incorporate a human expert’s knowledge into the data transfor-mation process. The results of this research can be used to solve similar tasks on users’ behavioral model detection from session data and to improve the process of data mining.  

*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.