Temporal Decay Functions for Clickstream Session Representation: A Comparative Analysis of Linear and Exponential Weighting Strategies
Workshops of 26th International Conference on Computational Science and its Applications, ICCSA 2026, Braga, Portekiz, 30 Haziran - 03 Temmuz 2026, cilt.16762 LNCS, ss.599-618, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Cilt numarası: 16762 LNCS
- Doi Numarası: 10.1007/978-3-032-30524-4_37
- Basıldığı Şehir: Braga
- Basıldığı Ülke: Portekiz
- Sayfa Sayıları: ss.599-618
- Anahtar Kelimeler: behavioral clustering, clickstream analytics, session representation, temporal decay functions, user behavior modeling
- Yıldız Teknik Üniversitesi Adresli: Evet
Özet
Clickstream data from e-commerce platforms offers valuable insights into how users browse. Applications such as user analytics, recommendation systems, and behavioral segmentation depend on modeling these sequential interactions. However, many existing session representation techniques use straightforward averaging to produce session embeddings. These tactics ignore the timing of interactions within a session. In fact, events that occur near the end of a session often provide stronger indications of the user’s immediate intent. This implies that the creation of useful session representations may depend on time information. This work examines the impact of temporal decay functions on the development of session embeddings for clickstream analytics. The study examines the effects of various decay strategies on behavioral clustering performance. We propose a session modeling framework that combines explicit temporal weighting techniques with transformer-based event embeddings to address these issues. Linear and exponential temporal decay functions are studied to understand how different weighting strategies affect clustering quality. Experiments on real-world clickstream datasets show that session representations with temporal weighting generally outperform traditional averaging methods. In particular, exponential temporal decay offers the best clustering performance across different embedding models. These results highlight how important it is to consider time in session representation learning. They also show that concentrating on recent interactions greatly improves behavioral clustering in clickstream analytics.