AI Research Answer
Patterns of eHealth Website User Engagement Based on Cross-site Clickstream
3 cited papers · May 5, 2026 · Powered by Researchly AI
🧠
TL;DR
Cross-site clickstream analysis offers a powerful lens for understanding how users navigate digital health resources, capturing sequences of actions across web…
Cross-site clickstream analysis offers a powerful lens for understanding how users navigate digital health resources, capturing sequences of actions across web and application interfaces. Mohajer (2020) The study of user engagement in digital health contexts has grown increasingly important as mobile and web-based health platforms proliferate, with behavioral logs serving as a primary data source for understanding interaction patterns.1
1
User Engagement in Mobile Health ApplicationsBabaniyi Yusuf Olaniyi, Ana Fernández del Río et al.2022arXiv
View - Clickstream Data — Sequential log events collected at different levels of web or application usage, representing walks through a finite state automaton (FSA) of an application, where actual user walks are typically far fewer than all possible paths. Mohajer (2020)
- Variable Length Markov Chain (VLMC) Models — Models used to represent and analyse user web navigation sessions, where prediction accuracy has been shown to increase linearly with summarisation ability. Borges & Levene (2006)
- Web Usage Mining (WUM) — A method operating on web server logs to capture user navigation patterns and generate recommendation lists based on similar usage profiles.
- User Engagement in Digital Health — Measured through probabilistic and survival analysis applied to behavioral logs transformed into daily time series, enabling personalized engagement metrics and churn detection.
Want to research your own topic? Try it free →
Diagram
[User Interaction Layer] | v [Web/App Server Logs / Clickstream Collection] | v [Preprocessing & Session Segmentation] | v [Pattern Extraction Layer] | | | [Sequence [Graph/FSA [Markov Chain Mining] Modeling] Modeling] | v [Behavioral Pattern Repository] | v [Analysis & Output Layer] | | | [Engagement [Churn [Recommendation Metrics] Detection] Generation] | v [Personalized Intervention / Content Delivery]
Several methodological approaches have been applied to clickstream-based engagement analysis. Mohajer (2020) proposes representing clickstream sequences as graph structures corresponding to FSA walks, arguing that users traverse only a small subset of all possible paths, and that cycles within FSA graphs are the key structural unit for pattern identification . In contrast, Borges & Levene (2006) employ variable length Markov chain models, demonstrating through extensive experimentation that prediction accuracy scales linearly with a model's summarisation ability, making VLMCs particularly suited for next-link prediction in navigation sessions . Saxena & Shukla (2010) take a time-series approach using the One Pass Frequent Episode Discovery (FED) algorithm, folding time-series data over periodicities (day, week) to identify significant intervals and frequent patterns for forecasting user behavior, diverging from traditional Apriori-class algorithms1
.
1
Significant Interval and Frequent Pattern Discovery in Web Log DataKanak Saxena, Rahul Shukla2010arXiv
View Table
| Approach | Method | Key Output | Application |
|---|---|---|---|
| Graph/FSA (Mohajer 2020) | FSA cycle detection | Behavioral pattern platform | App clickstream |
| VLMC (Borges & Levene 2006) | Variable-order Markov | Next-link prediction | Web navigation |
| FED (Saxena & Shukla 2010) | Frequent episode mining | Significant intervals | Web log forecasting |
| WUM (Dixit & Gadge 2010) | Two-tier architecture | Recommendation lists | Web server logs |
Want to research your own topic? Try it free →
The evidence blocks retrieved do not include studies specifically focused on cross-site eHealth clickstream analysis, meaning direct empirical benchmarks for that specific context cannot be cited from the available literature. Additionally, semantic analysis of browsing patterns — while demonstrated on general Web of Data logs (DBPedia, Semantic Web Dog Food) — has not been validated in dedicated eHealth website contexts within the retrieved evidence, leaving a gap in domain-specific applicability. Hoxha et al. (2012)
- Clickstream sequences can be modelled as FSA walks, with real users traversing only a small fraction of all possible paths, enabling efficient pattern compression.
- Prediction accuracy in web navigation models increases linearly with summarisation ability in variable length Markov chain frameworks.
- Behavioral logs from digital health apps can be transformed into daily time series to build personalized engagement metrics and detect churn.
- Semantic formalization of usage logs using ontologies and RDF enables expressive querying of temporal and semantic browsing patterns.
- Web usage mining with two-tier architectures improves accuracy in capturing user intuition and generating navigation-based recommendations.
Want to research your own topic? Try it free →
- "Cross-site clickstream analysis methods for eHealth portals and patient engagement"
- "Markov chain and deep learning hybrid models for health website navigation prediction"
- "Churn detection and dropout patterns in digital health intervention platforms using survival analysis"
Research smarter with AI-powered citations
Researchly finds and cites academic papers for any research topic in seconds. Used by students across India.