Clickstream Analysis

This theme focuses on analyzing clickstreams in streaming media to understand client interactive behaviors such as pausing, rewinding, or jumping within media files. Such characterization aids in generating realistic synthetic workloads for performance evaluation of streaming systems and informs caching strategies. Addressing various content domains (education, entertainment audio/video) recognizes the differing interactive patterns, which matter for designing scalable and efficient streaming protocols.

Data preprocessing transforms raw web server logs into analyzable formats crucial for valid mining of user clickstream patterns. Reconstruction of web visitor sessions involves identifying unique users and segmenting their navigational sequences despite complications like shared IPs and crawlers. Precise session delineation directly impacts the quantity and quality of extracted behavioral rules and pattern analyses, influencing web personalization and portal optimization.

This research theme investigates the fusion of server-side clickstream logs with ethnographic data from direct user observation and surveys to uncover user web browsing strategies and their evolution over time. The combined approach overcomes the limitations of clickstream data alone, which lack insights into user intentions and interactions like use of the back button, as well as the scalability constraints of ethnographic studies. Understanding user strategies guides better website design and usability evaluation.

This theme explores the application of advanced machine learning models—including neural networks, clustering algorithms, and ensemble methods—to extract and predict meaningful user navigation and browsing patterns from clickstream data. Reducing model complexity while maintaining predictive accuracy is a key concern addressed through pattern extraction (e.g., longest repeating subsequences) and classifier use. Accurate modeling supports improved recommendation systems, personalization, and web service optimization.

This area deals with analyzing network traffic to distinguish between natural user inactivity (think times) and disruptions caused by network problems that degrade streaming or browsing experience. The differentiation is critical for ISPs and service providers to identify network faults and improve QoE. Methodological innovations include ON-OFF modeling and wavelet-based criteria that operate on packet traffic flows without deep packet inspection.

Eye-tracking offers granular quantitative data on visual attention during web interactions, but the interpretation of gaze trajectories into distinct behaviors remains challenging. This research theme investigates using advanced machine learning classifiers (LSTM, random forest, MLP) on scanpath data to discriminate among different web browsing tasks, offering a novel quantitative complement to traditional qualitative eye-tracking analyses and enhancing user experience evaluation.

This research investigates dynamic profiling of viewers based on continuous streaming of interaction data such as ratings and preferences, using incremental learning methods tailored for stream data. Accurate viewer models underpin personalized recommendation systems that adapt in real time, helping to improve prediction accuracy and viewer engagement in multimedia streaming platforms.

Metrics-based web analytics involve defining and leveraging quantitative indicators to measure website performance and visitor behavior. Proper metric selection and analysis enable identification of popular pages, behavioral transitions, and bottlenecks, guiding optimization efforts to improve user experience and business objectives.

This theme examines the validation and application of remote proxy-based mouse tracking to investigate culturally influenced browsing behaviors, comparing groups such as Chinese and European users. Mouse-tracking provides a scalable and less intrusive approach than eye-tracking, enabling large-scale behavioral comparisons that can inform culturally sensitive web design.