A Two-step Approach for Interest Estimation from Gaze Behavior in Digital Catalog Browsing

Area of Science:

• Human-Computer Interaction
• Information Retrieval
• Data Science

Background:

• Eye gaze data offers insights into user interests for digital catalog content.
• Conventional behavior analysis struggles with dynamic attention shifts and irrelevant attributes.

Purpose of the Study:

• To identify user comparison phases during browsing.
• To detect specific attributes and values reflecting user interest, even indirectly.
• To enhance the accuracy of inferring user interests from eye gaze data.

Main Methods:

• A novel two-step approach combining short-term analysis with probabilistic latent semantic analysis.
• Likelihood-based short-term analysis to detect comparison phases and focus attributes.
• Probabilistic latent semantic analysis for subsequent interest estimation.

Main Results:

• The short-term analysis significantly improves the accuracy of the subsequent analysis step.
• The framework effectively extracts attribute combinations (aspects) relevant to user interest.
• The approach accurately estimates user interest based on identified aspects.

Conclusions:

• The proposed method accurately identifies user comparison behaviors and attribute-level interests.
• This approach overcomes limitations of conventional methods in dynamic browsing scenarios.
• The framework provides a robust solution for personalized content recommendation and analysis.