2900. Enhanced Behavioral Weighting and Explainable Clustering for Adaptive Recommendation Systems
Invited abstract in session WA-34: Advancements of OR-analytics in statistics, machine learning and data science 6 , stream Advancements of OR-analytics in statistics, machine learning and data science.
Wednesday, 8:30-10:00Room: Michael Sadler LG10
Authors (first author is the speaker)
| 1. | ATAKAN YILMAZ
|
| Data Science, Trendyol | |
| 2. | Serhat Mirkan Acar
|
| Data Science, Trendyol | |
| 3. | Egemen Karabiyik
|
| Data Science, Trendyol |
Abstract
We propose an advanced recommendation framework that refines user similarity modeling through action-aware graph analysis and explainable clustering. Our method assigns distinct importance to different user actions—such as clicks, basket additions, and orders—in the product co-occurrence graph, preserving fine-grained behavioral data while emphasizing more meaningful interactions. To balance the impact of highly connected nodes, we introduce a modified label propagation algorithm that dynamically adjusts edge influence, promoting fairer cluster formation. Additionally, we incorporate explainability into the similarity graph, enabling transparent reasoning behind recommendations and clarifying how user behaviors align. Clusters update frequently to accommodate evolving preferences, and recommendation lists are composed of items with high intra-cluster relevance yet unexplored by the target user. Extensive evaluations on large-scale e-commerce environment demonstrate improved engagement and diversity, underscoring the effectiveness of our action-sensitive graph construction, bias-resistant clustering, and explainable user similarity modeling for enhanced recommendation performance.
Keywords
- E-Commerce
- Machine Learning
- Algorithms
Status: accepted
Back to the list of papers