Code-Mixed Reviews Harm E-commerce Recommendations in Bangladesh

A new research paper explores the challenges of personalized recommendations on e-commerce platforms in Bangladesh, specifically focusing on the code-mixed language used in product reviews. The study, conducted on data from Daraz Bangladesh, evaluates various recommendation models and finds that Item-based Collaborative Filtering is the most stable approach. The research highlights that the fragmentation of the Banglish vocabulary, a mix of Bengali and English transliterated into Latin script, significantly degrades recommendation quality. AI

IMPACT Highlights challenges in recommendation systems due to linguistic diversity in user-generated content.

RANK_REASON Research paper published on arXiv detailing findings on recommendation systems. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.IR (Information Retrieval) →

• arXiv
• Banglish
• Bengali Unicode
• Daraz Bangladesh
• English
• Explicit Matrix Factorization
• Hugging Face
• Implicit Matrix Factorization
• Item-item collaborative filtering
• NDCG@10

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →