EComReviews-BD: A benchmark dataset for aspect-based sentiment analysis on multilingual Bangladeshi E-Commerce reviews

The rapid growth of e-commerce has generated a large volume of customer reviews that serve as an important source of information for understanding consumer opinions and product quality. Aspect-based product recommendation and sentiment analysis relies on high-quality annotated datasets that capture customer opinions on specific product attributes. The proposed dataset was derived from the most popular e-commerce platform Daraz. It features a diverse amount of product reviews which includes 20,685 reviews with 304 products. All the reviews are in Bangla, English, Banglish format which are given by customers. After that we include sentiment and extract aspects for each product category as we have 19 categories of product. In the overall dataset, 74 aspects were extracted and manually annotated. All the reviews and product details were collected through web scraping. This fine-grained annotated dataset enhances comprehensive analysis in marketplace and product details, supporting future improvement, evaluating existing products. The dataset features multilingual reviews, product names, product categories, ratings and fine-grained aspect annotations, which provide a valuable resource for many different research areas. Along with aspect-based sentiment analysis, it can also support research in natural language processing (NLP), information retrieval (IR), recommendation systems, large language models (LLMs), and explainable artificial intelligence (XAI), which could open up pathways for different kinds of intelligent applications and future research.