Product attribute extraction from C2C social media messages

Abstract

On social media, people could share information related to their desire to purchase, sell, or consume products or services, which serves as a marketplace for C2C e-Commerce. However, the message post by the social media users will not reach the potential buyer/seller out of your followers’ circle. Furthermore, due to the difficulties of interpreting the semantics of social media posts, extracting product attribution from them is also difficult. To fix these issues, our research proposes a framework for extracting product attributes from microblogging messages about product selling and buying in this paper. First, we use a hybrid approach that includes Knowledge Base (KB), rule-based, Conditional Random Field (CRF), and Logistic Regression to extract the semantics of messages using named entity recognition. The dataset was created using raw social media messages, product descriptions from ecommerce sites, and KB because there was no product attribute annotated training dataset. When applied to a real-world dataset, the proposed approach achieves high accuracy, with classification and CRF models achieving 95 and 82 percent accuracy, respectively.