Structured Feature Extraction Using Association Rules

As of today, opinion mining has been widely used to identify the strength and weakness of products (e.g., cameras) or services (e.g., services in medical clinics or hospitals) based upon people's feedback such as user reviews. Feature extraction is a crucial step for opinion mining which has been used to collect useful information from user reviews. Most existing approaches only find individual features of a product without the structural relationships between the features which usually exists. In this paper, we propose an approach to extract features and feature relationship, represented as tree structure called a feature hierarchy, based on frequent patterns and associations between patterns derived from user reviews. The generated feature hierarchy profiles the product at multiple levels and provides more detailed information about the product. Our experiment results based on some popularly used review datasets show that the proposed feature extraction approach can identify more correct features than the baseline model. Even though the datasets used in the experiment are about cameras, our work can be applied to generate features about a service such as the services in hospitals or clinics.