Aspect-Based Sentiment Analysis and Summarization of Online Product Reviews
Disciplines
Artificial Intelligence and Robotics
Abstract (300 words maximum)
Online shopping has grown more popular and accessible over the past decades, with countless products being available to consumers at the click of a button; however the sheer volume of similar products available can complicate the shopping experience. User-submitted reviews offer some insight into the quality of a product, but reading through several reviews to compare similar products can be time-consuming and frustrating for the consumer. This project looks to reduce the time spent comparing products through the implementation of natural language processing techniques. We developed a system to generate a user-friendly report of the pros and cons of a given product, using information extracted from user-submitted reviews, to avoid the need for the consumer to read through the reviews themself. An entity tagger was built to identify and extract frequently mentioned aspects and the associated sentiment phrases from reviews for laptops on Amazon.com. By training a classifier to perform aspect-based sentiment analysis on the extracted data, we are able to determine the overall customer sentiment towards each aspect. The identity of each aspect, the context in which it appears, and the determined sentiment were provided to a summarizer model to create a short report of the positive and negative attributes of the product. We found that this generated text was a much more consumer-friendly way to learn about the pros and cons of a specific product, which should make deciding between similar products much simpler for potential buyers. Alongside the increased convenience for consumers, this system has business applications for gauging general customer opinion towards a specific product, or for comparing multiple available products in the same market.
Academic department under which the project should be listed
CCSE - Computer Science
Primary Investigator (PI) Name
Md Abdullah Al Hafiz Khan
PowerPoint presentation for the symposium
Aspect-Based Sentiment Analysis and Summarization of Online Product Reviews
Online shopping has grown more popular and accessible over the past decades, with countless products being available to consumers at the click of a button; however the sheer volume of similar products available can complicate the shopping experience. User-submitted reviews offer some insight into the quality of a product, but reading through several reviews to compare similar products can be time-consuming and frustrating for the consumer. This project looks to reduce the time spent comparing products through the implementation of natural language processing techniques. We developed a system to generate a user-friendly report of the pros and cons of a given product, using information extracted from user-submitted reviews, to avoid the need for the consumer to read through the reviews themself. An entity tagger was built to identify and extract frequently mentioned aspects and the associated sentiment phrases from reviews for laptops on Amazon.com. By training a classifier to perform aspect-based sentiment analysis on the extracted data, we are able to determine the overall customer sentiment towards each aspect. The identity of each aspect, the context in which it appears, and the determined sentiment were provided to a summarizer model to create a short report of the positive and negative attributes of the product. We found that this generated text was a much more consumer-friendly way to learn about the pros and cons of a specific product, which should make deciding between similar products much simpler for potential buyers. Alongside the increased convenience for consumers, this system has business applications for gauging general customer opinion towards a specific product, or for comparing multiple available products in the same market.