Natural Language Understanding (NLU) for Synthesizing Online Product Reviews

As a form of user-generated content, online reviews serve as a major source of information for buyers evaluating products and services. At the same time, these reviews provide sellers with valuable insights into consumer perceptions and evaluations, offering opportunities to refine and improve their offerings. However, given the sheer volume of reviews for products in many cases, effectively harnessing this information can be challenging for both buyers and sellers. Clearly, processing and distilling such large amounts of largely unstructured data exceeds the cognitive capacity of any individual. Over the years, researchers have developed various methods to extract insights from large volumes of online reviews using various analytics techniques, focusing on aspects such as consumer sentiment, satisfaction, product strengths and weaknesses, product feature gaps, and overall sales performance. Such insights are typically obtained through labor-intensive and computationally demanding processes, relying primarily on conventional natural language processing (NLP) techniques and statistical methods. The introduction of Generative AI (GenAI) tools like ChatGPT has now created new opportunities to extract knowledge from large repositories of online reviews, insights that would be difficult to obtain using traditional methods. Focusing on consumer sentiment and satisfaction, as well as product strengths and weaknesses, we present a study that explores how GenAI, particularly techniques from the domain of Natural Language Understanding (NLU), can enhance our comprehension of online reviews. In this study, we analyze a dataset of online reviews from a prominent manufacturer and retailer of home and office furniture. We first apply conventional NLP-based analysis methods, then compare the results to those generated by GenAI, particularly when its foundational knowledge is augmented with new information using Retrieval-Augmented Generation (RAG) techniques (IBM Corporation, 2023). We report and compare the performance of the two approaches. Overall, this exploratory study aims to understand how GenAI and NLU can be leveraged to better synthesize and extract meaning from large volumes of product-related user-generated content, thereby making the voice of consumers in online reviews easier to interpret and act upon.