Some time ago the Economist has published an article entitled “Sentiment Analysis – The Difference Engine: The Wisdom of Crowds. Mining the Web for opinions could be a boon for business, politics and consumer affairs”. In the past, we used to seek advice from people close to us like family and friends about what to buy, where to buy, and when to buy. We also read newspapers and magazines to form opinions on events, books, politicians, and acted based on this information. Businesses relied on extensive surveys, focus groups, and consultant services. With the rise of the social Web, this all changed.
Due to the Web, we are flooded with information from millions of opinionated users that we have never met and provide information that is difficult to assess. This phenomenon has been labeled as crowd sourcing. Finding the true facts is difficult, but this is not problematic for many application domains. On the markets, for example, it is not the facts, but the emotions that matter, as long as these emotions are shared by a large crowd. Finding the crowd’s opinions underlying emotions is a hot research topic.
Currently, there is a lot of work done on natural language processing, as it provides techniques that can be used to mine the crowds’ opinions. These techniques have been embraced not just by social scientists, but also entrepreneurs who see opportunity in building tools that provide information on what customers really think. One such example is Clarabridge of Reston Virginia that produces opinion mining tools used by large companies as AOL, Marriott, Nissan, Wal-Mart, Wendy’s, United Airlines and many more.
Sentiment analysis aims to identify the sentiment from an emotionally charged content in an automatic manner. The sentiment is often hidden in blogs, tweets, forums, etc., and manually extracting this information and subsequently annotating it with sentiment is hard to achieve. As professor Bing Liu from University of Illinois at Chicago points out, sentiment mining is a very complex process aiming to identify “quintuples” composed of the target object, the aspect of the target object that is evaluated, the sentiment value, the person expressing the sentiment, and the time when the sentiment was expressed.
Aspect-based sentiment analysis is the new frontier for the field of sentiment analysis. While many research groups have focused on pure machine learning approaches, other schools, including ours, have focused on blending state-of-the-art machine learning solutions, like deep learning, with expert knowledge, often expressed in domain ontologies, obtaining competitive results. Combining a data-driven solution with expert information allows one to obtain and reuse better quality data patterns for sentiment analysis.
In the future, we foresee the rise of sentiment engines that will be able to extract sentiment information in a real-time fashion from various online sources. Such engines will not only help average users to seek the desired information, but also companies to gather signals on people’s sentiment about their brands and products. This information will help customers make more informed decision and companies develop products that better suit customer needs, contributing thus to the efficiency of our economies.
About Flavius Frasincar
Flavius Frasincar is assistant professor economics & informatics and is affiliated with the Econometric Institute of Erasmus University Rotterdam. He is a member of the editorial board of Decision Support Systems and International Journal of Web Engineering and Technology. His research focuses on the confluence between economics and informatics. In particular he studies the application of information systems and artificial intelligence for the development of intelligent decision support systems.
About Kim Schouten
Kim Schouten is a PhD candidate at Erasmus University Rotterdam. He is affiliated with the Econometric Institute and Erasmus Q-Intelligence. Kim defends his dissertation ‘Semantics-Driven Aspect-Based Sentiment Analysis’ on November 16, 2018 (click here for more information about his defense). In his research, Kim explores the use of different signals for aspect-based sentiment analysis, both from within and outside of the text. He is particularly interested in using information from semantic lexicons to improve machine learning results.
- More information