Unum is a Chattanooga-based company that sells various types of insurance (disability and life, for example) to other companies for their employees. Last summer, Unum hired an intern who came up with a creative idea to improve the organization’s highly manual, subjective underwriting process. Currently, Unum underwriters must conduct extensive searches to obtain and assess recent news articles, financials, and additional information pertaining to potential client companies. Based on these assessments, they give recommendations as to whether or not Unum should insure these companies. The intern suggested that, instead of continuing to utilize this time-consuming, biased process, Unum might investigate implementation of sentiment analysis techniques to conduct potential client company assessments. This is what our Capstone team is working on this semester.
Sentiment analysis—also known as opinion mining—involves the use of natural language processing tools to determine the connotation associated with a block of text. It involves using a computer to assess whether a block of text (a Tweet, a movie review, or a news article, for example) is generally positive or negative. Once everyone on the team was able to wrap his or her mind around this concept, the real work began.
So far, our team has integrated not one, but two sentiment analysis libraries (Stanford Core NLP and Alchemy API) into a back-end Java application. We have utilized MVC and C# to construct a front-end GUI application and created a customized RESTful web service to communicate between these back and front-end applications. An underwriter can search for a potential client company in the GUI, and, in less than a minute, receive a list of recent news articles pertaining to that company. Specifically, this list contains each article’s title, URL, content, and sentiment score; additionally, it holds an overall sentiment score for the company (which is the average of the article scores). Ultimately, this application will add value to Unum by expediting, standardizing, and increasing the objectivity of its potential client company assessment process.