Text analytics broadens its reach
Academic program helps companies launch text analytics
Students in the Oklahoma State University master’s program in analytics are on the leading edge of using text analytics in real-world settings. The students conduct their research in the context of a large analytics and data mining program led by Dr. Goutam Chakraborty, a professor at the university and author of Text Mining and Analysis: Practical Methods, Examples and Case Studies Using SAS.
The research focuses on three general areas: distilling meaning from a large group of documents by clustering and topic extraction; creating a hierarchy of categories for comments that express positive or negative sentiment, rather than simply scoring them; and advancing knowledge in the area of combining text analytics with quantitative data.
“Our work focuses on practical applications of analytics rather than on developing theories,” says Chakraborty. “We want our students to be prepared for real-world projects, and we want clients of the program to receive measurable benefits.”
The projects described represent each company’s first venture into the use of text analytics to augment other analytics activities being carried out. One involved a retail fuel company that tracked the comments of big rig truck drivers. The drivers have a mobile app they can use to input their comments when they go into the stores associated with the fuel stations. An automatic call back to the store manager delivers the comments, which are flagged as positive or negative.
At a financial company, offers were being made to customers based on a standard numerical model that included customer profiles of credit history and transaction data. A research project explored the effect of incorporating text from call center records to determine the offers made, and it was found that substantial lift resulted.
A healthcare provider was using numerical ratings of customer satisfaction but was not using a full-fledged quantitative model in its analytics. Students first developed a model and evaluated its predictive capability. In a second step, they analyzed text data and brought it back into the model. Text analytics explained the variance in customer satisfaction, and once incorporated, improved the predictive accuracy.
Each of those pilot studies provided the company with insights into what text analytics can offer. From that point, the companies can then continue the existing study in-house or go on to broader use of the technology.
One hazard of using analytical techniques, whether quantitative or text-based, is the hope that automation alone will provide an answer. “The importance of human input cannot be overstated,” says Chakraborty. “For example, after initial topics are defined and machine learning produces the initial results, the categories may need to be redefined. But humans who understand the context of the business domain must do that. The software cannot do it all.”