Natural Language Processing (NLP): Applying Sentiment Analysis to Improve Government Customer Experience

For those of you who keep up with tech news or flex your Pythonista skills, “sentiment analysis” is a common term. For those who don’t know the term, sentiment analysis is when data scientists apply natural language processing (NLP) to analyze content from websites, e-mail, and social media.

NLP is a form of data analysis applied to language. A simple word count is one thing, but NLP allows us to summarize and classify documents, and identify clusters or topics to draw relationships and trends. “Sentiment analysis” is the process that identifies and categorizes opinions expressed in text to determine whether the tone towards a topic is positive, negative, or neutral.

At the Bureau of Fiscal Service, we are excited about our progress in developing NLP solutions to improve customer experiences. Consider the two use cases below, involving customer e-mails and telephone calls.

 Understanding Customer Feedback

A team of data scientists conducted sentiment analysis on e-mails to an inbox for a Fiscal Service website, FiscalData.Treasury.gov to find out why customers were contacting them. 

The team was asked to conduct research using NLP to determine emotional/sentiment scores, and to distinguish whether the body of language in the e-mails were questions or comments.

The results, as shown in Figure 1, show that most messages scored as emotionally neutral because they asked a question or contained financial jargon. But there was some negative sentiment shown in the category “Technical Issues”.

Figure 1

Source: 2022 FDG Inbox Sentiment Analysis, Research & Analytics, Office of Chief Data Officer

Optimizing Call Centers

Last year, the Bureau embarked on a journey to see how we might improve the customer contact experience. Customer contact is defined when a customer reports a problem or need which they are looking to resolve or find a solution.

We recognized there would be a need for NLP as a tool in advanced analytics that could be used to help solve these challenges, specifically to determine why customers are driven to call Fiscal Service.  Our first step is to get and analyze the data (get audio files and e-mails) from our 21 nationwide call centers.

Once data are collected, our data scientist will convert the audio files to text and apply sentiment analysis to help Fiscal Service understand why members of the public contact us throughout the year. That analysis will be used to develop a plan and prioritize our future investments to improve the customer experience and deliver on what our customers need most.

As we learn more about our customers’ journeys, we are excited to start applying technology innovations such as NLP and other developing technologies which we believe will enhance the customer experience across government!

Stay tuned for more updates on this journey.