June 24, 2014

How Natural Language Processing is Changing Business Intelligence

Written by

From onsite customer behavior, to seasonal or even daily trends, the typical data warehouse can contain an eclectic mix of information. The insights gain from these data have propelled businesses into a new realm of customer understanding, but limiting analytics to this type of highly structured format excludes the majority of the data that’s being created today.

80 percent the data being created is unstructured. It’s generated from conversations between customer service reps and on social media, among other places. To mimic the advantages gained from harnessing transactional data, organizations are turning to natural language processing (NLP) technology to derive understanding from the myriad of unstructured data available online and in call-logs.

As early as 1960, engineers were working to design programs that could derive meaning from language. By the 1980s, natural language processing had grown enough to harness some meaning from conversation, but only in the form of rigid IF-Then rules. This format was incredibly time consuming to write, not to mention limited in its scope.

Driven by highly structured languages, it’s always been difficult for machines to grasp the context of human speech. But machine learning has helped computers parse the ambiguity of human language. With the advent of advanced statistical algorithms, programs are now capable of using statistical inference to make predictions on what was meant in conversation by calculating the probability of certain outcomes. And the brilliance of inference and machine learning is that the program can continually improve itself the more data it processes.

What does this mean for business? It signifies that all the insights hidden in unstructured data are becoming more attainable with each technological advance. It means that qualitative data is now quantifiable.

How Businesses Use NLP

A subset of natural language processing, natural language understanding is concerned with the reading comprehension of machines. By using the aforementioned statistical inference model, software developers are helping make natural language understanding a reality.

The most common application of natural language understanding is text analysis, also known as sentiment analysis. While transactional data helps organizations predict what actions customers will take, it fails to offer much insight into how they felt during the process, leaving significant gaps in understanding the customer relationship. That’s why businesses are most concerned with comprehending how their customers feel, not just how they’re going to act.

Sentiment analysis can most commonly be put to work gathering insight from social media. With millions – in some cases even billions – of current and potential customers online, there’s tons of data being created each day that brands can harness. Basic sentiment analysis tools like Digimind can search the web for mentions of your brand and quantify whether the context was positive, neutral, or negative. Digimind digs deeper into context by ranking the importance of the source based on their social media clout.

Email filters are another common application of NLP. By analyzing the emails that flow through their servers, email providers can calculate the likelihood that an email is spam based its content. Call centers are another area rich with unstructured data. Whenever customer representatives engage callers, those callers list specific complaints and problems. Mining this data for sentiment can lead to incredibly actionable intelligence that can be applied to product placement, messaging, design, or a range of other use cases.

Natural language processing and sentiment analysis has even found its way into higher education. Seattle Pacific University’s School of Education grades their teaching candidates based on the performance of their respective classrooms. During their candidacy, applicants maintain a personal blog to reflect on their teaching experiences, but with upwards of 120 blogs per applicant, the Associate Dean of the School of Education was overwhelmed with the amount of work before him.

By utilizing Semantria, a text analysis tool set up to work with Excel, the Dean was able to isolate key noun phrases from his students’ blogs and aggregate them in Excel for Semantria to analyze. With the aid of Semantria, Seattle Pacific’s Associate Dean of Education was able to connect specific noun phrases with both quality reflection and classroom performance. Specifically, he found a greater range of phrases was correlated with better classroom performance. Now, Seattle Pacific encourages their School of Education candidates to write more elaborate posts, thereby encouraging more in-depth analysis of their teaching habits.

Such qualitative behavioral factors are often beyond the reach of purely transactional data, which is why Semantria’s CEO believes that within two years 100% of predictive analytics programs will incorporate text analytics. If transactional data gives organizations an understanding of how their customers are acting, it seems unstructured data gives them the why.

Does your company use sentiment analysis or text analytics software? Have you seen results? Let us know in the comments!