|
By Lauren E. Bielski
Want to know what your customers are saying about your company? Forget looking for the elusive truth in tea leaves. The answer just might lie online and in other sources of text.
Speed reading isn't the answer, but analytics just might be. While it isn't a mainstream practice yet, text analytics is gaining interest and greater use, as early adopters are beginning to automate the evaluation of unstructured data, which represents the majority of what companies collect.
"Certainly, the prominence of Web 2.0, where everybody who wants to express an opinion can do so, has promoted wider interest in analyzing text," says Nick Patience, research director for information management at the 451Group, which has offices in New York, Boston, San Francisco, Washington and London.
At the recent 5th Annual Text Analytics Summit in Boston, the 451Group analyst discussed options for vendors in the emerging field. Patience jokes that one theme in his presentation was "only the paranoid survive," adding that sentiment analysis has its place in the broader field of customer experience management, which involves systematic efforts to please and engage customers by understanding what they experience in their interactions with the stores or customer service channels. One of the vendors that does this, in part by determining whether an author's opinion is positive, negative or neutral on a topic, is Lexalytics, acquired in 2008 by British document management firm Infonic.
Growing out of a cross section of tools that essentially performed a kind of "extract, transform and load" function to normalize unstructured data, text-based analytics looks at language, interprets it grammatically and establishes other important relationships based on what's written in a given piece.
In contrast to numerical data, which has long been the fodder of ambitious business intelligence programs, the field of text analysis has been gradually developing out of natural language processing tools that convert samples of language into structures, such as trees, which computers can process. Normalized data then can be consumed by broader applications.
Now, according to the 451Group, text-based analytics is a small but viable industry with companies that include Attensity Group, Temis, Nstein, Clearwell, Recommind and SAS, and Stratisfy.
Challenges and opportunities
Customer sentiment may be driving current interest, but text-based analytics can be used to address other problems, including locating documents within organizations. For instance, Autonomy, with U.S. offices in San Francisco, offers a data processing server called IDOL, which outputs data that is then applied to problems of e-discovery in the banking and insurance industry.
Assessing employee sentiment is another area of interest.
"We see interest in text analysis as part of broader efforts to engage in improved data governance and keep track of who works with what and where the data is," says Patience.
Information overload is a chronic issue with most businesses and business people. "Today, companies collect masses of data as part of their process-the notes that go along with filling out a form on a customer service call, for example," explains Dean Abbott, president of Abbott Analytics, based in San Diego. "Marketing people know that they could mine that information and use it to improve service or sales calls. It's a matter of how to harvest it in a time and cost effective way."
Not that automating data analysis is easy. Abbott, who has worked on traditional data mining projects for the government and many businesses, knows that text-based analytics is still a form of business intelligence. People who work with it need to know what they are looking for, and have a knack for forensics for establishing patterns.
"If a company doesn't have an effective data governance practice and a good track record with traditional analytics, they probably wouldn't do well with text-based analytics," he said in response to a question about whether text mining could develop separately from traditional BI efforts.
Moreover, language has its own ambiguity. Slang, misspellings, poor usage and texting are some of the hitches that get in the way of straightforward pattern recognition. (SAS Text Miner, for example, can handle multiple languages.)
Despite all the challenges, Abbott thinks that more companies would be interested in working with vendors in the space. "Over time, handling unstructured data correctly can help streamline data collection."
He offers an example: analyzing written employment performance assessments. From that analysis, certain standardized terms can develop and be recommended or utilized in an HR application that helps the business manager put together a performance review.
Another Text Analytics Summit presenter, Seth Grimes, principal consultant with Washington, D.C.-based consultancy Alta Plana, noted that text analytics can boost business results via established data mining programs or independently. Contact center notes, surveys, forum postings, email and other online sources can yield attitudinal information that numbers alone can't.
Nick Patience believes that over time, suites that blend structured and unstructured data analysis capabilities will become more common. "Certainly, nobody wants more complexity and greater numbers of operational silos," he says, referring to historic issues connected with business intelligence projects.
Only registered users can write comments.
Please login or register. |
