News Stay informed about the latest enterprise technology news and product updates.

Best practices for choosing your text analytics software

Expert analysts and consultants weigh in on the best practices for choosing text analysis software and offer vital steps for picking from an array of content, sentiment and text analytics products.

Mining valuable information from email, blog posts and other unstructured corporate data depends upon text analysis software, but with so many content and sentiment analysis tools available, selecting the right software can be a significant challenge.

Taking the time for some self-discovery and identifying all the channels from which information flows into the organization can bring success in reach, according to Tom Reamy, chief knowledge architect and founder of KAPS Group, a text analytics consultancy based in Oakland, Calif.

“The first and No. 1 step is to spend a lot of time on self-knowledge, looking at your company and asking what questions you want text analytics to solve,” Reamy said, noting that an in-depth evaluation of a company’s information environment will reveal where many of those questions lie.

But choose information sources carefully, said Seth Grimes, founder of consulting firm Alta Plana Corp. in Takoma Park, Md.

“Not every organization needs to analyze Facebook fan-page posts or call center notes,” he said. “Further, you may not collect the most important content in-house, in which case you need to consider new, outside sources.”

Identify the types of content and its users
A critical next step is to “map out the different kinds of content and identify who is using it and how it is used,” Reamy said.

Jamie Popkin, a vice president and distinguished analyst with Gartner in Stamford, Conn., agreed. Any evaluation of text analytics software needs to take into account not only the types of content, but the avenues from which it comes.

Approach analytics as a platform
Think about text analytics as a platform and enabling technology, not as something that’s going to be used in a single application,” Reamy said. Even if you initially consider just one application of the software, "you’ll probably come up with new applications down the road and you’ll need some of those [other] capabilities.”

Most organizations, however, would do well to position text analytics as part of a broader operational or analytical solution, according to Grimes. He suggested that organizations that want to move beyond project-based text analytics “should look for solutions that build text analytics into key line-of-business applications and BI/analytics solutions. You want comprehensive, integrated analytics, not another siloed system.”

If a company already uses an enterprise content management platform with search capabilities, it should consider sticking with that vendor for the analytics part.

“That vendor may very well have text analytics, and you’ve already made that investment,” Popkin said. If organizations ignore an enterprise approach, ad hoc adoption might result and that can lead to multiple departments using multiple products. “Then they decide that they should use a single product and tell the IT department, ‘You need to take this over and they need to be integrated.’ When this happens, you often almost need to start again.”

Solve an existing business problem
On the other hand, it might first make sense to identify a real-world business issue where text analysis software can play a role.

“Start with a focused business problem and solve it in order to gain experience and build support,” suggested Grimes. “Find a doable project that will produce results within a relatively short time frame and just go for it.”

Grimes noted that this is a very different approach from a search for a technology aimed at meeting multiple departmental needs.

Reamy agreed that a proof of concept with actual examples would yield the best evaluation. “It’s about language and semantics and meaning, and the only way to test that is with real-life language,” he said. “It’s important to test with all the possible use cases that you’ve got.”

Many vendors will provide limited proof-of-concept trials at no or little cost, Grimes said. There are also many Software as a Service and open source software options to consider.

Do at least three rounds of development
Proof-of-concept trials are typically a six-to-eight-week process, and by then, Reamy said, training is already well under way. “You learn a whole lot more about the software and how you are going to use it.”

“Sometimes people don’t realize what it is they are actually searching for until they start using the tool,” Popkin said. He said it was also key – during evaluation and development – to figure out the skill levels that are required of users: “Depending on how sophisticated you want to get, you might need someone trained in linguistics, for example.”

Use software capabilities and features as evaluation filters
“Text analysis software is different from traditional software, and it’s important to remember that when you go about evaluating the software,” Reamy said. “In text analysis, scorecards are absolutely meaningless.”

One vendor’s product might only do sentiment analysis, while another’s might be limited in analyzing in different languages, Reamy said. “It’s important to find what fits all of your various needs.”





Dig Deeper on Text analytics and natural language processing software

Start the conversation

Send me notifications when other members comment.

Please create a username to comment.