"We've all been here and things look good, but we're never sure if it's going to scale and really work," he said. "Our primary consideration was [extract, transform and load]. It's always the ETL that'll sink ya. We wanted to make sure the exhaustive extraction process was scalable -- hundreds of thousands of lines of text, versus hundreds of lines."
Another of Schwab's key considerations was finding a system with an open architecture to avoid vendor lock-in and scalability. Because financial services are heavily regulated, the company also needed robust group and role security features. While the user interface and reporting capabilities were factors, they were not as important as the open architecture, Bayes said. Finally, he wanted a system that Schwab could maintain itself.
"This was one of my big fears -- we don't want to hire any linguists," he said. "We want to do projects without re-engaging consulting services. We're going for that competency center role."
That's an important point for any company considering text analytics technology, noted Sue Feldman, vice president for content and digital marketplace technologies at Framingham, Mass.-based IDC.
"It's unlikely most companies will have the linguistic expertise to implement and maintain a text analytics application," Feldman said during an expert panel discussion. "That means a long-term consulting engagement or relying on Software as a Service."
Key considerations for a text analytics initiative
There are various questions companies need to ask themselves when evaluating text analytics, according to Halper, because it remains a diverse and complicated market. Firms should consider what kind of analysis they need, she said. Do they need to extract statistical information and marry it with their structured data to run predictive models? Or is just analyzing the text -- such as an insurance firm analyzing claim forms -- enough?
Also, budgets are a significant factor. Some of the syndicated services can provide analysis for $5,000 to $10,000 a month. Software as a Service applications will typically cost between $5,000 and $20,000 per month, and licensing the software can reach into the six figures, Halper said. 数据挖掘论坛
They should also ask whether other departments within the organization are looking at text analytics, or could potentially make use of it, and piggyback on those projects. Support for the languages the software needs to support is a key decision as well.
Early adopters of text analytics have seen challenges with getting data from source systems to analysis systems, developing taxonomies and establishing rules built around the built-in functionality, Halper added.
"If you want to buy enterprise software, ask [whether you] need a taxonomy. Some vendors say you don't," she said. "Where am I storing all that data if I don't have a good data warehouse in place?"
|
