The engine, called IBM OmniFind Personal Email Search (IOPES), allows users to search their mail based on concepts, such as dates and phone numbers, according to IBM. It also allows searchers to define their own concepts.
Once the software is installed, it indexes and analyzes the user's e-mail store. Searches are conducted through a browser interface that delivers results through a stripped-down, Google-like interface.
Users can enter simple keyword-based queries or ones using basic natural language constructions. For example, to find e-mails from a friend named Mark Smith, you could simply enter "from Mark Smith."
But to find only the e-mails Smith sent in a certain month, a query might be constructed as "Mark from January 2007." You could find his phone number by typing "Smith's phone number."
The results don't show a list of e-mail headers or display the messages in full. Instead, the software extracts the passage it believes contains the right answer, and highlights what it deems to be the specific information requested, such as a phone number.
Users can also search for attachments, with search results providing direct links to the documents in question.
E-mail is a good target for developing a semantic search engine because users frequently repeat certain phrasings and words and repeatedly exchange the same type of information. "There is a fairly large number of things that are so e-mail specific," said Shivakumar Vaithyanathan, the project's technical lead.
Researchers in a number of IBM labs worked on the project for the past year and a half, according to Vaithyanathan. The product has been quietly available on the company's alphaWorks site for a couple of months, but only now is IBM attempting to drive widespread adoption, according to a spokeswoman.
"To be able to solve all these problems in some meaningful way, we want some feedback," Vaithyanathan said.
IBM also released the tool internally to its employees and said it has received mostly positive responses.