Avatar Seeks Semantic Search

While a lot of attention has been heaped upon Powerset and its almost-here natural language search, IBM has been working on a similar technology that may or may not be as close to public debut.

IBM calls their effort Avatar Semantic Search. Right now it doesn’t even have the nice minimalist home page Powerset has for early peek signups, but since everyone’s done reading ‘Harry Potter and the Deathly Hallows’, a little text to read is a good thing.

“Ongoing research in Avatar is at the cusp of a number of disciplines ranging from search and information retrieval to machine learning, information extraction, and probabilistic databases,” IBM announced on the project’s page.

We’ve looked at earlier IBM efforts to pull information out of unstructured resources. Their UIMA developments now occupy a place in the freely available IBM Omnifind Yahoo Edition enterprise search product, for example.

But UIMA is so 2005. While Powerset has drawn upon research performed by the Palo Alto Research Center, aka PARC, IBM reached out to the academic community to complement Avatar’s internal team.

They have approached the semantic search issue in three ways. Developing an information extraction system will allow Avatar to plunge into mounds of raw text, and emerge with structured data based on rules-based annotators.

IBM claimed this extraction system will permit unsophisticated users to build an annotator with Avatar and pull out the desired information from email, web pages, business reports, etc.

Through semantic search, the researchers think they can interpret queries people make, and model the real intent behind a query.

The real challenge comes from an effort they refer to as managing uncertainty and probabilistic databases. They’ve stepped deeply into theory here, well beyond any help Douglas Adams can provide for me.

IBM built momentum with UIMA starting well before I’d interviewed Marc Andrews about it in December 2005. It led to the co-branded, freely available Omnifind product I mentioned earlier, and I have to think Avatar may be on a similar track today.

It had been about a year since IBM’s Marc Andrews, director for strategy & business development for content discovery, chatted with WebProNews about UIMA. I talked with him while he was enjoying the always-cheerful environs of a busy airport somewhere about the latest news about UIMA.

The technology behind UIMA offers the promise of going beyond the typical search capabilities seen in search engines like Google and Ask today. UIMA understands unstructured information, stored in a multitude of formats, by its concept rather than just as keywords to match.

In January 2006, IBM placed UIMA on SourceForge, giving open source developers access to its framework.

The tech world and beyond have taken notice; two major projects, at the Mayo Clinic and the Memorial Sloan-Kettering Cancer Center, have been implementing UIMA-based solutions to aid in finding and facilitating deeper and more encompassing medical research.

Andrews said IBM has seen adoption of UIMA throughout the academic, government, and business worlds over the past year since we’ve spoken.

IBM’s open source efforts continued with their recent steps into that community. Apache has launched a project, Apache UIMA, which should drive more community involvement with UIMA.

Ideally, Andrews would like to see what the free/open source software community can do with creating components that take advantage of UIMA to fulfill a given need. “Value comes from components that others can draw from,” he said.

The OASIS efforts at developing a standard for UIMA are at a very early stage. Andrews said the official call for participation will come from the OASIS UIMA Technical Committee, comprised of heavy hitters like Carnegie Mellon University and the Army Information and Intelligence Warfare Directorate.

Those who would like to experiment with UIMA at Apache’s project site will find IBM has contributed UIMA’s version 2.0 source code to the incubator. Carnegie Mellon created a UIMA Component Repository, and analytics tools from sources like the UK’s General Architecture for Text Engineering (GATE) and OpenNLP may be obtained freely.

—

Add to Del.icio.us | Digg | Reddit | Furl

Bookmark WebProNews:

David Utter is a staff writer for WebProNews covering technology and business.

At Virginia Tech, the System X supercomputer consists of a cluster of 1,100 of those Apple Xserve G5 machines.

The PhysOrg website reported on how researchers at the school have developed a method of finding answers by combing through seemingly disparate events.

Those researchers call the search capability they’ve created, “Storytelling.” It is a lot more than keyword search, and sounds more like the conceptual search IBM has been working with its UIMA project.

The Storytelling process has been described as finding a “chain of concepts” between specified start and end points.

A researcher on the project described the process:

It’s pretty amazing technology at work, and it sounds like they can find answers to questions that are a little deeper than my last web search for Midnight Oil lyrics.

—

Add to Del.icio.us | Digg | Yahoo! My Web | Furl

Bookmark WebProNews:

David Utter is a staff writer for WebProNews covering technology and business.

Departmental projects and small to mid-sized companies may need products to help them manage information across the business. Not all of them need a sizable install of IBM’s high-powered solutions to do search or content integration.

IBM wants to embrace that market, a sensible move because while there are a lot of bigger companies in the world, there are plenty of smaller firms that a high-end offering simply does not match.

The IBM WebSphere Information Integrator OmniFind Starter Edition addresses that market for search products. It can analyze and index information across a business from internal portals, databases, and other sources.

Companies needing expanded capabilities in OmniFind Starter can plug-in the Unstructured Information Management Architecture (UIMA) framework. We have discussed UIMA before, and its ability to sift through concepts and not just keyword search.

Content management systems exist in many forms, beyond those used by online publishers. To address them and have information available beyond their confines, a business can use the WebSphere Information Integrator Content Starter Edition.

IBM said the Information Integrator works with distributed content “as if it were stored and managed in a single repository.” Companies that have considered shifting to a service-oriented architecture (SOA) will find the Information Integrator can extend “the information as a service framework to include unstructured information.”

Through the use of out-of-the-box connectors, or custom ones created with the toolkit that comes with Information Integrator, users can access multiple data repositories from a single interface built to use the Integrator’s behaviors.

—
Tag:

Add to Del.icio.us | Digg | Yahoo! My Web | Furl

Bookmark WebProNews:

David Utter is a staff writer for WebProNews covering technology and business.

Upon seeing reports like the one in the Times Online UK, we had to wonder if Microsoft president for EMEA Neil Holloway was discussing a Microsoft integration of UIMA into its search capabilities. Holloway told conference attendees in Paris that Microsoft will unveil a new search engine in Britain and the US in six months before unleashing it on the rest of Europe.

And it’s going to make everyone forget about Google:

The last point echoes the long-ago battle between Philips and Sony over videotape format standards. Although Sony’s Betamax won over fans with its superior audio, VHS became the standard through its broader acceptance.

Holloway noted that integration of that search will take place in some of Microsoft’s popular programs:

The information focus sounds very familiar. IBM, the company that scares Microsoft more than Google, released its UIMA framework last year and made it freely available.

In an interview with Marc Andrews, director for strategy & business development for content discovery at IBM, he disclosed that IBM was focusing on concept as the operative function in search. More from that article:

If Microsoft does come up with a better search, to achieve the goal Holloway stated, UIMA may be the mechanism they use to get there.

—

Drag this to your Bookmarks.

Add to | DiggThis | Yahoo! My Web

David Utter is a staff writer for WebProNews covering technology and business.

IBM unveiled its source code release on SourceForge, the biggest open source development site, in a statement today. Developers can find the UIMA framework at Sourceforge, and additional items like the IBM UIMA SDK, with additional facilities and components, can be downloaded for free from IBM’s website.

The move demonstrated IBM’s confidence in the open source movement, and in its UIMA technology. In December 2005, IBM’s director for strategy & business development for content discovery Marc Andrews discussed UIMA within the context of search, and noted how UIMA is “information aware” rather than just data or application aware.

Unstructured data is the main idea to IBM’s concept approach. Whether information exists in an email, a text file, or a piece of rich media, UIMA’s framework can enable the construction of applications to retrieve the information from the container. Later in 2006, IBM plans to make the UIMA project a “full open source community development model.”

UIMA first started picking up notice in February 2005, and more formally disclosed at SES San Jose 2005 in August of that year. At that time, IBM promised to unveil UIMA on SourceForge by the end of the year.

Several firms like Factiva and Cognos created UIMA compliant solutions, IBM noted in its statement. Ongoing UIMA development in the medical field at places like the Mayo Clinic and Memorial Sloan-Kettering holds promise, too. Mayo wants to extract information from about 20 million clinical notes, while Sloan-Kettering has an even more ambitious plan:

—

Drag this to your Bookmarks.

Add to | DiggThis | Yahoo My Web

David Utter is a staff writer for WebProNews covering technology and business.

If you remember something about a particular movie, a quote or a scene, but not the movie itself, a search engine may be very helpful in making a connection. But if it’s something very obscure and a Google or a Yahoo can’t find it, it isn’t a real big deal.

What if it’s your physician trying to find out more about symptoms of an illness you possess? Does finding data related to those symptoms, locating information based on the symptoms, become a big deal for you?

Most likely the answer will be yes. The idea of “concept” as the operative function in search has become the place where IBM wants to take enterprise search. For IBM, it’s stopped being about keywords.

It’s about discovering the concepts in content.

Marc Andrews, director for strategy & business development for content discovery, spoke to me about how discovery works, and it all starts with UIMA, Unstructured Information Management Architecture.

I asked him if UIMA was application aware as well as data aware, and he indicated it went beyond that. “It’s information aware, whatever format information comes in, wherever it’s stored across the enterprise, and it’s really going beyond the traditional search.”

Andrews noted how search solutions don’t really go beyond web and file servers when it comes to spidering. “They’ve really ignored all of the enterprise knowledge that is being managed in their content management environment; that’s being stored in databases supporting their different applications, and potentially even in mainframes.”

Content discovery offers a more holistic view of data across the enterprise. “We’ve been focusing on enabling organizations to do a lot more than just search,” Andrews said. “One of the major limitations organizations have today is they’re limited to keyword-based search capabilities. That ends up falling short of most organizations’ needs.”

UIMA, which has been in development for about three years, could fulfill those needs. Andrews said development of the architecture began about three years ago. “It started out as a project in IBM Research because we had over 200 researchers working across 8 labs in six or seven different countries.

“They were all developing different types of text analytics. We needed to be able to have those different components interoperate with and build upon each other,” Andrews said.

By developing UIMA, IBM created an open framework that could accept those text analytic components as plug-ins, interpret the meaning of unstructured information, and identify concepts and facts that allow for the search for more than mentions of words.

Now let’s come back to the scenario where you visit the doctor with a variety of symptoms, a list that has even the most talented of clinicians shaking his or her head. It’s an appropriate topic since the earliest adopters of UIMA have been organizations like the Mayo Clinic and Sloan-Kettering Cancer Center.

“An expert knows to go in and search for ten different ways of describing the symptom. Your typical patient or even doctor refers to things in different ways. They want to be able to search for a symptom or search for any drugs that relate to the symptom and find all of that information,” Andrews said.

“They’re doing it for clinical trials research and drug research, to be able to identify concepts and facts. So those organizations are leveraging UIMA today to incorporate these types of analytics.

“Doctors and patients wanted to find out more about different clinical trials that are going on,” he continued. “These clinical trials are being managed by various organizations, information is scattered across NIH (National Institute of Health) and other different databases, and included clinical trials being conducted” by various pharmaceutical firms.

Medical centers and universities like Stanford, Carnegie-Mellon, and Columbia were running into search-related challenges. So was the organization that gave birth to the Internet, the Defense Advanced Research and Projects Agency (DARPA).

It was DARPA that sponsored the first working group for UIMA. From that working group, the first version of UIMA became available at the beginning of 2005. In August 2005, IBM announced its intent to make UIMA available as open source. Andrews said there have been about 3,000 downloads of the UIMA framework.

There’s some proof-of-concept work going on at some Fortune 500 companies with UIMA, but Andrews wasn’t in a position to discuss names yet. More information on that should be coming early in 2006.

Email the author here.

Add to | Yahoo My Web

David Utter is a staff writer for WebProNews covering technology and business.

However, because corporate documents do not contain a natural structure, meaning they are more unique and un-related, indexing the large amounts of business documents can be an arduous task. It’s this concept that drives Arthur Ciccolo, one of the chief developers of IBM’s Unstructured Information Management Architecture (UIMA) project.

While enterprise search may be considered a niche topic, many of the developments coming from UIMA, if applied to web-based search, could have incredible ramifications. However, this is not IBM’s goal. During a phone interview, Ciccolo stated in no uncertain terms that IBM’s goal for their search technology is the enterprise level, not web search.

To define UIMA and its function, IBM offers this:

Unstructured information represents the largest, most current and fastest growing source of information available to businesses and governments An Unstructured Information Management (UIM) application may be generally characterized as a software system that analyzes large volumes of unstructured information (text, audio, video, images, etc.) to discover, organize and deliver relevant knowledge to the client or application end-user.

In order to accomplish this task, Ciccolo and his team are putting their efforts developing different framework structures to perform text analysis, semantic comprehension, and natural language support. By doing so, IBM’s UIMA utilities can better perform the tasks of indexing and comprehending the different types of enterprise business documents.

To understand why enterprise search can be such a complicated excursion, you must first understand the different types of unstructured data that has to be indexed. With web-based documents, there is a much more narrow focus because document types are more limited (html, pdf, fla, etc.) and they usually contain links, which lends itself to easier indexing.

These off-page attributes provide the structure and makes web indexing less complicated than the unstructured environment of business documents. While web documents are more confined, enterprise documents run the gamut from word processing documents to video and sound files, and the off-page attributes that provide structure are absent.

The capabilities of UIMA, some of which are still in the developmental stages, attempt to address these concerns. For instance, the focal point of UIM is to focus on text analytics and the semantics contained within. By understanding the contents of the text being indexed, developing natural language search capabilities (“What is the formula for product X?”) is an attainable goal.

However, to understand the difficulties involved in developing a natural language search feature that works, consider this: the reason Microsoft will be releasing Longhorn without the search feature has to do with the developing a new file structure that supports natural search queries.

With UIMA, once a document is indexed, searching for it should be easier than it would if it was web indexed. Ciccolo indicated that once an item is indexed, their technology automatically generates editable meta data, which makes discovery much simpler. By integrating UIM technology with IBM’s Intranet search utility WebSphere Information Integrator OmniFind, they are able to provide an exciting era of enterprise-related search.

With regards to the current abilities of UIMA, the potential future developments are quite impressive, and if adopted, could cause huge ripples in the search technology status quo. While the list of possible developments is fairly long, there are two possible developments that could have long-reaching ramifications.

The first area of interest has to do with video search. Ciccolo and his team are developing video search methods that could revolutionize the whole concept. Normally, most video indexing is done by spidering the closed-caption text contained within the film. However, Ciccolo’s vision has to do with actually analyzing the picture to extract whatever relevant content is contained within. This data would then be indexed and have meta data generated (which would be editable using IBM’s service), making retrieval methods even better.

The other area of interest has to do with providing the ability to perform trans-lingual queries. What drives this development is the following concept:

- User A enters a query in English language
- UIMA translates query into target language
- UIMA then searches target language documents and,
- Returns search result in whatever language initiated the query. In this case, English

Arthur indicated that during the testing phases, he discovered search results were actually more relevant after the translation took place, meaning UIMA performs the translation after the query is entered. If something similar to this was adopted by search as a whole, it could and would alter the entire landscape of possibilities.

Other features and ideas of interest include the ability for administration members to write their own search algorithms, which can be implemented on top of existing framework. Speaking of admins, Ciccolo made sure to point out UIMA was developed with IT departments in mind. The technology is purposely made to be easy to install and to tweak. This makes adapting their search technology to fit your business much easier.

Another area that the team is focusing on is the medical industry. The ability to catalog and differentiate between the mountains of journals and documents over-running medical institutions is completely welcomed. Currently, IBM has an agreement with the Mayo Clinic to implement and test UIMA. If the medical field as a whole would adopt such technology, finding patient information, journals on particular illnesses, and pharmaceutical information would be much easier to accomplish. This in turn would undoubtedly improve the medical industry’s ability to treat and care for patients, as well as share information.

To understand the goal of the UIMA project, Ciccolo offers these thoughts, “IBM’s goal for UIMA is that it becomes widely accepted as a new class of middleware for analytics and that it enables the next generation of search: semantic search.”

For much more information about the project and other areas of IBM’s approach to enterprise search, please visit the following areas:

UIMA Homepage
The research proposal
About the authors
Information about OmniFind

While the subject can be tricky to navigate through, I would recommend reading the journals and documentations of Art Ciccolo and his team.

Chris Richardson is a search engine writer and editor for WebProNews. Visit WebProNews for the latest search news.