Presenting at San Diego Semantic Web Meetup

In a few weeks, I will be giving a seminar for the San Diego Semantic Web Meetup group. Its going to cover a fair amount of the same material we covered at the NCBO webinar, but you should come anyway because a) I have new data and b) I'm going to be there live (so you can bring your rotten tomatoes!).

** Update, I am now excited to be presenting with Denny Vandrecic, an expert on ontology evaluation and one of the founders of the Semantic Media Wiki.

Peter Mika at SD Semantic Web Meetup tomorrow

For anyone in San Diego with an interest in the Semantic Web, tomorrow's meetup is a must. The presenter for the day will be Peter Mika from Yahoo! Research in Barcelona, stopping in on his way to SemTech. Peter gained early career fame for investigations into the relationships between social networks and the semantic web such as the award winning "Ontologies are us: a unified model of social networks and semantics". Abstract from the Meetup site below, hope to see you there.

"While current search techniques aim at ever more sophisticated methods for searching over hypertext, the Semantic Web promises to break boundaries in search by transforming the content itself into a form that is more easily processable by machines.

In this talk, we will discuss some of the possible technologies for annotating content for machine processing and showcase some of the ways that semantic annotations can improve the search experience for users. In the first part of the talk, we will describe existing and upcoming formats for embedding metadata inside web content, including microformats, RDFa, Facebook's Open Graph format and HTML5's microdata. We illustrate these formats with practical examples, show some of the tools that can help authoring and give guidance on best practices and point out some of the pitfalls, drawing on our experience from working with large publishers. In the second part of the talk, we will discuss research in semantic search and demonstrate its applications in Semantic Web search engines and web search engines with semantic extensions, including Yahoo's Search Monkey and Google's Rich Snippets."

Jerzy Lewak at SDSW Meetup

Image via Wikipedia

I just got back from another interesting San Diego Semantic Web Meetup, here are my notes before I forget.

The presenter this evening was Jerzy Lewak, a former theoretical physicist, a professor emeritus at UCSD, and cofounder of (at least) Nisus Software and SpeedTrack Inc.. (Once again, I continue to be impressed at the level of the speakers that are recruited for these events!). Jerzy presented the history and current applications of a novel human interface for databases that he calls GIA (Guided Information Access) that is made possible by the underlying TIE (Technology for Information Engineering) framework.

Jerzy began his talk as a true computer scientist by defining 'The Problem' that originally inspired this work (back in 1991). At that time email was just starting to pick up steam and already the task of finding old emails was becoming unmanageable. So, he set out to find a better way. (Always nice to be working on a solving a problem for yourself.) His search for a solution took a decidedly familiar path:

1. Hmm, why don't I set up some nice hierarchies of concepts to place my emails into so it will be easier to find them later?
2. Darn.. that really isn't working out very well. The more data I get, the harder it is to organize and I keep running into the problem that almost every single item in my collection might be placed under more than one category. Perhaps a physical filing cabinet is actually a terrible thing to base a completely virtual information storage and retrieval system on... (Though he didn't bring it up, he was describing exactly the same thing that Clay Shirky got so excited about in the 'Ontology is Overrated' essay in 2005 that - in turn - got me all excited in 2006, except of course Jerzy was thinking in 1991. And, as pointed out to me by my LIS friend Joe, this basic problem and the following conclusion were pretty well fleshed out by Ranganathan in the 1930's...).
3. After experimenting with plain content search (a la Google) he arrived at faceted classification as the most powerful and flexible way to describe and access data.

So far so good, I am paying attention.

Now the problem arises that many combinations of facets actually produce zero results. As he noted, just 200 facets (he uses the word 'selectors') is enough to uniquely describe every particle in the universe. This became the real problem. The breakthrough that got him going and has led to SpeedTrack and everything else he presented was the idea of dynamically limiting the potential facets based on those that are already selected. In the interfaces he demonstrated, he would:

1. Choose some database field like 'last name' and type in a name like 'Smith'
2. Show two things - one, the number of results went down and two, the number of possible values for the other fields (e.g. 'first name', 'height', 'date', etc.) would immediately be constrained to only show values for objects linked to Smiths in the database.
   

His interfaces might be described as an advanced multi-parameter type-ahead. They work by guiding the user (GIA) in the creation of potentially very complex queries that are guaranteed to return results. This is achieved by dynamically exposing the underlying indexes that drive boolean queries.

The system works on both structured and unstructured data (he showed a quick example of newspaper articles) but requires fairly heavy manual labor to get it started. Overall it looked like it would be useful and fun to use and I can imagine many potential directions they could take it.

My only complaint for the talk was that there was absolutely zero Web in it - no mention of any native ability to consume or produce RDF, no mention of OWL, no discussion of scaling possibilities, and the proverbial elephant in the room of large-scale data access was left more or less untouched. I guess that might be one sign of a good talk - I got interested and it left me thirsting for more..

Jans Aasman From Franz Inc. Presents at San Diego Semantic Web Meetup

Yesterday, about 20 people gave up 4 hours of sunny Saturday afternoon to gather in Carlsbad to listen to Jans Aasman present a tutorial on AllegroGraph - what Franz Inc. is touting as "Web 3.0's Database".  The attendees, some of whom drove all the way down from Santa Barbara, included a number of people interested in semantic search and search engine optimization.  There were also software developers, CEOs of software development companies, bioinformaticians, an orthopedic surgeon, business development specialists, and entrepeneurs.  A challenging crowd to give a tutorial too that all would find useful.

Since I ended up having to take my notes on my iphone and the presentation lasted nearly 4 hours, I'm only going to hit a couple of the things that stood out for me here.  For more info. on the products presented, the Franz Inc. website has extensive information - and free downloads!  I also hope to find a link to the slides at some point.

First of all, wow - well done organizer!  Its not everyday that you get to hear a presentation by the CEO of one of the oldest (is it the oldest?) living companies involved in artificial intelligence.  (Yeah I said it.. thats what we used to call this stuff.)  Franz has apparently weathered the storm and now that semantics is coming back to life under a new name - has come out doing very well with clients including: Lilly, GlaxoSmithKline, Adobe, Raytheon, Kodak, Boeing, Cisco, Mayo clinic, Novartis, and many others.   These clients are paying for the pro versions of tools like AllegroGraph and Gruff as well as consulting services related to the use of these tools to solve specific problems.  Jans began his presentation with a really rapid explanation of why these companies are spending their money like this rather than on continued development with the encumbent technology - relational databases.

Jans suggested that triple stores like AllegroGraph are more useful then relational databases when any of the following constraints are met:

1. You have many classes of complex objects
2. The properties/definitions of these classes change frequently
3. You want to work with rules/reasoning
4. You have a big graph in your data that you want to analyze 

Of those he hit on #2 the most often.  Basically, if you can perfectly conceive of the entire world that you need to model at the time you create your database schema you are probably going to be better off using the more mature rdbms technology.  If not - and this is the likely case for more projects then your database architect would have you believe - the triple tech. will make change much easier.

After that brief, high-level motivator he dived into a fairly extensive series of demos/examples.  While the subject matter varied from news stories to clinical trials to digital photo websites, there were two consistent themes that I found interesting.

1. Every example he gave involved a named entity extraction step at one point or another.  This indicates that, while there is a growing amount of structured data out there, there is still way too  much information held in text to ignore when doing any real analysis.  So..  you are going to need to get friendly with some people that do natural language processing - perhaps like this company Alchemy.
2. Most of the demos were conducted in Gruff, the visual interface to AllegroGraph.  In Gruff, he repeatedly showed off the very cool trick of selecting two, apparently unrelated nodes in the graph (say a two different drugs) and asking for connections between them.  AllegroGraph responded to this request ~instantaneously with relevant and useful results.  Very impressive technically (this was a graph with millions of triples and the system was running on his laptop) and I think very useful for discovery.

Wish I could write more but its the weekend after all..  For more information on this stuff, go to Franz inc., grab yourself a free download or two, and try it out yourself!  I'd be curious to hear how you find it.

Eric Little presents at San Diego Semantic Web Meetup

Just returned from my first Semantic Web Meetup in San Diego, here are some impressions:

In comparison to the Vancouver SW meetups that I have attended

1. Quite a different crowd. Larger group, zero t-shirts, generally older,
   stronger industry representation, heavily weighted towards biomedical
   folks - even an actual surgeon in attendance.
2. As with everything in Southern California, it involved a fairly long drive.
3. If I was a small fish in Vancouver, I'm a microbe here.

A few bullets from the (very good) talk

1. The presenter, Eric Little, did his PhD with Barry Smith (in formal biomedical ontology) and came from a background in philosophy. He is now the chief knowledge engineer at CTG - a fairly large (revenue in the 100's of millions) Information Technology and Consulting company.
2. He talked for a while about how ontologies should provide "actionable intelligence" and gave some nice stories about the benefits gained when links are formed across semantic groups - for example, they created a diagnosis ontology by linking the disease ontology to the LOINC (logical observations, identifiers, names, and codes). Kind of the same old story here, but good examples.
3. He then got to the meat of the talk (according to the abstract), which was about their MedMap application. He, correctly IMHO, pointed out that pretty much every attempt at an interface for dealing with large ontologies more or less sucks. No offense... He then proceeded to claim that their solution in the MedMap application was much better. The comparison to things like Protege and Concept Map builders was a bit unfair as the MedMap application is targeted towards end users rather than ontology engineers but, I can't say I disagree with the picture he painted.
4. The MedMap application is a very nifty widget-based approach built using Adobe Flex on the top and with combination of Top Quadrant products underneath (notably SPARQLMotion). It looks a lot like the iGoogle interface, with a user-customizable collection of what he called "analysis portlets". Each of the portlets has a dedicated task, but they all sit on top of the same massive knowledge base. The examples he gave were clinical - for example, he showed one scenario where patients were being automatically classified into different risk groups with a nice little graphical representation of the different classes; however, the technology and many of the interface modules are clearly fairly general purpose. He mentioned that they are in fact working in quite different domains such as the petroleum industry.
5. To get your own MedMap-like solution for your problem, starting rates are on the order of $300-400,000 for about a 3 month long push for a first usable product. They are after big fish like the health insurance industry.
6. His opinion was that the only companies that are likely to be interested in buying semantic IT solutions are those that a) have a lot of money, b) are more forward-thinking than average, and c) have already tried other approaches (e.g. relational databases) and watched them fail.

Personal notes/reactions

1. The sheer slickness of that Flex interface still has me all in a quiver.. It makes me want to delete everything we have been doing on my current attempt at a semantic application (I am now embarrassed to put in the link) and start over - or perhaps it is just time to concede that its impossible to build Ferraris like I what just saw without employing a staff of people that know what they are doing.
2. It was cool to reconnect with Barbara Starr, my former supervisor at SAIC - which was prior to my MSc, my foray into bioinformatics, and my PhD! Grey hairs...
3. It was also very cool that she is about to start working on an application of the GoodRelations ontology by Martin Hepp. I am proud to say that Martin sent me unsolicited positive feedback about my first research project in grad school several years ago. He concurred with my claim that ontology development/maintenance practices were too expensive given typical knowledge engineering practices and that techniques that tapped into the collective knowledge of the web (a la 2.0) were the way to go. (Note that this contention is in stark contrast to the ontologies of "reality" from Barry Smith.)

All in all a very good meeting and I am very much looking forward to the next one.

Authority, ontology, and Freebase

Something I knew about but hadn't thought through until Kirrily Robert pointed it out to me yesterday at the Vancouver Freebase meetup is the importance of the 'as_of_time' parameter in freebase.  This allows developers to access views of freebase data at any point in time.  The important consequence of this is that it makes it easy to create authoritative versions of data in freebase.  This means that if, for example, the gene ontology were loaded into freebase on a particular day by a particular trusted authority, users of freebase could access versions of the GO topics exactly as they appeared at that moment.  This answers the common worry from biologists about trust and authority.  The 'oh no, I wouldn't use anything that could be edited by just anyone' phenomenon goes away and the best of both worlds becomes a possibility.  With an open app like freebase you get the wikipedian potential for rapid, parallel, collaborative knowledge formation.  With this kind of versioning, it should be possible to create trusted, authoritative versions at any point in time.

huge.