Mark Greaves of Vulcan at ISWC
I’m at a panel of Industry Talks at the International Semantic Web Conference.
The first presenter is Mark Greaves from Vulcan (Paul Allen’s research lab), who is talking about Halo, an interface to a Semantic Wiki that they are funding (but not building). About blending Semantic web as (1) “enterprise data integration framework” and as (2) “web scale big messy data” aimed at giving users a better online experience—enhance content creation, publishing, linking, forming communities. He asserts that strand 2 still requires strand 1 (which I don’t believe, but let’s see where he goes with it). Mediawiki is web-scale consensus authoring around text. Semantic wiki is for same around structured data. Hypothesyses that wiki-style consensus does emerge around structured data. Halo is a Semantic MediaWiki extension. Ads ontology browser, semantic toolbar for authoring, semantic retrieval capabilities. The real question: do we actually achieve semantic convergence (if not, all you get is chaos in the data). Did test, of grad students in chemistry domain; used AP exam. In initial version people would not even use each others’ property names (no convergence). Went back, revised all UIs. In new version, are getting semantic convergence. People were starting to agree on vocabulary and data classes. Lessons learned:
- UI design matters. People must be brought to convergence on common vocabulary by proper UI
- formal usability study was necessary to discover this (surprise!)
- Schema last data engineering
- User created ontologies “aren’t very good.” flatter than normal (no deep hierarchies). but sufficient for common uses. (I’m not sure what he means by not good, if they are what the user wants. Does the user want a different ontology than the one they created? Why? How do you know?)
Project Halo (which I guess is different from the Halo UI for semantic mediawiki) is a project aimed at scientific question answering. Inspired by Dickson’s Final Encyclopedia. Text search can’t answwer scientific questions because they cannot reason about the subject matter, e.g. what is the outcome of a particular chemical reaction. Goal is to build a question answering system that can be filled out by subject matter experts rather than knowledge engineers. They combined work on project Aura (an old chemistry knowledge capture project) onto semantic mediawiki, and it worked (I got distracted by a reboot when he was explaining the details).