Knowledge from the Web: Understanding and Reasoning
Gjergji Kasneci (Microsoft Research)
Info
|
Date |
10th May 2011 (week , Trinity Term 2011) |
|
Time |
11:30 |
|
Place |
147 |
Abstract
In recent talks, the Vice President of Bing Search, Harry Shum, has repeatedly expressed the ambition of
turning Bing into an intent, knowledge and decision engine. The proposed strategy is a shift from the current hit-or-miss
model (in which the user intent is either captured or missed by the 10 blue links) to a dialog model, which would allow
users to specify their information need in several steps in an entity-centric way. If this strategy is to be taken seriously,
we need to think about new techniques for understanding the knowledge on the Web and reasoning about it.
From
a research perspective, in this talk I argue that in order to overcome the hit-or-miss model and go beyond single-page Web
search, there is a need for combining knowledge fragments (i.e., facts about entities and relationships) from various Web
sources. For recommendation and decision support, reasoning about the uncertainty of such knowledge fragments is crucial;
I will present a family of increasingly complex probabilistic models that allow us to quantify their uncertainty by efficiently
leveraging the 'wisdom of the crowds'. This can be done by jointly modelling the reliability and expertise of judges together
with the truth values of knowledge fragments. I will conclude the talk by presenting other on-going projects towards a more
knowledge-oriented Web.
Further info
|
Related series |
|
