Campus Event Calendar
Event Entry
What and Who
Infinite Hidden Relational Models
Zhao Xu
siemens
Talk
AG 1, AG 2, AG 3, AG 4, AG 5, SWS, RG1, RG2
Public Audience
English
Note: We use this to send email in the morning.
Date, Time and Location
Tuesday, 12 June 2007
11:00
-- Not specified --
E1 4
433 (Rotunde 4th floor)
Saarbrücken
Abstract
Statistical relational learning analyzes the probabilistic constraints
between the entities, their attributes and relations. We apply nonparametric
Bayesian mixture modeling to relational data and propose infinite hidden
relational models (IHRM). An IHRM extends the expressiveness of a relational
model by introducing for each entity an infinite-dimensional hidden variable
as part of a Dirichlet process mixture model. There are mainly three
advantages. First, this reduces the extensive structural learning, which is
particularly difficult in relational models due to the huge number of
potential probabilistic parents. Second, the information can globally
propagate in the ground network defined by the relational structure. Third,
the number of mixture components for each entity class can be optimized by
the model itself based on the data. IHRM can be applied for entity
clustering/classification and relation/attribute prediction, which are two
important tasks in relational data mining. We also develope the inference
algorithms for IHRM, including Gibbs sampling with the Chinese restaurant
process, Gibbs sampling with the truncated stick breaking construction, and
the corresponding mean-field approximation. The performance of IHRM is
evaluated in two different domains: a recommendation system based on the
MovieLens data set, and prediction of the function of yeast genes/proteins
on the data set of KDD Cup 2001.
The experimental results show that IHRM gives significantly improved
estimates of attributes/relations and highly interpretable entity clusters
in complex relational data.
between the entities, their attributes and relations. We apply nonparametric
Bayesian mixture modeling to relational data and propose infinite hidden
relational models (IHRM). An IHRM extends the expressiveness of a relational
model by introducing for each entity an infinite-dimensional hidden variable
as part of a Dirichlet process mixture model. There are mainly three
advantages. First, this reduces the extensive structural learning, which is
particularly difficult in relational models due to the huge number of
potential probabilistic parents. Second, the information can globally
propagate in the ground network defined by the relational structure. Third,
the number of mixture components for each entity class can be optimized by
the model itself based on the data. IHRM can be applied for entity
clustering/classification and relation/attribute prediction, which are two
important tasks in relational data mining. We also develope the inference
algorithms for IHRM, including Gibbs sampling with the Chinese restaurant
process, Gibbs sampling with the truncated stick breaking construction, and
the corresponding mean-field approximation. The performance of IHRM is
evaluated in two different domains: a recommendation system based on the
MovieLens data set, and prediction of the function of yeast genes/proteins
on the data set of KDD Cup 2001.
The experimental results show that IHRM gives significantly improved
estimates of attributes/relations and highly interpretable entity clusters
in complex relational data.
Contact
Ellen Fries
502
--email hidden
passcode not visible
logged in users only
Ellen Fries, 06/04/2007 12:29 -- Created document.