previous

Introduction

Probabilistic Soft Logic (PSL) is a software package for reasoning about similarity and uncertainty in relational domains where exploiting dependencies and correlations yields better performance on prediction tasks. Such tasks occur in many areas such as information integration, computer vision, natural language processing, and machine learning in general.
PSL is a general-purpose framework for expressing, reasoning about and learning structural dependencies. PSL provides a declarative language tailored to relational domains that require reasoning about similarity and/or probability. Some of the novel aspects of PSL include a representation based on continuous valued random variables, efficient polynomial-time inference algorithms, native support for reasoning about sets, and the ability to estimate confidences values for predictions.

The following presentation provides a detailed account of PSL:

Publications

About PSL

  1. Probabilistic Similarity Logic, Matthias Broecheler, Lilyana Mihalkova and Lise Getoor, Conference on Uncertainty in Artificial Intelligence 2010
  2. Computing marginal distributions over continuous Markov networks for statistical relational learning, Matthias Broecheler, and Lise Getoor, Advances in Neural Information Processing Systems (NIPS) 2010
  3. A Scalable Framework for Modeling Competitive Diffusion in Social Networks, Matthias Broecheler, Paulo Shakarian, and V.S. Subrahmanian, International Conference on Social Computing (SocialCom) 2010, Symposium Section
  4. Probabilistic Similarity Logic, Matthias Broecheler, and Lise Getoor, International Workshop on Statistical Relational Learning 2009

PSL Applications

  1. Decision-Driven Models with Probabilistic Soft Logic, Stephen H. Bach, Matthias Broecheler, Stanley Kok, Lise Getoor, NIPS Workshop on Predictive Models in Personalized Medicine 2010

Release

PSL is implemented in Java and released under the Apache 2 license.

Use the links at the top of the page to get started.

You can join the psl-announce mailing list to stay up-to-date on PSL news. Messages are sent (infrequently) to announce new version releases and other important events.

Data Sets

Collective Classification of Wikipedia Documents:
We collected all Wikipedia articles that appeared in the featured list in the period Oct. 7-21, 2009, thus obtaining
2460 documents. After stemming and stop-word removal, we represented the text of each document
as a tf/idf-weighted feature vector. Each document belongs to one of 19 distinct categories, which were obtained
by using the category under which each featured article was listed. The data contains the relations
Link(fromDoc; toDoc), which establishes a hyperlink between two documents; Talk(document; user), which
states that the user edited the “Talk” page of the given document; and HasCat(document; category), which states
that the document has a particular category.

  1. The wikipedia data set as described above and used in the experiments of the original PSL UAI 2010 paper can be downloaded here.
  2. For the experiments on marginal computation in PSL we used a slightly modified subset of the above data set which can be downloaded here. This subset has pre-computed folds, uses only links as relational information, and uses a subset of the categories only.

Ontology Alignment
The Ontology Alignment Evaluation Initiative (OAEI) publishes different sets of ontology pairs together with reference alignments for evaluation and comparison purposes. For our PSL UAI 2010 paper we used the 2008 benchmark ontologies.

next