zhaoqj2016/pra

Repository for the Path-Ranking Algorithm

Repository for the Path-Ranking Algorithm

This repository contains the code used for the following papers:

• Relational retrieval using a combination of path-constrained random walks
• Random Walk Inference and Learning in A Large Scale Knowledge Base
• Learning relational features with backward random walks

If you use this code or results in your research, and wish to cite as appropriate:

@Article{lao2010pra,
  author="Lao, Ni
  and Cohen, William W.",
  title="Relational retrieval using a combination of path-constrained random walks",
  journal="Machine Learning",
  year="2010",
  month="Oct",
  day="01",
  volume="81",
  number="1",
  pages="53--67",
  issn="1573-0565"
}
@inproceedings{lao2011kb,
  title={Random walk inference and learning in a large scale knowledge base},
  author={Lao, Ni and Mitchell, Tom and Cohen, William W},
  booktitle={Proceedings of the Conference on Empirical Methods in Natural Language Processing},
  pages={529--539},
  year={2011},
  organization={Association for Computational Linguistics}
}

Instructions

Compiled jar file can be found at http://www.cs.cmu.edu/~nlao/data/pra-src-20140421.jar Data files can be found at http://www.cs.cmu.edu/~nlao/data/

1. Produce edges file from a NELL knowledge dump

arg1=NELL dump file arg2=probability threshold arg3=output format)

java -cp pra-src-20140421.jar edu.cmu.pra.data.WKnowledge createEdgeFile NELL.08m.165.cesv.csv 0.8 edges

1. index edge file to a compact graph reprensentation

arg1=graph folder which contains *.edges arg2=edges/db)

java -cp pra-src-20140421.jar edu.cmu.pra.SmallJobs indexGraph ../graphs/NELL165/ edges

1. create Queries

arg1=graph folder arg2=query folder arg3=whether its for training (boolean) arg4=whether adding the range of relation as part of the query (boolean)

java -cp pra-src-20140421.jar edu.cmu.pra.SmallJobs createQueries ../graphs/NELL165/ ./queriesR_train/ true false

1. Train model (or do predictions) for a relation with parameters specified by ./conf

java -cp pra-src-20140421.jar edu.cmu.pra.LearnerPRA

1. Train models (or do predictions) in batch mode.

arg1=path to java class files arg2=text file containing a list of relation names

java -cp pra-src-20140421.jar edu.cmu.pra.SmallJobs batchLearnerPRA pra-src-20140421.jar selected_relations