MLN4KB: an efficient Markov logic network engine for large-scale knowledge bases and structured logic rules

Introduction

This repository contains the code for our paper:

Title: MLN4KB: an efficient Markov logic network engine for large-scale knowledge bases and structured logic rules.

Authors: Huang Fang, Yang Liu, Yunfeng Cai, Mingming Sun.

Affiliation: Baidu Research, Cognitive Computing Lab (CCL).

Quick start

Open Julia in terminal under this folder and go to the package REPL by pressing ], type activate . to activate the package. Then go back to Julia REPL by pressing the backslash.

The "smokers and friends" toy example:

Load packages:

push!(LOAD_PATH, pwd())
using Revise, mln4kb, Printf

MLN inference:

factFile = "./examples/smoke/facts.txt"
ruleFile = "./examples/smoke/rules.txt"

mln = MLN(factFile, ruleFile);
PrepareMLN!(mln)

# Extract facts
ExtractFacts(mln.kb, "Friends")

# Inference
objList, numViolatedList = WalkSAT!(mln, maxIter=Int(1e2), warmupPeriod=Int(1e2))

Weight learning:

iterate = zeros( Float64, length( mln.rules ) )
lr = 1e-1
optimizer = AdaGradOptimizer(lr, iterate)
OptimizePseudoLogLikelihood!( mln, optimizer, numNegativeSamples=1, maxIter=Int(1e2), resetMLN=false )

More test examples can be found in ./examples/run_examples.jl.

Citation

If you find this project helpful, please cite the code with the following bibtex.

@inproceedings{fang2023mln4kb,
  title={MLN4KB: an efficient Markov logic network engine for large-scale knowledge bases and structured logic rules},
  author={Huang Fang and Yang Liu and Yunfeng Cai and Mingming Sun},
  booktitle={The International World Wide Web Conference 2023},
  year={2023}
}

Contact

Please feel free to send your comments and contact us by fangazq877@gmail.com. We are considering to develop a C/C++ version of MLN4KB (with multi-CPU parallelization), please let us know if you find MLN4KB.jl is still too slow for your application.