/SpRL_TextOnly

This repo is for spatial role labeling using text only data.

Primary LanguageScala

Spatial Role Labeling

This is a text-based baseline model for spatial role labeling task.

Prerequisites:

  1. HetSaul Library
  2. Gurobi optimization tool

--This project uses HetSaul library as a dependency. Before running project you need to clone to https://github.com/HLR/HetSaul.git and assuming you have sbt, do sbt publish-local, this should be all you need to do to be able to use the HetSaul project as a dependency.

--To download and install Gurobi visit Gurobi Website Make sure to include Gurobi in your PATH and LD_LIBRARY variables

export GUROBI_HOME="PATH-TO-GUROBI/linux64"
export PATH="${PATH}:${GUROBI_HOME}/bin"
export LD_LIBRARY_PATH="${LD_LIBRARY_PATH}:${GUROBI_HOME}/lib"

Dataset

Please download the dataset from [here] and copy in the root folder to have:SpRL_TextOnly/data and make sure to download google vectors and have it in /data/GoogleNews-vectors-negative300.bin. The layout of the data should be as follows:

Data/
|-- mSpRL      
|   |-- saiapr_tc-12
|   |  |   
|   |  |-- newSprl2017_train.xml
`------`-- newSprl2017_gold.xml

## SpRLApp

All configurations needed to run the main application are placed in 
[`tripletConfigurator`](tripletConfigurator.scala). In order to run the application for training, set `IsTraining` to `true` and for testing set it to `false`. 

### Setting models
 [BM] is the baseline model and is set using `val model = FeatureSets.BaseLine` and disable using images and preposition classifiers
 `val usePrepositions = false` and `var populateImages = false`. 
 You need to set the `isTrain= True` and run the MultiModalTripletApp to train models and when training finished change to `isTrain= False` 
 and rerun the MultiModalTripletApp. You can see the results on screen as well as in `SpRL_TextOnly/data/mSpRL/results/triplet_BM.txt` file.

## Running

To run the main app with default properties:

```sbt "project SpRL_textOnly" "runMain edu.tulane.cs.hetml.nlp.sprl.TextOnly.SpRLApp"

results will be saved in data/mSprL/results folder as text files corresponding to the model selected in the config file.

Results on CLEF 2017 dataset

Here are the summarized results of relation classifier for a Baseline model and the baseline plus constraints:

-----------------------------------------------------------------------------------
BM                              65.640     60.226     62.817     885        812
BM+Constraints                  70.036     66.554     68.250     885        841