Skip to content
This repository has been archived by the owner. It is now read-only.
Torch implementation of Severyn and Moschitti's SIGIR 2015 CNN model for question answering
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

Reimplementation of Severyn and Moschitti's CNN for Ranking Short Text Pairs

NOTE: This repo contains code for the original Torch implementation from SIGIR 2017. The code is not being maintained anymore and has been superseded by a PyTorch reimplementation in Castor. This repo exists solely for archival purposes.

This repo contains code for the reproduction of Severyn and Moschitti's CNN for Ranking Short Text Pairs from SIGIR 2015 [1]. Our findings were published in a SIGIR 2017 short paper [2].

Getting Started

1. Please install the Torch library by following instructions here:

2. Using following script to download and preprocess the Glove word embedding:

$ sh

Please make sure your python version >= 2.7, otherwise you will encounter an exception when unzip the downloaded embedding file.

3. Before you run our model, please set the number of threads >= 5 for parallel processing.

$ export OMP_NUM_THREADS=5


Run the original SIGIR'15 model with external IDF features on TrecQA raw dataset:

$ th trainSIGIR15.lua -dataset TrecQA -version raw -train train-all -model conv -sim bilinear -ext

Run linear model with external features:

$ th trainSIGIR15.lua -dataset TrecQA -model linear -ext

All model options described in our paper [2] can be specified through the above command line parameters.


[1] Aliaksei Severyn and Alessandro Moschitti. Learning to Rank Short Text Pairs with Convolutional Deep Neural Networks. SIGIR 2015.

[2] Jinfeng Rao, Hua He, and Jimmy Lin. Experiments with Convolutional Neural Network Models for Answer Selection. SIGIR 2017.

You can’t perform that action at this time.