Cross-lingual Learning-to-Rank

• This repository is about Cross-lingual Learning-to-Rank with Shared Representations. Shota Sasaki, Shuo Sun, Shigehiko Schamoni, Kevin Duh and Kentaro Inui. NAACL2018
  • https://www.aclweb.org/anthology/N18-2073/

Table of contents

• Usage
  • Requirements
  • How to inference
• Resource

Usage

Requirements

• Python version 2.7
• chainer == 5.4.0
• numpy == 1.15.0

How to inference

$ python src/inference.py \
--vocab_path [path_to_vocaburaly_directory] \
--model_path [path_to_model_snapshot] \
--data_path [path_to_target_data] \
--doc_lang [document_language] \
--n_hdim [for_feed_forward_layer]

For example,

$ cd resource
$ wget https://github.com/losyer/clir/releases/download/results_v1.0/sw.tar.gz
$ tar -zxvf sw.tar.gz
$ wget https://github.com/losyer/clir/releases/download/vocab_v1.0/vocabulary.zip
$ unzip vocabulary.zip
$ cd ..

$ python src/inference.py \
--vocab_path resource/vocabulary \
--model_path resource/sw/in_language/deep_model_100/model_epoch_19 \
--data_path resource/input_examples/train_sw_l100.txt \
--doc_lang sw \
--n_hdim 100

Resource

See https://github.com/losyer/clir/tree/master/resource