Tutorial

This repository contains the implementation for two Chinese word embedding learning methods using semantic and phonetic components, i.e., rPCWE and rsp2vec.

0. Environment

• Linux
• gcc
• cmake
• gensim==4.0.0

1. Download the dataset

The processed Chinese wikipedia dataset can be downloaded here.

Please unzip the downloaded data under the folder data/.

2. Train

rPCWE

To train the rPCWE model, run the following commands under the rPCWE/src/ folder.

$ chmod 777 -R .
$ bash run.sh

rsp2vec

To train the rsp2vec model, run the following commands under the rsp2vec/ folder.

$ chmod 777 -R .
$ bash run.sh

3. Evaluate

Evaluate the performance of the trained embeddings on the relevance and analogy tasks by running the following commands:

$ python eval.py rPCWE/src/word_vec-3-win7-size150-it75.txt
$ python eval.py rsp2vec/rsp-150.vec