Embedding/Chinese-Word-Vectors

Project overview

This project provides over 100 pre-trained Chinese word vectors with varying representations and corpora, along with the CA8 analogy dataset and a Python evaluation toolkit. Researchers and developers can use these resources to assess dense and sparse word vectors for downstream NLP tasks.

Project type

Model Development · Data Processing

Deployment

Refer to project documentation

License

Apache-2.0

Best for

• Researchers, developers, and data teams focused on model training and customization who need diverse pre-trained Chinese word vectors and evaluation tools.
• Users needing a Chinese analogical reasoning benchmark covering comprehensive morphological and semantic relations.

Key capabilities

• Provides over 100 pre-trained Chinese word vectors trained with different representations (dense and sparse), context features (word, ngram, character), and corpora.
• Provides a Chinese analogical reasoning dataset CA8 specifically designed for Chinese, containing 17813 analogy questions covering comprehensive morphological and semantic relations.
• Provides a Python evaluation toolkit to evaluate the quality of dense and sparse word vectors using analogy tasks.
• Provides pre-trained dense word vectors trained with SGNS and sparse vectors trained with PPMI.

Limitations and risks

• The CA-translated benchmark only covers 134 Chinese words and three semantic questions.
• GPU and minimum hardware requirements for running the evaluation toolkit and processing the vectors are not documented.

Getting started

• Running the evaluation toolkit has an easy setup difficulty. First success path involves downloading the pre-built vector files and running the command: python ana_eval_dense.py -v <vector.txt> -a CA8/morphological.txt.

Evidence and sources

• README: This project provides 100+ Chinese Word Vectors (embeddings) trained with different **representations** (dense and sparse), **context features** (word, ngram, character, and more)…
• README: The pre-trained vector files are in text format. Each line contains a word and its vector. Each value is separated by space. The first line records the meta information: the first…
• README: In contrast, CA8 is specifically designed for Chinese language. It contains 17813 analogy questions and covers comprehensive morphological and semantic relations.
• README: $ python ana_eval_dense.py -v <vector.txt> -a CA8/morphological.txt
• README: All word vectors are trained by ngram2vec toolkit.

Reviewed at: 2026-07-19T18:45:13Z