The followings are some of our representative research papers.

Notes: The Language collumn in the following tables indicates that the models are evaluated on those languages in the paper. It does not mean the model will not work on other languages.

Model Recommendation: DSG provides 200-dimensional word embeddings for around 8M Chinese words. ZEN 2.0 provides large pre-trained language models (the large version uses 24 layers of self-attentions with 1024 dimensional hidden vectors) for Arabic and Chinese. The models are trained on large corpus and enhance text modeling through n-grams. ChiMed-GPT is a Chinese medical large language model (LLM) built by continually training Ziya-v2 on Chinese medical data, where pre-training, supervised fine-tuning (SFT), and reinforcement learning from human feedback (RLHF) are comprehensively performed on it.

Model Recommendation: WMSeg and McASP contains easy-to-use CWS and joint CWS and POS tagging models that are based on BERT and ZEN. Models trained on different datasets are available for downloading.

Model Recommendation: SAPar provides constituent parsers (which are based on BERT, XLNet, and ZEN) for Arabic, Chinese, and English; DMPar provides code for dependency parsing; NeST-CCG offers BERT-based models for English CCG supertagging. Both repositories provide pre-trained models and they are easy-to-use.

Model Recommendation: SANER use pre-trained language models and word embeddings in text modeling, with the semantic of similar words are used to enhance text understanding. Pre-trained models are available for downloading and they are easy-to-use.

Model Recommendation: Pronoun-Coref uses GloVe and ELMo embeddings in text modeling. The model is light and easy-to-use.

Model Recommendation: DGSA provides an end-to-end solution (the model are based on BERT) for aspect-level sentiment analysis, which can be directly used to process raw text.

Model Recommendation: RE-AGCN provides BERT-based models for relation extraction, where the model leverages the auto-parsed dependency tree of the input text to have a better understanding to the text.

Model Recommendation: T-DNA is a Transformer-based language model for domain adaptation, which can be used easily.