Code for ArXiv submission titled "TegFormer: Towards Better Topic-Consistency and Text-Logic on Topic-to-Essay Generation" (https://arxiv.org/abs/2212.13456)
CS Concepts: Natural Language Processing --> Natural Language Generation --> Topic-to-essay Generation
Step1: Download data from https://pan.baidu.com/s/1_JPh5-g2rry2QmbjQ3pZ6w
Step2: Download gpt2 from https://huggingface.co/uer/gpt2-chinese-cluecorpussmall and put downloaded files into TegFromer/Essay/model/gpt2/
Step3: cd TegFormer/Essay/data, run TextRank.py and then run Extend_topics.py
Step4: cd TegFormer/Essay, run train.py
Step5: put test.txt into TegFormer/Essay/data
Step6: cd TegFormer/Essay, run infer.py and you can get generated text from TegFormer/Essay/data/gen.txt
Step1: Download data from https://pan.baidu.com/s/1eC4gb_We33kr-ZbHn3KdIA
Step2: Download gpt2 from https://huggingface.co/uer/gpt2-chinese-cluecorpussmall and put downloaded files into TegFromer/Zhihu/model/gpt2/
Step3: cd TegFormer/Zhihu/data, run TextRank.py and then run Extend_topics.py
Step4: cd TegFormer/Zhihu, run train.py
Step5: put test.txt into TegFormer/Essay/Zhihu
Step6: cd TegFormer/Zhihu, run infer.py and you can get generated text from TegFormer/Zhihu/data/gen.txt