How do I write a small program that can automatically edit new year greetings

Implementation notes ：

Step one ： Only by building an artificial intelligence model can machines learn to speak like human beings , ad locum , Use python Language scripting , Stack based Transformer（ A module in a neural network , Self attention mechanism is used to realize the relevance of words in sentences ） The decoder module of GPT-2 Model , Adopted BERT The participator of （ Divide the words in the sentence ） Build a large-scale deep learning model .

Step two ： After building the model , There will still be some output results Bug appear .

At this time, we need to let computer algorithms learn how to speak like people , It means training him with a lot of data （ Learn how to talk ）, We first adopted 100GB Chinese corpus CLUECorpus2020 data （ It covers all kinds of written materials collected on the Internet ） pretraining , Reuse over 10 Ten thousand Chinese lyrics （ You can only speak well if you are well informed ）, Then collect all kinds of new year blessings , Fine tune the model （fine-tune - Fit the context ）

Step three ： The trained model can output the results written by artificial intelligence , There are still some problems with this result , But they are all small problems , You need to do some cutting and formatting on the string .

Step four ： When it's done , You can see the example of black box printing , But this example is not user friendly enough , And only one person can use , How can we make people use it on their own computers , We need to turn it into something that everyone can access .

Step five ： So I used Tencent cloud GPU The server , It has been configured on Tencent cloud server . Turn the artificial intelligence program we have written into one that everyone can access web service , Is to ask for it from all over the world , Ask him to output something , He can return a result .

Step six ： But it's not easy enough , We use vue The framework compiles a set of human-computer interaction interface , Call the just written web Interface , Then export the results .

Step seven ： Then we will release our human-computer interface , Another configuration has been carried out on Tencent cloud , Our website is finally online ！ Everyone can easily and happily visit and share on wechat .

Finally, dezai wants to say , Natural language generation is inseparable from the excellent work of various researchers in recent years , At present, if artificial intelligence is used to generate different words （“ Based on artificial intelligence to create, rather than a simple combination of vocabulary ”） It has become possible .

Scientific research achievements used ：

Radford, A., Narasimhan, K., Salimans, T., & Sutskever, I. (2018). Improving language understanding by generative pre-training.

Devlin, J., Chang, M. W., Lee, K., & Toutanova, K. (2018). Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805.

Xu, L., Zhang, X., & Dong, Q. (2020). CLUECorpus2020: A large-scale Chinese corpus for pre-training language model. arXiv preprint arXiv:2003.01355.

Ethayarajh, K. (2019). How contextual are contextualized word representations? comparing the geometry of BERT, ELMo, and GPT-2 embeddings. arXiv preprint arXiv:1909.00512.

Developer：

Tianrun Chen（ Neural network algorithm ） Zhengjie Zhang（web Human computer interaction page ）