RFC：issue-answer for osschat

[wj-Mcat]

Abstract

Newbies always have some many common problems which had been existed in issues or faqs. So when there is a new baby developer ask one of them, chatbot should give the right question which contains some references. And these references always contain the latent golden answer.

Motivation

Artificial intelligence has developed rapidly in recent years, especially in the field of chatbots. how to find the right answer for user is a fundamental feature which can be applied with many tools. In my opinion, it should first be used in the technical open source community.

Solution

There are two problems to be resolved:

1. Sync issue data
2. search the answer.

The first one is too simple to be discussed, so we will focus on how to search the answer.

There are some powerful deeplearning frameworks, eg: pytorch, jax, Paddle and awesome tools, eg: PaddleNLP, transformers to create our own nlp applications. So we can use one of them to create semantic searching model.

We can create the semantic searching model with the following steps:

1. create encoder model

2. create recall & re-rank stage to find the solid answer.

3. Use vector search engine to speed up searching.

Issue-answer can be implemented with different tools but the same steps.

Conclusion

Issue-answer is the basic feature for technical open source community, but not implemented util now. So, let’s create it.