In a sense, RFCs - the specifications of Internet protocols that are published by the Internet Engineering Task Force (IETF) - are like cooking recipes: they explain how to implement a protocol, but do not usually explain why certain design decisions were made.
However, RFCs are also often are the result of several years of discussions, and most of the discussion text is available, e.g. via github, mailing lists or meeting minutes.
For some RFCs, we have discussion text associated with paragraphs available, since we were able to download the issue text associated with pull requests. By analyzing the length, number of participants, and the sentiment (applying sentiment analysis from a Natural Language Processing (NLP) model), we can get an idea of how heated the debate was - and a tool could, for example, give paragraphs a different colour depending on the length and "heat" of the debate, along with a link to the debate itself.
In this master thesis, we want you to write such a tool. We envision the tool to offer such information about the discussion underlying paragraphs for where we have them, and Internet-drafts (earlier versions of RFCs) for when we only have mailing list data available. The idea is to facilitate getting a better understanding of a protocol - e.g.:
- which aspects of the design are compromises?
- which parts were just written like they are by the authors, and never even discussed?
- does the protocol have elements that were "pushed through" by the authors despite criticism which was simply ignored?
As a by-product of the development of this tool, we expect to better understand the limitations of existing pre-trained NLP models for sentiment analysis.