Answerable is a python program that finds unanswered questions on Stack Overflow that you could be able to answer. The learning is performed on the similarity of the new questions with the one you've answered.
Pondering by the reputation associated is in the wish list for now.
At first I didn't know about the stackexchange API, so I planned to do it via web scraping. However, when I saw the amount of traffic that could mean, I decided did a little research and found this site.
Web scraping has already been replaced by the API and RSS.
The reason to use the RSS feed instead of regular API requests is the quota. I want this software to be free and open, so it won't use access_token. As a consequence, the API will be used only to recover the minimal information necessary to elaborate the learning model. Any thoughts on this are welcome.
Supporting other stackexchange sites is not a priority right now, but maybe when a fully functioning version is finished, I'll work on enabling it.
It is free software under the MIT License.
The code is fully in Python 3.8, but I still plan to experiment with Prolog to make some semantic analysis in a future separate branch.
If you want to contribute, feel free to open an issue, make a PR (Check the contributing guidelines in the repository) or send me an email.
Update: Now that there is a fully functional version, I will focus on improving the recommendations. Here I'll keep the observations of the current model.
Observations about the current model
- It doesn't make a distinction between text that is code and text that is natural language.
- Short questions tend to have priority over long questions.
- The general similarity to the questions you have answered may have more weight than it should, compared to the similarity with the more similar question that you have answered. This can reduce the quality of the results when the user has a long activity history.