Need for a better Autoresponder for the discussions

[caco3]

Originally posted by @friedpa in #2637 (comment):

@SybexX The problem with the tool-tips/wiki is, that you and @caco3 spending a lot of time for explaining things over and over again. It seems, that a lot of users are simply lazy, why reading the tool-tips/ wiki, if I can get an answer without using any brain-power...
In other Git-hub forums there is a standard text send out pointing to the tool-tips/ wiki... just my fifty cents....

[caco3]

@friedpa Thanks for your input. I already added a system with auto responses based on labels, see eg. here: #2280 (comment)
How ever it is quite complicated to maintain and the labels must be set manually, thus it was mainly me who was using it.

What we would need is a more intelligent system. The tricky part is to analyze the users comment and automatically finding out what he actually needs. I think we could setup a better system which just would search for trigger words (eg. "reflection") and the respond with a link to the documentation. How ever setting up this word list would be manual work and I would need help on building it up.

Idea:
A response could look something like this:

---

Hi [user]

I see you used the words [trigger words].

Did you jet check the documentation?
If you have not done yet, I suggest to study the following documentation sections: [Link to documentation page or parameter or, ...]

If this all does not help and you need a developers to look into it (after you really studied the documentation), write a reply with the text dev-help-needed.

---

The tricky part is: I usually sporadically check the discussions. But to save time I only open those with zero responses. This would not work anymore this way. That's why I would add the dev-help-needed which could trigger another action which sets a label so we devs can easier find it or even mentions us so we get an email.

[friedpa]

@caco3 I could help with searching for the key-words used. I have stored in my Git-Hub Inbox always the last year of all comments.
So I could start analyzing the questions and the "standard" answers, if any. Your opinion?

[caco3]

I have stored in my Git-Hub Inbox always the last year of all comments.

I was able to export all discussions as JSON, so we have all data available and can test the algorithm offline.
Let me implement a PoC and then I would be very happy with your help tu build up the database.

[friedpa]

English or Deutsch zuerst?

[caco3]

Für den Anfang mal englisch.

Hier mal ein Beispiel mit dem PoC: caco3#46
Du kannst einfach Diskussionen erstellen.
Wenn eines der Schlüsselworte (trigger words) gefunden werden, wird darauf reagiert.

Die Schlüsselwort-Liste ist momentan noch direkt in der Python-Datei, wird aber später noch eine separate Datei.

[caco3]

So, ich hab das ganze mal noch etwas erweitert.
Die Wort liste ist jetzt eine separate Datei: https://github.com/caco3/AI-on-the-edge-device/blob/main/discussion-bot/response_data.txt
Und es gibt ein Script test/test.py, welches basierend auf den bisherigen Diskussionen (799 Stück) eine Tabelle generiert, anhand derer man sieht, wie die response aussehen würde:

Am einfachsten ist es, wenn man es lokal ausführt, siehe im readme.
Oder man es ins GIT committet wird der Test automatisch ausgeführt und die HTML-Seite erstellt: https://github.com/caco3/AI-on-the-edge-device/actions/workflows/test-discussion-bot.yaml

@friedpa Spiel doch bitte mal damit rum.

@SybexX, was denkst Du dazu?