0
$\begingroup$

I have a software that is designed for students to have chat rooms with their classmates, and I want to process the datas being exchanged in the chat-rooms using AI model. The issue is, I don't know the specific model that is best fit or generally good fit for this type of datas. The chat-room can be considered as a normal discord channel without any toxicity, but I would like to learn the patterns, the trending topics or conversations, the common converstations they are having. What are the recommendations?

This is my idea and please be understanding of my extend of my knowledge. I am trying to figure out ways myself and I thought I would ask the community for directions. The software is for students that autamatically assigns a chat-room per class they are enrolled in, and their classmates are within the same chat-room(given they have same exact class).

I want to process all and any type(if the existing technology allows my greed) of data coming from the chatrooms and display the processed datas and use it as necessary. The AI model I have in mind is a passive one that simply processes the datas and outputs the processed version. You may say this is too vague, but I don't know what directions to look.

Improve this question
$\endgroup$
3
  • $\begingroup$ Can you please put your specific question in the title? "Application of AI" is very generic and not a question. Moreover, I'd recommend you tell us your knowledge of AI or machine learning and what you've found so far, if you have done any research, which I assume you did. $\endgroup$
    – nbro
    Commented Jan 26 at 11:02
  • $\begingroup$ Could you look at it again and see if you can add more? @nbro $\endgroup$
    – Erden E
    Commented Jan 26 at 18:19
  • $\begingroup$ So are you looking for a tool/model that in a way summaries the chat room conversations or gives you the main topic of the conversations? Could you please further change the title to ask this specific question? Anyway, if my assumption is correct, you may be interested in "topic modelling", although you could also try to use the OpenAI models. I think may may be able to do this job, although you will first need to fetch the data from the chat rooms before giving it to the models, which you need to do anyway. $\endgroup$
    – nbro
    Commented Jan 29 at 9:54

1 Answer 1

Reset to default
0
$\begingroup$

This is a bit of guess, given you vague question, but you might be interested in doing the following:

  1. first is to embed the separate discussions using an embedding model (word2vec, BERT, Ada2)
  2. Then cluster these embeddings, you could use K-means clustering, in order to figure out what the main groups of conversations are.
  3. You could also perform NER (Named Entity Recognition), in order to determine the main key words in the debates.
  4. Once these are done, you can analyse the data since the key words found from NER can be used to classify the exchanges (through presence, or through semantic similarity using the embeddings)

Alternatively (less technical):

  1. you could ask an AI to summarize each exchange first.
  2. Then given this list of summaries, ask an AI to give you a list of topics. (you did the summaries before in order to shorten the token limit)
  3. Finally, for each chat, you can ask an AI to classify if it belongs to any one of the previously found topics.

Hopefully this vague answer will spark a lingr series of corrections that will be helpful.

Improve this answer
$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .