How to classify a post in a social network?

Question

Machine learning

Good day
, I have the task of identifying negative posts on social networks
. Let's say I take and train my model using ready-made datasets for classifying the text of a post . I
also train another model for classifying an image there
. Now I don’t understand how to connect it.
There is info about the name of the group in the post. where it was posted, the date, the number of likes and reposts
And I want to use this data somehow in the analysis,
but there is no ready-made dataset for this, and marking everything with pens is very boring and long
How would you do it in this case

Answer 1

Your question is somehow very chaotic.
What did you get out of it.
1. You know how to identify the tone of the text. Let's assume - "negative-positive". Most likely, the result can be normalized to the range (-1; +1)
2. You know how to classify images " there " (by the way - where is "there"?) Well, they were classified by class. Maybe dichotomous, maybe multiple. Received some assessment of belonging to the class, which they themselves previously identified. In any case, the resulting estimate can be normalized if necessary.
3. What are you going to associate with what? If we assume that both the first and second tasks scattered your posts into classes, then you can now move on to the classic multidimensional (and very likely - just two-dimensional) classification problem, which can be solved by any known method. True, you will have to try different methods, since there is no universal one. But there is hope that anything in the range from kNN to random forest will work for you.
4. If there is additional information - well, in the worst case, this will increase the number of features that the classifier works with.
5. How is there no ready-made dataset? You wrote " I take and train my model using ready-made datasets". That is, there is a dataset of posts with texts, obviously - and with pictures, and since you have already trained - it means it is marked up. And you yourself write "to the post there is info about the name of the group where it is posted, the date, the number of likes and reposts" . Well, do not discard this information, but pass it along with the results of sentiment analysis and image classification to the final, generalizing classifier. What's the problem?