Some popular Instagram accounts receive hundreds of comments below each post. Such account owners need to know if the post has been positively accepted by the auditory, while it is nearly impossible to read all the comments one by one.
We used the LSTM neural network to determine the tonality of each comment. To train the network, we used two datasets: a dataset of Twitter posts and a dataset of IMDb reviews.
We also added several heuristics, like, for example, emoji recognition.
As a result, we have made an application that provides its users with information on the average tonality of the comments on their Instagram posts, reveal the most positive comments (to encourage the user) and the most negative comments (to let the user react to the problems).