I am trying to make aspect based sentiment analysis model for a bank. Please help how to proceed with it. any help is appreciated.
Here are some of the resources that might interest you.
ABSA has become recently popular, even SemEval (Google it!) has some competitions on it, and teams score fairly high accuracies.
Coming to the point, First you need to extract Aspects, for which you can use CNN, CRFs, etc. (you can find research papers on it), gives fairly high accuracies ~80%.
After extracting aspects, you now have to get sentiments of these aspects, for which you can use RNNs, LSTMs, GRUs, etc. (sentiment depends on words nearby, thats why RNN family of neural networks are needed.) Last time i ran code, I used Google Embeddings for words, plus their POS tags(Stanford POS Tagger)…
Also, there was a research paper focusing on dependency trees for extracting dependency between words for better sentiment analysis (It didn’t improve accuracy much because people tend to use SLANG language many a times.)…
I hope it have helped!