Currently I am working on a project where I have ASR on which I am performing semantic analysis to extract meaning out of it. The ASR text contains huge amount of vague conversational text which needs to be eliminated using some algorithm. Trial 1: I tried to use NER to find some important entities in text and then I eliminated the sentences which did not have any important entities. It worked for some samples but failed for majority because it couldn't retain meaningful sentences. Trial 2: Tried summarizing it using Seq2seq model supervised learning, but it was an entire failure. It couldn't even generate proper sentences. Trial 3: Tried to perform Sementic Analysis of the text, couldn't find a way to perform it successfully.
I need suggestions for something to try and also any help or examples for performing Sementic Analysis for extracting meaning out of it.
Hope anyone would have performed similar research.