Currently we are using a simple CountVectorizer for extracting features from text. It might be a good idea to experiment with using word embeddings instead. The rationaly behind this is that count vectorization doesn't factor in the context of the text where there might be semantic context important for the triaging model.
Currently we are using a simple
CountVectorizer
for extracting features from text. It might be a good idea to experiment with using word embeddings instead. The rationaly behind this is that count vectorization doesn't factor in the context of the text where there might be semantic context important for the triaging model.