[ad_1]
Having round 16+ years {of professional} expertise and was working within the Expertise Companies division of my firm which gives software program and {hardware} technical help for our clients. Learn additional to study my journey with Nice Studying’s PGP Synthetic Intelligence and Machine Studying Course.
It was found that the purchasers have been beneficiant in offering greater satisfaction scores (starting from 1-5). Nonetheless, there have been eventualities the place suggestions within the free textual content discipline was not as constructive. This was a possible space that may very well be addressed by the group to enhance buyer satisfaction and thereby stop sudden income erosion.
We used ‘Sentiment Evaluation’ to get extra insights into buyer suggestions aside from the score offered. For coaching knowledge, we needed to create a coaching set manually primarily based on the suggestions that was traditionally offered by the shopper and categorized as “Constructive or Impartial” or “Unfavorable.” We used tokenizer & glove embeddings for knowledge pre-processing and skilled a Bidirectional LSTM mannequin utilizing Adam optimizer. We chosen this mannequin as we discovered that this carried out comparatively higher with lesser complexity. The mannequin was used to foretell if the suggestions was “Constructive or Impartial” or “Unfavorable.”
The problem of constructing this mannequin was primarily to manually create giant coaching knowledge set primarily based on the historic knowledge out there. Additionally, we needed to do knowledge clean-up earlier than pre-processing to take away particular characters and typos. We additionally needed to fine-tune the coaching set, and its quantity as preliminary outcomes on the validation knowledge set was not very passable.
Based mostly on the suggestions class prediction by the mannequin, extra engagements have been initiated with the highest clients. We expect this initiative to enhance our share of pockets with these clients.
[ad_2]
