Text Reviews in banking ML Models

Models for determining the sentiment analysis of text reviews for subsequent improvement of the company's service. Moreover, this service can be extrapolated from the banking sector to insurance and any sphere where there is a reflection of users, as well as the existence of a platform for posting reviews and the availability of reviews themselves (according to the principle of the more reviews, the better). The service will allow you to find the causes of negativity and give them both a social and economic assessment. The reported study was funded by RFBR, project number 20-310-70042. Models are applicable only to the Russian-speaking community, since the models were trained on Russian semantics. Model weights (CNN) stored in files: cnn-frozen-embeddings-08-0.94.hdf5 and n-trainable-03-0.94.hdf5 (CNN with frozen embedding layer, and defrosted retrained, respectively) responses_model.w2v, w2v_xgboost.w2v - saved Word2Vec models of reviews. xgb_model.pickle and nn_model.pickle are trained models.