DLNLPProject

Deep Learning Romanian Reviews Dataset Sentiment Analysis

Datasets

1. Movie and Products RO Reviews (ro_sent)

2. Products Ro Reviews (LaSoReDa)

Results

1. Comparison between multiple types of models:
   1. Different training/validation datasets
   2. Different deep learning techniques (maybe some ML ones too (NB))
   3. Different word/text representations (ngram, w2v, d2v, tfidf, *wwv)
   4. Different sanitization/preprocessing methods
2. Plotting and visualisation of the models performance
   1. accuracy, precision, recall
   2. f1
   3. ROC AUC curves

Goals

We propose a further inspection of the tasks presented by the datasets provided above. By combining the information and using advanced deep learning techniques, we will develop a more performant model relative to the ones already made for each of the tasks.

State of the art

State of the art link