TODOs - Githubissues

mschroederi / DL-sentiment-analysis

Movie Sentiment Analysis using PyTorch with Transformer, LSTMs and BoW models.

MIT License

0 stars 0 forks source link

TODOs #5

Closed mschroederi closed 4 years ago

mschroederi commented 4 years ago

Embeddings

Character-based vs word-based
- CountVectorizer
- TfIdfVectorizer
- HashVectorizer
Word2Vec
(BERT)

Models

Simple Linear
RNN (LSTM vs GRU)
(Attention?)

Other

NLTK

mschroederi commented 4 years ago

Highlights

LSTM model
documentation setup & first sections
Accuracies on test data:
- BagOfWords: ~95%
- LSTM: ~89%
Word2Vec

Plans for next week

[ ] Baseline approach (Python counting) [David]
[ ] BagOfWords (selectable vectorizer)
- CountVectorizer
- TfIdfVectorizer
- HashVectorizer
[ ] LSTM [Mats]
- uni-directional (think of hyper-parameter optimization)
- bidirectional & attention
[ ] (Transformer) [Jan]

Only for the best approach

[ ] classification of own review (write_review.py -> maybe select model)
[ ] which words are most important for class 0 and 1
[ ] investigate false positive and false negative reviews