A semantic analyzer of sentences and questions based on Language processing algorithms (with Pytorch :) ).
In this problem, we compare similarity between two questions. The dataset used here comes from the Quora Question Pairs challenge.
Our solution with a pretrained BERT model built with Pytorch.
How does it work?
python3 same_analyze.py "Am I wrong?" "Do you love ice scream?"
Same at 0.15%python3 same_analyze.py "How do I save videos from twitter?" "How do you upload videos from your camera roll onto Twitter?"
Same at 10.42%python3 same_analyze.py "How do I save videos from twitter?" "How do you upload videos from your camera roll onto Twitter?"
Same at 97.04%Details are in the notebook qqp_BERT.ipynb
Here, we evaluate how positive is a comment sent. The dataset used here to train the model come from The Stanford Sentiment Treebank dataset
Our solution is a BiLSTM model trained on a negative-positive classification task. We embed words with the Word2Vec Gensim model trained with the glove-wiki-gigaword-50 corpus.
How does it work?
python3 sent_analyze.py "I love this movie"
Positive at 100.0%python3 sent_analyze.py "A great idea becomes a not-great movie."
Positive at 0.07%Details are in the notebook sentiment_analysis_BiLSTM_v2.ipynb