Natural Language Processing

The first file uploaded was for predicting tags on StackOverflow using Linear Models
The second file uploaded was NER(Named Entity Recognition) on twitter using BiDirectional LSTM. It is not the best, achieving around 90% precision on train data set and 50% on test data set, with similar on validation. However implementing a hybrid model of LSTM, CNN and CRF(Conditional Random Field) may prove beneficial and I will try it soon (following the paper https://arxiv.org/abs/1603.01354)
The third file uploaded was to find duplicate questions on StackOverflow. It utilised the StarSpace embeddings (by Facebook) for this task.