action_detection_using_neural_network

A trained neural network model to detect whether a given sentence is an actionable item or not using some pre-tagged action item sentences dataset..

Requirement :

• To install dependencies run pip install -r requirements.txt

Dataset consists of labelled sentences :

* True  :   825
* False :  695

NaiveBayesClassifier is trained :

• Download the trained model here
  • On 10% test - 92.7% validation accuracy
  • On 20% test - 88% validation accuracy ) and paste it in "models" folder
  • To train the model run : python3 main_NBC.py /path/to/data/file
  • To predict individual sentences run : python3 main_NBC.py "example sentence to classify"

Convolutional Neural Network is trained :

• Download the trained model and the tokenizer here and paste them in "models" folder
• To train the model run : python3 main_CNN.py /path/to/data/file
• To predict individual sentences run : python3 main_CNN.py "example sentence to classify" /path /saved_model/ /path/saved_tokenizer
  • The current accuracy 95% on 20% validation set.
  • The table for precision, f1 score and recall.

Content	precision	recall	f1-score	support
0	0.98	0.90	0.94	69
1	0.92	0.99	0.95	82
accuracy			0.95	151
macro avg	0.95	0.94	0.95	151
weighted avg	0.95	0.95	0.95	151

LSTM Recurrent Neural Network is trained :

• Download the trained model and the tokenizer here and paste them in "models" folder.
• To train the model run : python3 main_LSTM.py /path/to/data/file
• To predict individual sentences run : python3 main_LSTM.py "example sentence to classify" /paths /saved_model/ /path/saved_tokenizer
  • The current accuracy 92% on 20% validation set.
  • The table for precision, f1 score and recall.

Content	precision	recall	f1-score	support
0	0.92	0.91	0.92	144
1	0.92	0.93	0.93	304
accuracy			0.92	304
macro avg	0.92	0.92	0.92	304
weighted avg	0.92	0.92	0.92	304

Need for improvement

• Including pretrained glove vector embeddings to boost accuracy.
• Increase the dataset, for deep learning large amount of data is required to boost accuracy.
• Include K-fold cross validation in LSTM training module.