WSD

WSD - Word-sense disambiguation is an open problem in Natural Language Processing concerned with identifying which sense of a word is used in a sentence. The major purpose of WSD technique is to accurately understand the meaning of a specific word in a text. We have proposed a method to find ambiguous words in a sentence employing a supervised machine learning approach. If the word has different meaning in both the sentences then it is domain dependent otherwise domain independent.

Step1: Import all the required libraries

Step2: Generate word embeddings, glove as well as ROBERTA

	- Lemmatize sentence1 and sentence2  
	- Drop all the irrelevant columns except Ambigous_Word, Sentence1, Sentence2. Store it as X  
	- Convert the labels column into 1 or 0 and store it as y.  
	- Perform train_test_split()  
	- Store the data in the pickle file.  
	- Store the whole data (X, y) in the pickle file.  

Load the data from the pickle file

Pickle file format:
	data = {  
        # Training Data  
        "training_sentences1": training_sentences1,  
        "training_embeddings1": self.training_embeddings1,  
        "training_sentences2": training_sentences2,  
        "training_embeddings2": self.training_embeddings2,  
        "train_ambiguous_word": training_ambiguous_words,  
        "train_ambiguous_word_embeddings": self.training_ambiguous_word_embeddings,  
        "training_labels": self.y_train,  
        # Testing Data    
        "testing_sentences1": testing_sentences1,  
        "testing_embeddings1": self.testing_embeddings1,  
        "testing_sentences2": testing_sentences2,  
        "testing_embeddings2": self.testing_embeddings2,  
        "test_ambiguous_word": testing_ambiguous_words,  
        "test_ambiguous_word_embeddings": self.testing_ambiguous_word_embeddings,  
        "testing_labels": self.y_test,  
        # Complete Data   
        "sentences1": sentences1,  
        "sentences_embeddings1": self.embeddings1,  
        "sentences2": sentences2,  
        "sentences_embeddings2": self.embeddings2,  
        "ambiguous_word": ambiguous_word,  
        "labels": labels  
    }  

Step3: Train a machine learning model (Random Forest, Extra Trees Classifier and Support Vector Classifier), with the following format:

	  input_data=<ambigous_word_embeddings, sentence1_embeddings, sentence2_embeddings>  
	  input_data=<GloVe Vector Embeddings,  ROBERTA,              ROBERTA>  

Step4: Evaluate the model using testing data and development data. Evaluation metrics = accuracy, precision, recall, f1 score

Step5: Prediction phase:

	  - Input two sentences from the user.  
	  - Convert the sentences to lowercase. Remove stopwords  
          - Lemmatize the words  
          - Get glove embedding for noun and the ROBERTA embeddings for sentences and predict using the classifier whether T or F.