Q$2.$ Neural Network Approach for WSD: Code $[10]$

In this question, you will build WSD using a Bi$-$directional LSTM$.$ You may use PyTorch for this task.

$1.$ Load SemCor corpus using NLTK$1$ with semcor.sents$().$ Similarly, load WordNet model in NLTK$2$ as import wordnet as wn$.$ Randomly select $50$ sentences as the test set for this question and remaining sentences will be the train set. Build a vocabulary of all the words in the SemCor dataset and assign unique index to each word.

$2.$ Extract WordNet tags as the labels for evaluation. Note that for stop words, the data does not have the tags.

$3.$ Build a neural network model with an Embedding layer of dimension $100,$ Bi$-$directional LSTM layers, a Dense layer, and an output layer with the SoftMax activation function.

$4.$ Train the model assuming your own hyper$-$parameters such as epochs, optimizer, learning rate, etc.

$5.$ Evaluate the model using test set with metrics precision, recall, and F$-$score. Report the hyper$-$parameters used for model training.

$6.$ With examples from test set, compare and analyse the results of the $3$ modelsl. Your analysis will include the performance of each model, if the model is able to disambiguate the senses$,$ if it fails then why, etc.