NLP-in-Tensorflow

Building natural language processing systems using TensorFlow starting from processing text to generating poetry with neural networks (LSTMs)

Skills

Natural Language Processing & Generating, Tokenizing, Generating n_grams, Padding, Embeddings, Bidirectional LSTM, Hyper-Parameters, Tensorflow, Keras, Python

These notebooks are submitted as part of assignments while completing a course in Coursera

• DeepLeaning.ai

Included notebooks are

1. Tokenization_Padding_in_Tensorflow.ipynb

   • Embarked on a journey into the world of Natural Language Processing (NLP) by learning how to tokenize words, transforming them into numbers that a machine can understand
   • This foundational knowledge paved the way for understanding how machines can begin to comprehend human language
   • Dataset - kaggle/bbc

2. BBC_News_Articles_Neural_Network_Classifier.ipynb

   • Explored Embeddings
   • These powerful tools map our vocabulary into higher-dimensional space, allowing the machine to grasp the subtleties of word meanings
   • Learned how words with similar sentiments are clustered together, and how the direction of these vectors can reveal the underlying emotions in text
   • The introduction of subword tokenization further highlighted the importance of not just the words themselves, but also the sequence in which they appear
   • Dataset - kaggle/bbc classifying articles into 5 categories ['tech', 'business', 'sport', 'sport', 'entertainment']

3. Text_Classification_Keras_LSTMs.ipynb

   • Explored various model formats that help capture context (forward & backward), allowing for a more nuanced understanding of sentiment in text
   • Learnt best practices for defining Hyper parameters
   • Adjusting hyper parameters & analyzing accuracy of neural network models with different Keras Layers like Conv1D, Dropout, GlobalMaxPooling1D, MaxPooling1D, LSTM and Bidirectional(LSTM)
   • Started experimenting with text prediction, laying the groundwork for creating entirely new sequences of words resulting in Poetry
   • Dataset - kaggle/bbc classifying articles into 5 categories ['tech', 'business', 'sport', 'sport', 'entertainment']

4. Creating_Poetry_With_Bidirectional_LSTMs.ipynb

   • Developed a creative tool by building a poetry generator
   • Drawing inspiration from traditional Irish songs and Shakespearean poetry, trained our model to generate beautiful verses
   • Tokenizing, Generating n_grams, Padding, defined model with Embedding layer & Bidirectional LSTM, Generated verses with 100 words by inputting seed texts
   • Dataset - shakespeare.org