/courses/machine-learning/nlp-deep-learning/Recent content in NLP with Deep Learning on DATATWEETSHugoenCopyright (c) 2025 DatatweetsFri, 14 Nov 2025 09:00:00 +0200/courses/machine-learning/nlp-deep-learning/lesson-1-introduction-to-natural-language-processing/Fri, 14 Nov 2025 09:00:00 +0200/courses/machine-learning/nlp-deep-learning/lesson-1-introduction-to-natural-language-processing/Discover what NLP is and why raw text is so difficult for machine learning models. You will meet the real Disaster Tweets dataset, explore its class balance and tweet lengths, and walk through the classic text preprocessing pipeline of tokenization, lowercasing, and stopword removal with Keras and TensorFlow./courses/machine-learning/nlp-deep-learning/lesson-2-text-vectorization-and-word-embeddings/Fri, 14 Nov 2025 09:00:00 +0200/courses/machine-learning/nlp-deep-learning/lesson-2-text-vectorization-and-word-embeddings/Learn how text becomes numbers for deep learning. You will use the Keras TextVectorization layer to build a vocabulary and produce padded integer sequences, then see why dense word embeddings beat one-hot encoding and watch an Embedding layer learn to place disaster words apart from everyday words./courses/machine-learning/nlp-deep-learning/lesson-3-building-text-classification-models/Fri, 14 Nov 2025 09:00:00 +0200/courses/machine-learning/nlp-deep-learning/lesson-3-building-text-classification-models/Turn raw tweets into a working classifier. You will assemble a TextVectorization -> Embedding -> GlobalAveragePooling1D -> Dense pipeline in Keras, train it on real disaster tweets, evaluate it with accuracy and AUC, and understand why averaging word vectors throws away word order./courses/machine-learning/nlp-deep-learning/lesson-4-building-sequence-models-for-text/Fri, 14 Nov 2025 09:00:00 +0200/courses/machine-learning/nlp-deep-learning/lesson-4-building-sequence-models-for-text/Discover why bag-of-embeddings models throw away word order, and how recurrent layers read text as a sequence. You will build an Embedding to Bidirectional(LSTM) to Dense classifier on the real Disaster Tweets dataset and watch test accuracy climb from 0.710 to 0.751./courses/machine-learning/nlp-deep-learning/lesson-5-building-text-models-with-transformers/Fri, 14 Nov 2025 09:00:00 +0200/courses/machine-learning/nlp-deep-learning/lesson-5-building-text-models-with-transformers/Understand the transformer idea: self-attention lets every token attend to every other token directly, with no recurrence. You will learn queries, keys, and values, multi-head attention, and positional information, then build a compact transformer block in Keras and evaluate it on the real disaster tweets dataset./courses/machine-learning/nlp-deep-learning/lesson-6-guided-project-classifying-disaster-tweets/Fri, 14 Nov 2025 09:00:00 +0200/courses/machine-learning/nlp-deep-learning/lesson-6-guided-project-classifying-disaster-tweets/Assemble everything from this module into one project. You will load the real disaster tweets dataset, clean and vectorize the text, train and compare three model families with Keras, then evaluate the best model with a confusion matrix and a careful error analysis.