How ahmedsamy02/nlp-project works
Project type and purpose
• This is an NLP research/prototype project (“Pizza Order Parsing System”) implemented as a series of Jupyter notebooks and scripts.
• Purpose: take free‐form pizza‐order text, preprocess it, run two sequential neural models (order classification + entity recognition), then output a structured JSON parse tree.
File structure
• Top‐level notebooks correspond to pipeline stages:
• preprocessing.ipynb – data cleaning, negation handling, abbreviation expansion
• classification_relations.ipynb – order‐boundary classification model
• enc-dec.ipynb & encoder-decoder.ipynb – experimentation with encoder–decoder architectures
• “final codes/” contains the demo pipeline and final‐model notebook.
• “sample_input_output/” holds example input.txt and output.json.
• No traditional “src/” directory; each notebook encapsulates a module.
README for insights
• Pipeline stages: Data Preprocessing → Order Classification Model → Entity Recognition Model → Parse‐tree generation.
• Key technologies: Python, Jupyter, FastText embeddings, Bidirectional LSTM (Keras/TensorFlow), regex labeling, RMSprop/Adam optimizers.
• Dataset: Amazon Pizza Order dataset (2.4M samples), with 500K‐sample training subsets for each model.
Detail without over-fragmentation
• Distinct pipeline stage as its own component.
• Both training and inference flows if needed.
• Do not split beyond preprocessing, classification, entity recognition, and parse generation modules.