NLP Pipeline: Data Acquisition, Augmentation and Labelling | Module 3 | NLP from Scratch series |
Notes: https://miro.com/app/board/uXjVIXaFM2...
--------------------------------------------------
GDrive link for notebooks given in the END.
---------------------------------------------------------------
Welcome to the second lecture of the "Natural Language Processing Learned From Scratch" series! In this comprehensive video, we recall the topics taught in the last lecture such as HMM, SVM, NBC, RNN, LSTM, CNN, Transformers and Autoencoders.
Follow instructor on Linkedin to get more updates on Technology:
Sharvesh Subhash: / sharveshsubhash
Below is an engaging, detailed breakdown with timestamps so you can jump to the topics that interest you the most:
0:00 – Introduction & Recap
A brief recap of the previous lecture covering the three approaches (CFG & rules, statistical methods, and deep learning with LSTMs) that set the foundation for our discussion today.
0:01:00 – Overview of the NLP Pipeline
Introduction to the complete NLP pipeline—from data acquisition to cleaning, pre-processing, feature engineering, model evaluation, and deployment. Understand the step-by-step process that transforms raw text into actionable insights.
0:03:00 – Data Acquisition Techniques
Detailed explanation of data acquisition methods:
Using public datasets
Web scraping (extracting text from HTML pages)
Product-based data collection (user feedback, reviews, surveys)
Data augmentation to artificially increase data volume using paraphrasing and synonym replacement.
0:08:00 – Web Scraping & HTML Parsing
Learn about techniques for scraping textual data from web pages using tools like Beautiful Soup. Explore how to extract only the necessary text (ignoring HTML tags) and perform cleaning to obtain high-quality input data.
0:12:00 – Text Normalization & Spelling Correction
Dive into text normalization strategies, including handling Unicode characters (emojis) and implementing spelling corrections through API-based tools and digitized dictionaries.
0:17:00 – Advanced Data Augmentation Methods
An in-depth look at techniques for creating augmented datasets:
Synonym replacement
Back translation (translating to another language and back)
Random insertion and deletion
Bigram flipping and entity replacement
These methods help increase data diversity and improve model robustness.
0:26:00 – Implementing Augmentation Techniques
Hands-on discussion of code snippets for data augmentation using libraries such as NLPog. Understand how object-oriented programming concepts (constructors and classes) are used to implement various augmentation methods.
0:35:00 – Text Extraction from Images & PDFs
Explore practical implementations:
Using OCR (pi-desert) for extracting text from images
Parsing PDFs with Tika to extract content accurately, along with discussions of common challenges (e.g., formatting issues).
0:44:00 – Web Scraping Implementation with Beautiful Soup
A step-by-step walkthrough of how to use Beautiful Soup for extracting and cleaning data from HTML content. Learn about handling server restrictions by mimicking web browser requests.
0:51:00 – Implementing PDF Text Extraction
Learn how to leverage Tika for extracting text from PDF documents, with code examples demonstrating error handling and text cleanup.
0:58:00 – Data Augmentation & Labeling in Practice
Detailed exploration of industry-standard data augmentation using NLPog for synonym and context-based word replacements.
Additionally, an explanation of automated data labeling techniques using weak supervision approaches (e.g, Snorkel) to generate training labels with minimal human intervention.
1:00:00 – Auto Labeling with Snorkel & AutoTrain
Learn about the concept of weak supervision and how tools like Snorkel can be applied for auto-labeling tasks. Discover how industry giants build proprietary solutions on top of these open-source tools and a brief look at using AutoTrain for low-code AI model deployment.
1:25:00 – Wrapping Up Part One
A summary of all the discussed topics from data acquisition to advanced labeling techniques. Final motivational advice to stay disciplined and consistent while mastering NLP and LLMs, with a sneak peek into the next part where we cover feature engineering and model training in detail.
This video is packed with valuable insights, detailed implementations, and practical coding demonstrations on Google Colab. Whether you're a beginner or looking to sharpen your NLP skills, you'll find this lecture both engaging and informative.
Google drive link for Notebooks:
Text extraction from webpage using BeautifulSoup, image using pytesseract, tika, NLPAug, Snorkel (Stanford AI Lab)
https://drive.google.com/drive/folder...
#naturallanguageprocessing #learnfromscratch #nlp #NLPAUG #webscraping #beautifulsoup #tika #machinelearning #DataAugmentation #data #dataanalytics #datascience #dataanalysis #education #coding #python #practical #handsonlearning