Lessons
Natural Language Processing (NLP): all lessons
From bags of words to embeddings and transformers.
6 modules · 19 lessons · 1 free to read now
About this track: outcomes, tools and priceModule 1Representing Text
3 lessonsFrom one-hot vectors to word embeddings to contextual embeddings — how machines represent the meaning of words.
From Words to Vectors
18 minModels cannot do arithmetic on words. How we turn words into numbers decides what a model can learn — and what it never can.
Tokenisation for NLP
16 minA model never sees words — it sees tokens. How text is cut decides the vocabulary, the sequence length, the API bill, and why Indonesian costs more than English.
Included with the track Simulation insideTF-IDF and Classic Text Features
16 minCount the words, weight them by rarity, and you have a text classification baseline that trains in seconds — and is often hard to beat.
Included with the track Simulation inside
Module 2Text Classification
4 lessonsSentiment analysis, topic detection and ticket routing — from TF-IDF + logistic regression to fine-tuned transformers.
Text Preprocessing Pipelines
14 minEvery preprocessing step destroys information. This lesson is about deciding deliberately what to merge, what to keep, and why the answer differs for TF-IDF and for transformers.
Included with the trackClassic Text Classifiers
18 minTF-IDF with logistic regression: trains in a second, its weights are readable, and it is often enough. This lesson builds one from 60 reviews all the way to a routing threshold.
Included with the track Simulation insideFine-Tuning Transformers for Classification
20 minBorrow a model that has already read millions of Indonesian sentences, attach a 2,307-parameter classification head, train briefly — then decide honestly whether the result is worth the cost.
Included with the track Simulation insideClassification with LLMs
16 minOne prompt can classify without a single training label. The question is no longer "can it", but when it is cheaper, when it costs more, and how you know it is right.
Included with the track Simulation inside
Module 3Information Extraction
3 lessonsNamed entity recognition, relation extraction, and turning free documents into structured data.
Named Entity Recognition
16 minClassification gives one label per document. Extraction needs a label per token: who, which organisation, where, when — and the BIO scheme is how you write that down so a machine can learn it.
Included with the track Simulation insideStructured Extraction with LLMs
18 minFrom invoices and contracts to JSON records ready for a database: a schema that guards the types, code that guards the truth, and a loop that decides when a person has to look.
Included with the track Simulation insideEvaluating Extraction
14 min"95% accuracy" says nothing about an extraction system. What matters: per entity or per token, per field or averaged, and which mistakes lose money.
Included with the track Simulation inside
Module 4Transformers for NLP
3 lessonsEncoders, decoders, BERT and GPT — the architecture that changed NLP, and how to use pretrained models.
Encoders and Decoders
16 minBERT understands, GPT generates — and why the difference is in the attention mask.
Included with the track Simulation insidePretrained Language Models
14 minPretraining, fine-tuning, and choosing a model from thousands available.
Included with the trackHugging Face in Practice
18 minPipelines, tokenizers and the Trainer — from zero to a running model.
Included with the track Simulation inside
Module 5Semantic Search and Sentence Embeddings
3 lessonsSentence embeddings, vector databases and evaluating retrieval — the foundation of modern search and RAG.
Sentence Embeddings
14 minFrom word vectors to sentence vectors, and models trained for similarity.
Included with the track Simulation insideVector Databases
15 minFast nearest-neighbour search, indexes and metadata filters.
Included with the track Simulation insideHybrid Search and Evaluation
17 minCombining keywords and meaning, then measuring the result.
Included with the track Simulation inside
Module 6NLP for Indonesian
3 lessonsSlang, code-mixing, regional languages and resources like IndoNLU — closing with a capstone project.
Challenges of Indonesian Text
15 minAbbreviations, slang, affixes, and mixed Indonesian–English text.
Included with the trackIndonesian NLP Resources
13 minDatasets, benchmarks and pretrained models for Indonesian.
Included with the trackCapstone: a Review Analyser
45 minSentiment and topics from e-commerce reviews, evaluated and ready to use.
Included with the track Simulation inside