NLP & LLMs in Finance

Read these first

1. 01BERT and Masked Language Models

   A model trained by blanking out words and making it guess them, using the words on both sides. That pre-training produces context-aware representations you can fine-tune on a few thousand labelled filings.

2. 02Building a News Sentiment Pipeline End to End

   A sentiment score is only as good as its weakest stage, sourcing, entity matching, scoring and timestamping each fail differently, and any one of them can quietly wreck the signal.

3. 03Coreference and Alias Resolution for Issuers

   A company appears in text under a dozen different names, and pronouns stand in for it a dozen more times, resolving all of that to one stable identifier is what makes text usable as structured data.

4. 04Large Language Models

   Models trained to do one thing, predict the next token, at a scale where that single objective produces summarisation, extraction and reasoning as side effects. Useful in research pipelines, dangerous in backtests.

5. 05Retrieval-Augmented Generation

   Instead of trusting a language model's memory, you search your own documents first and paste the relevant passages into the prompt. It turns an unverifiable fluency problem into a search problem you can measure.

6. 06Word Embeddings and Word2Vec

   A way of turning words into coordinates so that words used in similar contexts land near each other. Word2Vec learns those coordinates from raw text alone, and the geometry it produces turns out to encode meaning.

Then the rest

1. 07Bag of Words and TF-IDF
2. 08Text Preprocessing and Tokenization
3. 09Aggregating and Decaying Sentiment Scores
4. 10Central Bank Communication Analysis
5. 11Document Embeddings as Alpha Features
6. 12Dynamic Topic Models for Market Narratives
7. 13Earnings Call Transcript Analysis
8. 14Embedding-Based Topic Modelling
9. 15Event Taxonomies for Financial Text
10. 16Hallucination and Groundedness
11. 17LLM Evaluation and Benchmarking
12. 18Mining SEC Filings with NLP
13. 19News Deduplication and Story Clustering
14. 20News Event Extraction
15. 21Novelty and Staleness Detection in News
16. 22Prompt Engineering Patterns
17. 23Scoring News Relevance to a Ticker
18. 24Sentence Transformers and Contrastive Fine-Tuning
19. 25Text Classification for Filings
20. 26The Loughran-McDonald Financial Lexicon
21. 27Timestamp Hygiene for Text-Derived Signals
22. 28Topic Modeling with LDA
23. 29Vector Databases and Embedding Search
24. 30Weak Supervision for Financial Text Labels
25. 31LLM Agents and Tool Use
26. 32LLM Data Contamination in Backtests
27. 33Parameter-Efficient Fine-Tuning and LoRA
28. 34RLHF and Preference Alignment