Chapter 35 Further Reading: Natural Language Processing for Business Text

Official Documentation

spaCy https://spacy.io/usage The spaCy documentation is among the best-written technical docs in Python. Start with "Linguistic Features" for a thorough explanation of tokenization, NER, and dependency parsing. The "Usage Guides" section has practical how-to articles for most production use cases. If you are going to use NER seriously in a business context, read the "Training" section to understand when and how to fine-tune models on your domain-specific text.

NLTK Book (Free Online) https://www.nltk.org/book/ "Natural Language Processing with Python" by Bird, Klein, and Loper — available free online — is the classic introduction to NLP using NLTK. Chapters 1-3 cover tokenization, frequency analysis, and basic text processing. Chapter 6 covers text classification. The examples use somewhat dated Python syntax (pre-f-strings) but the concepts are timeless and explained with unusual clarity.

TextBlob Documentation https://textblob.readthedocs.io/ Concise and practical. The Quickstart and Tutorial sections cover everything you need for business use. The "Advanced Usage" section explains how to use custom sentiment analyzers if you need to move beyond the default lexicon.

scikit-learn Text Data Tutorial https://scikit-learn.org/stable/tutorial/text_analytics/working_with_text_data.html The official sklearn tutorial for text classification using TF-IDF and Naive Bayes. Working through this tutorial will solidify your understanding of the ML classification pipeline from Section 35.6.

Books

"Text Analytics with Python" by Dipanjan Sarkar (Apress, 2nd edition) The most practical book on Python NLP for data professionals who are not researchers. Covers preprocessing, sentiment analysis, topic modeling, text classification, and text summarization. Code examples are production-quality and the business framing is clear. A good next step after this chapter.

"Applied Text Analysis with Python" by Bengfort, Bilbro, and Ojeda (O'Reilly) Oriented toward applied scientists and analysts. Strong on building NLP pipelines and on moving from prototype to production. The chapter on corpus readers and the pipeline chapters are particularly relevant for business applications.

"Natural Language Processing in Action" by Hobson Lane et al. (Manning) More technical depth than the other books on this list, but excellent at explaining how the algorithms work rather than just how to call the APIs. Chapters on word vectors and topic modeling are particularly illuminating. Read after you have the basics from this chapter and Sarkar's book.

Online Courses and Tutorials

Kaggle NLP Tutorials https://www.kaggle.com/learn/natural-language-processing Kaggle's free NLP course covers text classification and word embeddings with hands-on notebooks. The exercises use real datasets which makes the learning stickier than synthetic examples.

Real Python: NLP with Python and NLTK https://realpython.com/nltk-nlp-python/ A thorough, well-structured tutorial series covering NLTK fundamentals. Good for solidifying the preprocessing concepts from this chapter.

spaCy 101 (Interactive Course) https://course.spacy.io/ The official free spaCy course. Three hours to complete. Highly recommended if you plan to use spaCy for any production NLP work. The interactive exercises catch mistakes immediately in a way reading documentation does not.

Academic Foundations (Optional)

If you want to understand why the algorithms work rather than just how to use them:

"Foundations of Statistical Natural Language Processing" by Manning and Schütze The authoritative reference on the statistical foundations of NLP. Not a business book — this is a graduate-level textbook. But if you want to understand why TF-IDF works, how Naive Bayes text classification is derived, and the statistical basis for LDA, this is where to look. Read selectively; you do not need Chapter 15 on stochastic grammars for business text analysis.

Original LDA Paper: "Latent Dirichlet Allocation" by Blei, Ng, and Jordan (2003) Available free: https://www.jmlr.org/papers/volume3/blei03a/blei03a.pdf The original paper introducing LDA. Sections 1 and 2 (introduction and intuition) are readable by non-specialists and will give you a genuine understanding of what the algorithm is doing when you call models.LdaModel().

Specific Topics Mentioned in This Chapter

TextBlob sentiment accuracy benchmarks For a more nuanced look at TextBlob's accuracy compared to other methods, search for "TextBlob sentiment accuracy comparison" on Google Scholar or towards data science blog posts. Accuracy varies substantially by domain — social media vs. product reviews vs. formal business text.

Gensim LDA documentation and tutorials https://radimrehurek.com/gensim/auto_examples/tutorials/run_lda.html The official Gensim LDA tutorial. More detailed than what we covered in this chapter, including how to use coherence scores to systematically evaluate topic quality.

spaCy Named Entity Recognition (NER) accuracy spaCy publishes benchmark results for its models: https://spacy.io/models/en The en_core_web_sm model (small, fast) achieves ~85% F1 on standard NER benchmarks. The en_core_web_lg model (large) achieves ~87%. For business documents — especially contracts, emails, and news — accuracy in practice often exceeds these benchmarks because the text is more formal and unambiguous than the test corpora.

Datasets for Practice

Amazon Product Reviews Available on Kaggle and Hugging Face datasets. Millions of real product reviews with star ratings. Ideal for practicing sentiment analysis validation: score the reviews with TextBlob and compare against the star ratings.

Enron Email Dataset The famous Enron email dataset contains ~500,000 emails from Enron employees. Useful for NER extraction, keyword analysis, and topic modeling on real business email text. Available from various academic mirrors.

IMDB Movie Reviews A classic benchmark dataset: 50,000 labeled movie reviews (positive/negative). Useful for practicing and evaluating sentiment classifiers. The labels let you measure actual accuracy.

What This Chapter Did Not Cover

This chapter focused on classical NLP techniques. Two important areas are beyond its scope:

Transformer Models and Large Language Models: Modern NLP is increasingly dominated by transformer-based models (BERT, RoBERTa) and large language models (GPT-4). These achieve significantly higher accuracy on most NLP tasks but require more computational resources and are more complex to fine-tune. They are covered in the Advanced Python for Business module.

Multilingual NLP: The tools in this chapter work well for English text. For other languages, spaCy offers models for 20+ languages with varying quality. The general preprocessing and analysis patterns apply, but you need language-specific models and stopword lists.