Summary of an article published on The Web Scraping Club on 2025-12-07 (free to read). Read the full article

When you scrape text from several sources, you often need to know what is inside it: which company, person, product, price or date a snippet refers to. Named Entity Recognition (NER) turns that unstructured text into labeled, structured data. It matters for consolidating products across stores, tracking news about a CEO or a company, and as a first step before pattern detection or sentiment analysis on what you collected.

I cover what entity extraction is and compare the main options, from NLTK and Hugging Face transformers to spaCy and the cloud APIs from Google, Amazon and Azure. I then go through use cases in finance, e-commerce, HR and brand monitoring. In the tutorial we ran spaCy’s pre-trained English model on two scraped datasets: headphone descriptions from different e-commerce sites and a set of news summaries. We checked how well the labels matched what we expected in each case, and what to do when a general model does not fit your domain.

What’s inside

  • The options for entity extraction are ordered from research use to production use: NLTK, Hugging Face transformers, spaCy and managed cloud APIs.
  • Four industry examples show which entity types matter in each: finance, e-commerce reviews, job postings, and social media brand monitoring.
  • The tutorial uses the same spaCy script structure on two CSV files, one with product descriptions and one with news text.
  • If a pre-trained model does not fit your domain, there are two ways out: a different pre-trained model such as BERT, or fine-tuning an LLM.

In the full article: The two spaCy scripts, the setup steps, the entity labels printed for each e-commerce and news record, and my take on which fix to choose when the general model falls short.

Read the full article on The Web Scraping Club