모델을 불러와 문서 처리하기

Before you load anything, you download a model once. The en_core_web_sm model is small, fast, and perfect for getting started.

You bring spaCy to life with spacy.load. It reads the model from disk and hands you an nlp object ready to process text.

That nlp object is your whole pipeline wrapped up. Calling it on text runs every stage: tokenizer, tagger, parser, and entity recognizer.

To analyze a sentence, just call nlp() on a string. The result is a Doc, a rich container holding all the analysis.

That single call already ran tokenizing, tagging, and parsing. The Doc now carries every result, so you never repeat the work.

A Doc behaves like a sequence, so you can loop through it. Each item you get back is a Token with its own attributes.

The raw word lives in token.text. It is exactly the surface string spaCy found while splitting your sentence.

Loading a model is slow, so do it once at startup. Then reuse the same nlp object for every document you process.

For lots of texts, use nlp.pipe. It batches them efficiently and is far faster than calling nlp() in a plain loop.

Need only tokens? You can disable unused components when loading to skip work and run even faster.

You can also start from spacy.blank for a tokenizer-only pipeline. It is handy when you want to build everything yourself.

What do you get back from calling nlp() on a string?

You install a model, load it once with spacy.load, then call nlp() to get a Doc. One call runs the whole pipeline and you reuse it everywhere. 🎯