Natural Language Processing in Python using spaCy – Part II | IBKR Quant Blog (2024)

Part I in this series provides and overview of spaCy and demonstrates how to install the library.

Split text into sentences using spaCy

# Initialize the text that you want to analyzetext = "Ripple was founded in 2012 by Jed McCaleb and Chris Larsen (co-founder of E-Loan). But the gears were in motion back in 2004, four years before Bitcoin, when a Canadian programmer, Ryan Fugger, developed RipplePay.RipplePay, though not based on blockchain, was a secure payment system for a financial network. In 2012, Jed McCaleb and Chris Larsen started the company OpenCoin, renamed Ripple Labs in 2013."text# Source: https://blog.quantinsti.com/ripple-xrp/

show_text.pyhosted with ❤ byGitHub

Source:Exploring Ripple and XRP: What it is, Features, and More

Let us now pass this text to our model and split it into individual sentences.

doc = nlp(text)# Split the text into sentencessentences = [sentence.text for sentence in doc.sents]# Print the original textprint(f"Original text: \n\n{text}\n{'-'*50}\n")# Print the different sentences from the textprint( f"We have split the above into {len(sentences)} sentences:\n\n 1. {sentences[0]}\n2. {sentences[1]}")

split_the_text.pyhosted with ❤ byGitHub

Removing punctuation using spaCy

Before we proceed further with our analysis, let us remove the punctuations from our text.

token_without_punc = [token for token in doc if not token.is_punct]token_without_punc

remove_punctuation.pyhosted with ❤ byGitHub

Removing stop words using spaCy

We can remove the stop words likeand,the, etc. from the text because these words don’t add much value to it for analysis purposes.

For this, we first get the list of all the stop words for that language. There are two ways in which we can do it.

# Get the stop words: 1st methodall_stopwords = nlp.Defaults.stop_wordslen(all_stopwords) # Output: 326

remove_stop_words.pyhosted with ❤ byGitHub

This is the second method.

# Get the stop words: 2nd methodfrom spacy.lang.en.stop_words import STOP_WORDSlen(STOP_WORDS) # Output: 326

remov_stop_words_2nd_method.pyhosted with ❤ byGitHub

As we can see from the code below, the results of both the above methods are the same.

# Both are the sameall_stopwords is STOP_WORDS

check_methods_equality.pyhosted with ❤ byGitHub

Now that we have got the list of stop words, let’s remove them from our text.

# Remove the stop wordstoken_without_stop = [ token for token in token_without_punc if not token.is_stop]token_without_stop

remove_stop_words_from_text.pyhosted with ❤ byGitHub

POS tagging using spaCy

Let us now see how spaCy tags each token with its respective part-of-speech.

import pandas as pdtoken_pos = []for token in token_without_punc: token_pos.append([token.text, token.lemma_, token.pos_, token.tag_, token.dep_, token.shape_, token.is_alpha, token.is_stop])df = pd.DataFrame(token_pos, columns=['Text', 'Lemma', 'POS', 'Tag', 'Syntactic dependency relation', 'Shape', 'Is Alphabet', 'Is stop word'])df

pos_tagging.pyhosted with ❤ byGitHub

If you would like to know more about the different token attributes, please clickhere.

Named entity recognition using spaCy

What is a named entity? It is any ‘real-world object’ with an assigned name. It may be a country, city, person, building, company, etc. spaCy models can predict the different named entities in a text with remarkable accuracy.

ner = []for ent in doc.ents: ner.append([ent.text, ent.start_char, ent.end_char, ent.label_])df_ner = pd.DataFrame( ner, columns=['Text', 'Start character', 'End character', 'Label'])df_ner

named_entity.pyhosted with ❤ byGitHub

Dependency Visualization using displaCy

Adding another powerful functionality to it’s arsenal, spaCy also comes with a built-in dependency visualizer valleddisplaCythat lets you check your model's predictions in your browser.

To use it from the browser, we use the ‘serve’ method. DisplaCy can auto-detect if you are working on a Jupyter notebook. We can use the ‘render’ method to return markup that can be rendered in a cell.

Let us try the spaCy’s entity visualizer to render the NER results we got above.

For this, let us import displaCy which is ‘a modern and service-independent visualisation library’

from spacy import displacydisplacy.render(doc, style="ent")

dependency_visualisation.pyhosted with ❤ byGitHub

Output:

Getting linguistic annotations using spaCy

You can get insights into the grammatical structure of a text using spaCy’s linguistic annotations functionality. It basically tells you the part of speech each word belongs to and how the different words are related to each other.

We need to have installed spaCy and the trained model that we want to use. In this blog, we’ll work with the English language model, the en_core_web_sm.

Let’s look at an example.

# Pass the text you want to analyze to your modeldoc = nlp("Jennifer is learning quantitative analysis.")# Print the part-of-speech and the syntactic dependency relation for the tokensfor token in doc: print(token.text, token.pos_, token.dep_)

linguistic_annotations.pyhosted with ❤ byGitHub

Let’s visualize the syntactic relationship between the different tokens.

from spacy import displacydisplacy.render(doc, style="dep")

visualise_syntactic_relationship.pyhosted with ❤ byGitHub

Output:

spaCy examples on Github

We’ve covered some of the basic NLP techniques using the spaCy library. To delve into further details, you can head to thespaCy GitHub pagewhere you can find tons of examples to help you in your journey into the spaCy universe.

In addition to the development techniques, you can also try out theextensive test suiteoffered by spaCy which uses thepytestframework.

Conclusion

We’ve explored some of the main features of spaCy, but there is a lot more to be explored in this powerful Python natural language processing library. I hope that this blog has gotten you started with the initial few steps of this journey. So go ahead and explore the enormous world of human language and thoughts!

Want to harness alternate sources of data, to quantify human sentiments expressed in news and tweets using machine learning techniques? Check out this course onTrading Strategies with News and Tweets. You can use sentiment indicators and sentiment scores to create a trading strategy and implement the same in live trading.

Till then, happy tweeting and Pythoning!

Visit QuantInsti website for additional insight on this topic:https://blog.quantinsti.com/spacy-python/.