Natural Language Processing

Unit 4: Required

Materials We Provide

The Yelp dataset was chosen because of its rich and colloquial text attributes, in addition to how well it lends itself to sentiment analysis.

Note: This lesson also uses the Naive Bayes model MultinomialNB, which is often used for NLP applications, such as spam detection. An appendix is included at the end of the lesson for interested students. Supplemental materials are also offered if you want to explore Bayes-related topics.

Learning Objectives

By the end of this lesson, students should be able to:

• Discuss the major tasks involved with natural language processing

• Discuss, on a low level, the components of natural language processing

• Identify why natural language processing is difficult

• Demonstrate text classification

• Demonstrate common text preprocessing techniques

Student Requirements

Before this lesson, students should already be able to:

• Use Anaconda for package management

• Use train/test/split to create a set of features and target values

• Read data into a Pandas DataFrame

• Build and evaluate predictive models using scikit-learn

Lesson Guide

• Introduction to Natural Language Processing

• Reading Yelp reviews with NLP

• Text Classification

• Count Vectorization

  • Using CountVectorizer in a model

  • N-Grams

  • Stopword Removal

  • Count Vector Options

• Intro to TextBlob

  • Stemming and Lemmatization

• Term Frequency Inverse Document Frequency Vectorization

  • Yelp Summary using TFIDF

• Sentiment Analysis

• BONUS: Adding Features to a Document Term Matrix

• BONUS: More TextBlob Features

• APPENDIX: Intro to Naive Bayes & Text Classification

• Conclusion

Installation Notes

To procede through the lesson, first install TextBlob as explained below. We tend to prefer Anaconda-based installations, since they tend to be tested with our other Anaconda packages. However, in this case TextBlob is not available on some platforms with Anaconda (e.g. Win64). To install textblob:

1. conda install -c https://conda.anaconda.org/sloria textblob

Or:

1. pip install textblob

2. python -m textblob.download_corpora lite

Additional Resources

For more information, we recommend the following resources:

• Check out this Yelp blog post how they completed a classification task (with over 1000 response variables!) using restaurant review text

• Always check documentation: CountVectorizer, HashingVectorizer, TF-IDF

• Wikpedia's feature hashing and hash functions is a great place to turn for more info on hashing

• Charlie Greenbacher's Intro to NLP

• Wikipedia includes a walkthrough of TF-IDF

• Google's ngram tool

• An experiment using NLP and Eigenfaces (Eigenvalues for face recognition) for Tinder