Text as Data

A general pipeline for analyzing text data: Acquire, preprocess, process and analyze text data.

1. Get Text
2. Preprocess Text
3. Subset, Take a Random Sample, Summarize
4. Create tdm/tf-idf
5. Analyze
   • Analyze Text in R
   • Sentiment Analysis in Python

Get Text

You can get text data from scraping, APIs, searchable pdfs, images of paper, etc. Some examples:

• Get text from searchable pdfs. e.g. Get data from Wisconsin Ads storyboards using Python
• Get text from images of text using Tesseract from Python
• Get text from images of text using Abbyy FineReader Cloud OCR from R
• Get text from images of text using Captricity OCR from R
• Get Congressional Speech Data using Capitol Words API from the Sunlight Foundation

Preprocess Text

Preprocess text for text-as-data analysis.

Depending on the need, remove stop words, punctuation, capitalization, special characters, and stem.

• preprocess_csv takes a csv with 'raw' text and outputs a csv with processed text.

Get Summary of the Data, Subset Data

Output a simple or stratified random sample of a csv, and only the columns you need. Get summary of crucial aspects of the data. Takes a csv.

• Summarize and Subset.

Get TDM

Create a term-document-matrix and get some information about the matrix including frequent and infrequent terms. Options available for removing sparse terms etc.

• Get TDM, TF-IDF, Summary.

Sentiment Analysis in Python

• Basic sentiment analysis using AFINN

Analyze Text in R

• Classify text in R using SVM or Lasso. See Basic Text Classifier
• Worked out example of how to model words as a function of ideology using Congressional Speech. See Speech Learn

License

Scripts are released under the MIT License.