Text Analysis in R

Thursday: Introduction to R

Friday: Corpus Analysis & Topic Modeling

Saturday:

• Sentiment Analysis
• Machine Learning

Sunday:

• Basic visualization
• Semantic Network Analysis
• Graph Visualization

• What is the tone of a text?
• Techniques (e.g. Pang/Lee 2008, Liu 2012)
  • Manual coding
  • Dictionaries
  • Machine Learning
  • Crowdsourcing (Benoit ea 2015)

“The man who leaked cell-phone coverage of Saddam Hussein's execution was arrested”

• Language is subjective, ambiguous, creative
• What does positive/negative mean?
  • e.g. Osgood ea 1957: evaluation, potency, activity
• Who is positive/negative about what?
  • Sentiment Attribution

• Lexicon (dictionary)
• Annotated texts
• Tools / models

• Get list of positive / negative terms
• Count occurrences in text
• Summarize to sentiment score
• Possible improvements
  • Word-window approach (tomorrow)
  • Deal with negation, intensification

• Nothing new here!
• Directly count words in DTM:

library(slam)
reviews$npos = row_sums(dtm[, colnames(dtm) %in% pos_words])

• Count words in token list:

tokens$sent[tokens$lemma %in% pos_words] = 1

Lexical Sentiment Analysis

Thursday: Introduction to R

Friday: Corpus Analysis & Topic Modeling

Saturday:

• Sentiment Analysis
• Machine Learning

Sunday:

• Basic visualization
• Semantic Network Analysis
• Graph Visualization

• Statistical Modeling
  • Dependent Variable: sentiment, topic, frame
  • Independent Variables: words
• Focus: prediction rather than explanation
  • Millions of correlated independent variables

• Each text has a 'class'
• Training documents to fit model
• Test documents to gauge accuracy
  • (or use cross-validation)
• Choices:
  • What features?
  • Which model?

• Features: independent variables
• Basic approach: each word is a feature
• Other options e.g.
  • Collocations (n-grams)
  • LDA Topics
  • Feature selection

• Naive Bayes
• Maximum Entropy
• Support Vector Machines
• Neural Networks
• (deep learning)

• Ensemble Learning
  • Train multiple models
  • Decide by vote
• Active Learning
  • Code limited amount of material
  • Train+test model
  • Code most difficult cases, repeat

• Package RTextTools
  • Jurka et al, 2013
• Based on DTM plus coded classes
• Does learning, evaluation, prediction

(1) Create 'container' from DTM + coded classes

library(RTextToools)
c = create_container(dtm, classes, 
  trainSize=train, testSize=test, virgin=F)

(2) Train and test model

SVM <- train_model(c,"SVM")
SVM_CLASSIFY <- classify_model(c, SVM)

(3) Evaluate

analytics <- create_analytics(c, SVM_CLASSIFY)

is_coded = !is.na(classes)
c = create_container(dtm, classes, 
  trainSize=is_coded, virgin=T)
SVM <- train_model(c,"SVM")
SVM_CLASSIFY <- classify_model(c, SVM)
analytics <- create_analytics(c, SVM_CLASSIFY)
head(analytics@document_summary)

• Burscher et al 2014: Framing with SVM's

• Purpura & Wilkerson 2007: Active Learning for Agenda Coding

• Some online resources:

  • http://www.r-bloggers.com/sentiment-analysis-on-donald-trump-using-r-and-tableau/
  • https://www.cs.uic.edu/~liub/FBS/sentiment-analysis.html#lexicon
  • http://datascienceplus.com/sentiment-analysis-with-machine-learning-in-r/
  • https://sites.google.com/site/miningtwitter/questions/sentiment/sentiment

Text Classification for Sentiment Analysis

Break

Handouts:

• Sentiment Analysis Resources
• Lexcial Sentiment Analysis
• Machine Learning