The Goal

We have been learning how to work with text data, creating features and starting to build models. Today, we are going to apply everything we have learned so far to our first data analysis task.

This first Data Analysis will be more structured than Data Analysis 2 and 3. This is because part of the goal of this analysis is to help you practice the steps and structure of a data analysis.

The Data Set

The Federalist papers are a famous series of articles written in the 1700s in the early days of the United States. They debated and proposed ideas for the new government. Two famous authors of these papers were Alexander Hamilton and James Madison. Most of the papers were know to have been written by either Hamilton or Madison, but for several papers, the author who wrote them was not known. Hamilton declared at one point that he had written the articles. Since then, many scholars have worked to determine which of the two authors had written the anonymous papers.

Today, we are provided with the text from the Federalist papers known to have been written by Hamilton and Madison. The data come from https://github.com/nicholasjhorton/FederalistPapers, the GitHub repository of Dr. Nicholas J Horton.

Our goal is to use the text analysis skills we have learned so far to determine what characteristics might differ between the two authors’ papers. We will also be building a model to predict which author wrote the mystery papers.

To read in the data on the \(n = 65\) Federalist papers with known authors, use the following code:

# Load the data
Federalist <- read.csv("https://www.dropbox.com/scl/fi/5hzqwsvlnym5u1mhmn0jy/Federalist.csv?rlkey=55x0p9fl02zls9ixxek64vlqy&st=9ku4q61a&dl=1")

# Convert to a data frame
Federalist <- data.frame(Federalist)

# Make sure author is treated as categorical
Federalist$author <- as.factor(Federalist$author)

The columns are:

• paper: the number given to the paper; this of this like an identifier for the article.
• text: the text of the entire article.
• author: either Hamilton or Madison

Once you have loaded the data, load the packages you will need for this lab:

library(tidytext)
library(tidyr)
library(dplyr)
library(ggplot2)
library(tm)

Section 1: Word Count

At this point, we know that a common feature we use to compare documents is word count. Is it possible that the number of words can be used to determine which author wrote which paper?

Now that the feature has been created, we have the ability to use word length to build a logistic regression model to predict whether a paper is written by Madison or Hamilton.

Section 2: Word Frequency

In addition to the number of words, we have learned that we can count the number of times each words appears in a piece of text. This can be used to determine what words are more common in one type of text than another.

At this point, we are going to try something a little new. We are going to create features using these top 10 words, and then we will be using these in a logistic regression model to predict authorship.

# Find the top 10
Hamilton10 <- Federalist %>%
  filter(author =="Hamilton") |>
  mutate(text = removePunctuation(text)) |>
  unnest_tokens(word, text)|>
  anti_join(stop_words) |>
  # Remove the numbers 
  filter(!grepl('[0-9]', word)) |>
  # Count the number of words 
  count(word, sort = TRUE) |>
  # Re-order the words so the most popular is on top
  mutate(word = reorder(word, n)) |>
  # Choose the most 10 popular words 
  slice(1:10)

top10 <- union( Madison10$word, Hamilton10$word)

# DO NOT EDIT!!!
# THIS IS A FUNCTION 

topWordsData <- function( data, classifier, top ){
  corpus = Corpus(VectorSource(data)) 
  
  corpus = suppressWarnings(tm_map(corpus, content_transformer(tolower))) 
  corpus = suppressWarnings(tm_map(corpus, removePunctuation))
  corpus = suppressWarnings(tm_map(corpus, removeNumbers))
  
  # removing stop words 
  corpus = suppressWarnings(tm_map(corpus, removeWords, stopwords("SMART"))) 
  # removing white space 
  corpus = suppressWarnings(tm_map(corpus, stripWhitespace))
  
  matrix = as.matrix(DocumentTermMatrix(corpus)) 

  #converting matrix to dataframe 
  dtm_df <- as.data.frame(matrix) 
  
  # add on the classifier
  dtm_df$y <- classifier
  
  # Re-order columns
  dtm_df <- dtm_df[, c(ncol(dtm_df), 1:(ncol(dtm_df)-1))]
  
  # Keep only certain columns
  colnamesKeep <- which(colnames(dtm_df) %in% top)
  
  dtm_df <-dtm_df[, c(1,colnamesKeep)]
  
  dtm_df[,1] <-as.factor(dtm_df[,1])
  
  # Outut 
  dtm_df
}

Top10 <- topWordsData( Federalist$text, Federalist$author, top10 )

Section 3

At this point, we have seen how well our two different logistic regression models can do at predicting for the papers we do know the authors for. However, the whole point was to see if we could identify who wrote the papers we do not have authors for.

To load the text data for the papers we do not have authors for, use the following:

test <- read.csv("https://www.dropbox.com/scl/fi/yrfzwwn7olyaeq3mhb0bd/test.csv?rlkey=c6bs0ytyoomkq34aw7rx3bc62&st=tzvp12cp&dl=1")

test <-data.frame(test)

You will notice that this data set has paper number and the paper text, but it does not have the author, since that is what we are trying to predict.

This new data was not used in the model building process, but we are about to make predictions for this new data. This means these data are called test data because we are using them to test / assess the predictive abilities of our model.

The data we used to build the model originally are called training data.

Final Steps

You are now ready to submit! This will be graded like a project, so make sure you use complete sentences, check your spelling, format all your output professionally, and label your graphs appropriately.

References