step3.R

## Authorship prediction
## authorship of some Federalist Papers is unknown
## We use the 66 essays attributed to either Hamilton or Madison to 
## predict the authorship of the 11 disputed papers.

## Since each paper deals with a different topic, we focus on usage of articles, 
## prepositions, and conjuctions. We analyze the frequency of the following
## 10 words: although, always, commonly, consequently, considerable, enough, there, upon, while, 
## and whilst.

## We selected these words based upon an academic paper that inspired this exercise.

## As a result, we must the unstemmed corpus, corpus prep.

## We first compute the the term frequency (per 1000 words) separately for each term
## and document and then subset the resulting term-frequency matrix to
## contain only these words.

dtm1 <- as.matrix(DocumentTermMatrix(corpus.prep))
tfrequency <- dtm1 / rowSums(dtm1)  # term frequency
tfm <- tfrequency*1000 # term freqency times 1000 (i.e., per 1000 words)

## words of interest
words <- c("although", "always", "commonly", "consequently", 
           "considerable", "enough", "there", "upon", "while", 
           "whilst")

## select only these words
#tfm <- tfm[, colnames(tfm) == words]
tfm <- tfm[, which(is.element(colnames(tfm), words))]

## We can then calculate the average term frequency separately for 
## Hamilton and Madison across each author's entire body of documents.
hamilton <- c(1, 6:9, 11:13, 15:17, 21:36, 59:61, 65:85)
madison <- c(10, 14, 37:48, 58) # 

## average among Hamilton/Madison essays
tfm.ave <- rbind(colSums(tfm[hamilton, ]) / length(hamilton),
                 colSums(tfm[madison, ]) / length(madison))

tfm.ave