Introduction to clustRcompaR

In clustRcompaR: Easy Interface for Clustering a Set of Documents and Exploring Group- Based Patterns

clustRcompaR

An R package to cluster and compare text data.

Background

Document clustering is a common technique to discover topics in a corpus of texts. This package uses functions from the quanteda R package as the basis for two functions, cluster() and `compare(), to make document clustering and comparing topics identified through document clustering across factors straightforward.

Installation

Because this package is in development and is not yet available on CRAN, to install it, first install the devtools package using install.packages("devtools"), followed by the function devtools::install_github("alishinski/clustRcompaR"). After installing the package, use library(clustRcompaR) to load it in each session.

Workflow

• First, use cluster() on a data.frame with the first column a vector of strings and any subsequent columns vectors of factors.

• Optional arguments to the cluster() function include parameters for the minimum frequency with which a term must occur to be included in the analysis, the minimum number of terms in each document after processing, and additional stopwords. The output from the cluster() function can then be inspected to determine the interpretability of clusters and the suitability of the clustering solution.

• Next, use compare() with the output from the cluster() function along with a string for the factor to compare the frequency of clusters to.

• This output can be examined and used on its own, or be passed to two functions, compare_plot(), which plots the table using ggplot2, or compare_test(), which performs a chi-square test of proportions on the table and indicates across which levels of the factor clusters appear more or less likely than expected.

Example

Here is an example using the built-in inaugural_addresses dataset (from the quanteda package). This dataset consists of the inaugural addresses by every United States President.

First, we use cluster() to cluster the documents into three clusters. We include a new variable, year_before_1900, which we will later use to compare frequencies across clusters. Then we use extract_terms() to view the terms and term frequencies in the two clusters.

library(clustRcompaR)
library(dplyr)
library(quanteda)

d <- inaugural_addresses
d <- mutate(d, century = ifelse(Year < 1800, "17th",
                                ifelse(Year >= 1800 & Year < 1900, "18th",
                                       ifelse(Year >= 1900 & Year < 2000, "19th", "20th"))))

three_clusters <- cluster(d, century, n_clusters = 3)

extract_terms(three_clusters)

Second, we use the compare() function to compare the frequency of clusters across a factor, in this case, the century. We can then use the compare_plot() or compare_test() (which uses a Chi-Square test) function.

three_clusters_comparison <- compare(three_clusters, "century")

compare_plot(three_clusters_comparison)

clustRcompaR documentation built on May 1, 2019, 11:16 p.m.