Resampled normalized pointwise mutual information (rnpmi)

Explore correlations and sequences in qualitative and quantitative content analysis.

The package is in pre alpha state. Use with care. Numbers might be hurt.

Installation

library(devtools)
install_github("strohne/npmi", build_vignettes = TRUE)

Usage

# Packages
library(tidyverse)
library(npmi)

# Load example data
data <- read_csv2(system.file("extdata", "threads.csv", package = "npmi"))

# Cooccurence
pairs_coo <- data %>%
  select(item=id,feature,weight) %>%
  count_pairs()

# Resample npmi for cooccurrence
pairs_coo <- data %>%
  select(item=id,feature,weight) %>%
  get_cooccurrence()

# Sequences
pairs_seq <- data %>%
  rename(item=id,item_parent=parent_id) %>% 
  add_previous_item() %>% 
  count_sequences()

# Resample npmi for sequences
pairs_seq <- data %>%
  rename(item=id,item_parent=parent_id) %>% 
  add_previous_item() %>% 
  get_sequences()

# Plot absolute numbers
pairs_seq %>% 
  rename(value=n) %>% 
  matrixmap() 

# Plot probabilities
pairs_seq %>% 
  rename(value=p) %>% 
  matrixmap() 

# Plot rnpmi
pairs_seq %>% 
  rename(value=npmi) %>% 
  matrixmap()   
  
# Plot network
pairs_seq %>% 
  rename(value=npmi) %>% 
  network()

See the vignettes for further examples (either browse the vignettes folder or open vignettes from the package help). Vignettes are not polished yet.

Resampling can be parallelized, just call:

library(future)
plan(multisession)

Presentations

Jünger, Jakob (2021). Unseen correlations: On the identification of rare events and sequences in online comment threads. Annual Conference of the Methods Division of the German Communication Association (DGPuK), 2021, Virtual Vienna.