Analyse text reuse using minhash and locality-sensitive hashing (LSH)

gistfile1.txt

library(tidyverse)
library(textreuse)

# one row per student, to get the data, go to canvas -> quiz -> 'quiz stats' -> 'student analysis'
cnvs <- read_csv("quiz-responses-downloaded-from-canvas.csv")

# select only the column with the text we want to compare
cnvs_q5 <- 
cnvs %>% 
  select( q5 = contains("peers' reflections")) 

# set the hashing parameters
n_hashes <- 1000
n_bands <- n_hashes / 2

# check that number of hashes is evenly divisible by the number of bands
# this must be TRUE
n_hashes %% n_bands == 0

minhash <- minhash_generator(n = n_hashes, 
                             seed = 3552)
cnvs_q5_crps <- 
  TextReuseCorpus(text = cnvs_q5$q5,
                  minhash_func = minhash, 
                  keep_tokens = TRUE)

# lsh_threshold(h = n_hashes, b = n_bands)

#  locality-sensitive hashing algorithm,
buckets <- lsh(cnvs_q5_crps, 
               bands = n_bands)

candidates <- lsh_candidates(buckets)

# make a table of pairwise similarity scores
comparison_tbl <- 
lsh_compare(candidates, 
            cnvs_q5_crps, 
            jaccard_similarity) %>% 
  arrange(desc(score))

# inspect raw text of top scoring candidates to
# see exactly what they wrote:
content(cnvs_q5_crps[[comparison_tbl$a[1]]])
content(cnvs_q5_crps[[comparison_tbl$b[1]]])

content(cnvs_q5_crps[[comparison_tbl$a[2]]])
content(cnvs_q5_crps[[comparison_tbl$b[2]]])

content(cnvs_q5_crps[[comparison_tbl$a[3]]])
content(cnvs_q5_crps[[comparison_tbl$b[3]]])

content(cnvs_q5_crps[[comparison_tbl$a[4]]])
content(cnvs_q5_crps[[comparison_tbl$b[4]]])