Aggregate sentiment measures

Aggregates sentiment measures by combining across provided lexicons, features, and time weighting schemes dimensions. For do.global = FALSE, the combination occurs by taking the mean of the relevant measures. For do.global = TRUE, this function aggregates all sentiment measures into a weighted global textual sentiment measure for each of the dimensions.

# S3 method for class 'sento_measures'
aggregate(
  x,
  features = NULL,
  lexicons = NULL,
  time = NULL,
  do.global = FALSE,
  do.keep = FALSE,
  ...
)

Arguments

x

a sento_measures object created using sento_measures.

features

a list with unique features to aggregate at given name, e.g.,
list(feat12 = c("feat1", "feat2")). See x$features for the exact names to use. Use NULL (default) to apply no merging across this dimension. If do.global = TRUE, should be a numeric vector of weights, of size length(x$features), in the same order. A value of NULL means equally weighted.

lexicons

a list with unique lexicons to aggregate at given name, e.g.,
list(lex12 = c("lex1", "lex2")). See x$lexicons for the exact names to use. Use NULL (default) to apply no merging across this dimension. If do.global = TRUE, should be a numeric vector of weights, of size length(x$lexicons), in the same order. A value of NULL means equally weighted.

time

a list with unique time weighting schemes to aggregate at given name, e.g.,
list(tw12 = c("tw1", "tw2")). See x$time for the exact names to use. Use NULL (default) to apply no merging across this dimension. If do.global = TRUE, should be a numeric vector of weights, of size length(x$time), in the same order. A value of NULL means equally weighted.

do.global

a logical indicating if the sentiment measures should be aggregated into weighted global sentiment indices.

do.keep

a logical indicating if the original sentiment measures should be kept (i.e., the aggregated sentiment measures will be added to the current sentiment measures as additional indices if do.keep = TRUE).

...

not used.

Value

If do.global = FALSE, a modified sento_measures object, with the aggregated sentiment measures, including updated information and statistics, but the original sentiment scores data.table untouched.

If do.global = TRUE, a data.table with the different types of weighted global sentiment measures, named "globLex", "globFeat", "globTime" and "global", with "date" as the first column. The last measure is an average of the the three other measures.

Details

If do.global = TRUE, the measures are constructed from weights that indicate the importance (and sign) along each component from the lexicons, features, and time dimensions. There is no restriction in terms of allowed weights. For example, the global index based on the supplied lexicon weights ("globLex") is obtained first by multiplying every sentiment measure with its corresponding weight (meaning, the weight given to the lexicon the sentiment is computed with), then by taking the average per date.

Author

Samuel Borms

Examples

data("usnews", package = "sentometrics")
data("list_lexicons", package = "sentometrics")
data("list_valence_shifters", package = "sentometrics")

# construct a sento_measures object to start with
corpus <- sento_corpus(corpusdf = usnews)
corpusSample <- quanteda::corpus_sample(corpus, size = 500)
l <- sento_lexicons(list_lexicons[c("LM_en", "HENRY_en")],
                    list_valence_shifters[["en"]])
ctr <- ctr_agg(howTime = c("equal_weight", "linear"),
               by = "year", lag = 3)
sento_measures <- sento_measures(corpusSample, l, ctr)

# aggregation across specified components
smAgg <- aggregate(sento_measures,
                   time = list(W = c("equal_weight", "linear")),
                   features = list(journals = c("wsj", "wapo")),
                   do.keep = TRUE)

# aggregation in full
dims <- get_dimensions(sento_measures)
smFull <- aggregate(sento_measures,
                    lexicons = list(L = dims[["lexicons"]]),
                    time = list(T = dims[["time"]]),
                    features = list(F = dims[["features"]]))

# "global" aggregation
smGlobal <- aggregate(sento_measures, do.global = TRUE,
                      lexicons = c(0.3, 0.1),
                      features = c(1, -0.5, 0.3, 1.2),
                      time = NULL)

if (FALSE) { # \dontrun{
# aggregation won't work, but produces informative error message
aggregate(sento_measures,
          time = list(W = c("equal_weight", "almon1")),
          lexicons = list(LEX = c("LM_en")),
          features = list(journals = c("notInHere", "wapo")))} # }