Seeded-LDA for semisupervised topic modeling
seededlda is an R package that implements the seeded-LDA for semisupervised topic modeling using quanteda. The seeded-LDA model was proposed by Lu et al. (2010). Until version 0.3, that packages has been a simple wrapper around the topicmodels package, but the LDA estimator is newly implemented in C++ using the GibbsLDA++ library to be submitted to CRAN in August 2020. The author believes this package implements the seeded-LDA model more closely to the original proposal.
Please see Theory-Driven Analysis of Large Corpora: Semisupervised Topic Classification of the UN Speeches for the overview of semisupervised topic classification techniques and their advantages in social science research.
keyATM is the latest addition to the semisupervised topic models. The users of seeded-LDA are also encouraged to use that package.
Install
install.packages("devtools")
devtools::install_github("koheiw/seededlda") Example
The corpus and seed words in this example are from Conspiracist propaganda: How Russia promotes anti-establishment sentiment online?.
require(quanteda)
require(seededlda)Users of seeded-LDA must provided a small dictionary of keywords (seed words) to define the desired topics.
dict <- dictionary(file = "tests/data/topics.yml")
print(dict)
## Dictionary object with 5 key entries.
## - [economy]:
## - market*, money, bank*, stock*, bond*, industry, company, shop*
## - [politics]:
## - parliament*, congress*, white house, party leader*, party member*, voter*, lawmaker*, politician*
## - [society]:
## - police, prison*, school*, hospital*
## - [diplomacy]:
## - ambassador*, diplomat*, embassy, treaty
## - [military]:
## - military, soldier*, terrorist*, air force, marine, navy, armycorp <- readRDS("tests/data/data_corpus_sputnik.RDS")
toks <- tokens(corp, remove_punct = TRUE, remove_symbols = TRUE, remove_number = TRUE) %>%
tokens_select(min_nchar = 2) %>%
tokens_compound(dict) # for multi-word expressions
dfmt <- dfm(toks) %>%
dfm_remove(stopwords('en')) %>%
dfm_trim(min_termfreq = 0.90, termfreq_type = "quantile",
max_docfreq = 0.2, docfreq_type = "prop")Many of the top terms of the seeded-LDA are seed words but related topic
words are also identified. The result includes “other” as a junk topic
because residual = TRUE.
set.seed(1234)
slda <- textmodel_seededlda(dfmt, dict, residual = TRUE)
print(terms(slda, 20))
## economy politics society diplomacy
## [1,] "company" "parliament" "police" "diplomatic"
## [2,] "money" "congress" "school" "embassy"
## [3,] "market" "white_house" "hospital" "ambassador"
## [4,] "bank" "politicians" "prison" "treaty"
## [5,] "industry" "parliamentary" "schools" "diplomat"
## [6,] "banks" "lawmakers" "pic.twitter.com" "diplomats"
## [7,] "markets" "voters" "media" "like"
## [8,] "banking" "lawmaker" "reported" "just"
## [9,] "stock" "politician" "local" "now"
## [10,] "stockholm" "minister" "information" "think"
## [11,] "china" "european" "video" "even"
## [12,] "percent" "sanctions" "public" "trump"
## [13,] "chinese" "eu" "social" "going"
## [14,] "economic" "political" "court" "made"
## [15,] "india" "party" "women" "years"
## [16,] "year" "foreign" "man" "way"
## [17,] "oil" "prime" "report" "say"
## [18,] "project" "union" "found" "want"
## [19,] "billion" "moscow" "investigation" "many"
## [20,] "million" "trump" "department" "really"
## military other
## [1,] "army" "north"
## [2,] "terrorist" "nuclear"
## [3,] "navy" "korea"
## [4,] "terrorists" "south"
## [5,] "air_force" "iran"
## [6,] "soldiers" "trump"
## [7,] "marine" "korean"
## [8,] "soldier" "world"
## [9,] "defense" "israel"
## [10,] "syria" "deal"
## [11,] "syrian" "saudi"
## [12,] "forces" "kim"
## [13,] "security" "show"
## [14,] "nato" "israeli"
## [15,] "weapons" "agreement"
## [16,] "daesh" "program"
## [17,] "turkish" "cup"
## [18,] "turkey" "trump's"
## [19,] "international" "japan"
## [20,] "group" "peace"topic <- table(topics(slda))
print(topic)
##
## economy politics society diplomacy military other
## 136 181 262 158 144 119Examples
Please read the following papers for how to apply seeded-LDA in social science research:
Curini, Luigi and Vignoli, Valerio. 2021. Committed Moderates and Uncommitted Extremists: Ideological Leaning and Parties’ Narratives on Military Interventions in Italy, Foreign Policy Analysis.