In this Sentiment Analysis tutorial, You’ll learn how to use your custom lexicon (for any language other than English) or keywords dictionary to perform simple (slightly naive) sentiment analysis using R’s tidytext package. Note: This isn’t going to provide you the same accuracy as using the language model, but it’s going to get you to the fastest solution (with some accuracy tradeoff). This example deals with Turkish Sentiment Analysis Script. Please note this tutorial doesn’t include Text Pre-processing steps, but those are very important for any Text Analytics / NLP project.
Video Walkthrough
Steps
- Read the Input Text as a Dataframe
- Load the lexicon / new language dictionary
- Select the appropriate columns - in this case, word and polarity
- Join the tokenized words from the text dataframe with the lexicon dataframe
- Roll-up the result dataframe based on the grouping variable (row_number) to get sentence level aggregated sentiment score
Code
library(tidyverse)
#install.packages("tidytext")
library(tidytext)
sent <- read.csv('text.csv')
lexicon <- read.table("turkish_lexicon.csv",
header = TRUE,
sep = ';',
stringsAsFactors = FALSE)
lexicon2 <- lexicon %>%
select(c("WORD","POLARITY")) %>%
rename('word'="WORD",'value'="POLARITY")
sent %>%
mutate(linenumber = row_number()) %>% #line number for later sentence grouping
unnest_tokens(word, tweettext) %>% #tokenization - sentence to words
inner_join(lexicon2) %>% # inner join with our lexicon to get the polarity score
group_by(linenumber) %>% #group by for sentence polarity
summarise(sentiment = sum(value)) %>% # final sentence polarity from words
left_join(
sent %>%
mutate(linenumber = row_number()) #get the actual text next to the sentiment value
) %>% write.csv("sentiment_output.csv",row.names = FALSE)