Assignment 10A CodeBase

Objective

The objective of this assignment is to reproduce and extend the sentiment analysis example presented in Chapter 2 of Text Mining with R: A Tidy Approach by Julia Silge and David Robinson (2017).

Initial Approach: Sentiment Analysis with Text Mining with R

I started out by reproducing the example given in Chapter 2, Sentiment Analysis, in the book Text Mining With R.

Extend the example:

For my extended analysis, I downloaded the IMDb Large Movie Review Dataset from Zenodo. I selected a balanced sample of 50 movie reviews, consisting of 25 positive and 25 negative reviews. I used the Bing and AFINN sentiment lexicons to compare how each method evaluates the reviews.

• Bing showed whether a word is positive or negative.

• AFINN showed how strongly positive or negative a word is.

That’s the comparison we’ll make using the movie reviews.

My additional package is textdata, and additional lexicon is AFINN

Reproducing the code From the textbook:

library(janeaustenr)
library(dplyr)

Attaching package: 'dplyr'

The following objects are masked from 'package:stats':

    filter, lag

The following objects are masked from 'package:base':

    intersect, setdiff, setequal, union

library(stringr)
library(tidytext)
library(ggplot2)
#creating tidy_books dataframe fom austen_books. 
tidy_books <- austen_books() %>%
  group_by(book) %>%
  mutate(
    linenumber = row_number(),
    chapter = cumsum(str_detect(text, 
                                regex("^chapter [\\divxlc]", 
                                      ignore_case = TRUE)))) %>%
  ungroup() %>%
  unnest_tokens(word, text)

nrc_joy <- get_sentiments("nrc") %>%
  filter(sentiment == "joy")

emma_joy <- tidy_books %>%
  filter(book == "Emma") %>%
  inner_join(nrc_joy, by = "word") %>%
  count(word, sort = TRUE)

head(emma_joy, 10)

# A tibble: 10 × 2
   word          n
   <chr>     <int>
 1 good        359
 2 friend      166
 3 hope        143
 4 happy       125
 5 love        117
 6 deal         92
 7 found        92
 8 present      89
 9 kind         82
10 happiness    76

library(tidyr)
jane_austen_sentiment <- tidy_books %>%
  inner_join(get_sentiments("bing")) %>%
  count(book, index = linenumber %/% 80, sentiment) %>%
  pivot_wider(names_from = sentiment, values_from = n, values_fill = 0) %>% 
  mutate(sentiment = positive - negative)

Joining with `by = join_by(word)`

Warning in inner_join(., get_sentiments("bing")): Detected an unexpected many-to-many relationship between `x` and `y`.
ℹ Row 435434 of `x` matches multiple rows in `y`.
ℹ Row 5051 of `y` matches multiple rows in `x`.
ℹ If a many-to-many relationship is expected, set `relationship =
  "many-to-many"` to silence this warning.

ggplot(
  jane_austen_sentiment,
  aes(index, sentiment, fill = book)
) +
  geom_col(show.legend = FALSE) +
  facet_wrap(~book, ncol = 2, scales = "free_x")

Sentiment Analysis of Movie Reviews:

Corpus: IMDb Large Movie Review Dataset

library(readr)
library(dplyr)
library(stringr)

#loading IMDB database
imdb_data <- read_csv(
  "IMDB Dataset.csv",
  show_col_types = FALSE
)

#Still verifying the dataset
imdb_data %>%
  count(sentiment)

# A tibble: 2 × 2
  sentiment     n
  <chr>     <int>
1 negative  25000
2 positive  25000

set.seed(607)
imdb_sample <- imdb_data %>%
  group_by(sentiment) %>%
  slice_sample(n = 25) %>%
  ungroup() %>%
  mutate(
    review_id = row_number()
  )

Cleaning up the line breaks in HTML reviews

imdb_sample <- imdb_sample %>%
  mutate(
    review = str_replace_all(
      review,
      regex("<br\\s*/?>", ignore_case = TRUE),
      " "
    )
  )

Do Tokenization

imdb_words <- imdb_sample %>%
  unnest_tokens(word, review)

Sentiment Analysis Using Bing

bing_matches <- imdb_words %>%
  inner_join(
    get_sentiments("bing"),
    by = "word",
    suffix = c("_review", "_bing")
  )

Calculating the positive and negative words for each review:

bing_by_review <- bing_matches %>%
  group_by(review_id) %>%
  summarise(
    positive_words = sum(sentiment_bing == "positive"),
    negative_words = sum(sentiment_bing == "negative"),
    bing_score = positive_words - negative_words,
    .groups = "drop"
  )
head(bing_by_review)

# A tibble: 6 × 4
  review_id positive_words negative_words bing_score
      <int>          <int>          <int>      <int>
1         1              8              7          1
2         2              5              9         -4
3         3              3              9         -6
4         4              5              3          2
5         5              8              9         -1
6         6             10             14         -4

AFINN sentiment analysis

afinn_matches <- imdb_words %>%
  inner_join(
    get_sentiments("afinn"),
    by = "word"
  )

#calculating the setiment score for each review
afinn_by_review <- afinn_matches %>%
  group_by(review_id) %>%
  summarise(
    afinn_score = sum(value),
    average_score = mean(value),
    matched_words = n(),
    .groups = "drop"
  )
head(afinn_by_review)

# A tibble: 6 × 4
  review_id afinn_score average_score matched_words
      <int>       <dbl>         <dbl>         <int>
1         1          11         0.524            21
2         2           1         0.111             9
3         3          -3        -0.231            13
4         4           7         0.778             9
5         5           6         0.462            13
6         6         -10        -0.357            28

COmparing Bing and AFINN reivews

review_word_counts <- imdb_words %>%
count(review_id, name = "total_words")

Combing the results.

imdb_results <- imdb_sample %>%
  select(review_id, actual_sentiment = sentiment) %>%
  left_join(review_word_counts, by = "review_id") %>%
  left_join(bing_by_review, by = "review_id") %>%
  left_join(afinn_by_review, by = "review_id") %>%
  mutate(
    across(
      c(positive_words, negative_words, bing_score,
        afinn_score, average_score, matched_words),
      ~ tidyr::replace_na(.x, 0)
    ),
    bing_per_100 = 100 * bing_score / total_words,
    afinn_per_100 = 100 * afinn_score / total_words
  )

head(imdb_results)

# A tibble: 6 × 11
  review_id actual_sentiment total_words positive_words negative_words
      <int> <chr>                  <int>          <int>          <int>
1         1 negative                 227              8              7
2         2 negative                 138              5              9
3         3 negative                 119              3              9
4         4 negative                 246              5              3
5         5 negative                 127              8              9
6         6 negative                 367             10             14
# ℹ 6 more variables: bing_score <int>, afinn_score <dbl>, average_score <dbl>,
#   matched_words <int>, bing_per_100 <dbl>, afinn_per_100 <dbl>

comparison <- imdb_results %>%
  group_by(actual_sentiment) %>%
  summarise(
    number_of_reviews = n(),
    bing_average = mean(bing_per_100),
    afinn_average = mean(afinn_per_100),
    .groups = "drop"
  )

comparison

# A tibble: 2 × 4
  actual_sentiment number_of_reviews bing_average afinn_average
  <chr>                        <int>        <dbl>         <dbl>
1 negative                        25        -1.09         0.241
2 positive                        25         3.97        10.3  

Graph:

 # Preparing data for the graph

imdb_long <- imdb_results %>%
  select(
    review_id,
    actual_sentiment,
    bing_per_100,
    afinn_per_100
  ) %>%
  pivot_longer(
    cols = c(bing_per_100, afinn_per_100),
    names_to = "lexicon",
    values_to = "score"
  ) %>%
  mutate(
    lexicon = recode(
      lexicon,
      bing_per_100 = "Bing",
      afinn_per_100 = "AFINN"
    )
  )

# Creating the graph

ggplot(
  imdb_long,
  aes(
    x = actual_sentiment,
    y = score,
    fill = actual_sentiment
  )
) +
  geom_boxplot(show.legend = FALSE) +
  facet_wrap(~lexicon, scales = "free_y") +
  labs(
    title = "Sentiment Analysis of IMDb Movie Reviews",
    subtitle = "Comparing Bing and AFINN",
    x = "Original IMDb Review Label",
    y = "Sentiment Score per 100 Words"
  ) +
  theme_minimal()

Conclusion

Overall, this assignment helped me understand how sentiment analysis works and how different lexicons can produce different results. I found it interesting that Bing and AFINN did not always agree, even though they were analyzing the same movie reviews. Both showed higher sentiment scores for positive reviews, but AFINN actually gave the negative reviews a slightly positive average score. This made me realize that sentiment analysis isn’t always as straightforward as counting positive and negative words because context and how words are used also matter.