In this report,

# Package names
packages <- c("RedditExtractoR", "tidytext", "tidyverse", "textdata", "anytime", "magrittr", "wordcloud2",
              "ggdark","gofastr", "syuzhet", "sentimentr", "lubridate", "here")

# Install packages not yet installed
installed_packages <- packages %in% rownames(installed.packages())
if (any(installed_packages == FALSE)) {
install.packages(packages[!installed_packages])
}

# Load packages
invisible(lapply(packages, library, character.only = TRUE))
## ── Attaching core tidyverse packages ──────────────────────── tidyverse 2.0.0 ──
## ✔ dplyr     1.1.3     ✔ readr     2.1.4
## ✔ forcats   1.0.0     ✔ stringr   1.5.0
## ✔ ggplot2   3.4.3     ✔ tibble    3.2.1
## ✔ lubridate 1.9.3     ✔ tidyr     1.3.0
## ✔ purrr     1.0.2     
## ── Conflicts ────────────────────────────────────────── tidyverse_conflicts() ──
## ✖ dplyr::filter() masks stats::filter()
## ✖ dplyr::lag()    masks stats::lag()
## ℹ Use the conflicted package (<http://conflicted.r-lib.org/>) to force all conflicts to become errors
## 
## Attaching package: 'magrittr'
## 
## 
## The following object is masked from 'package:purrr':
## 
##     set_names
## 
## 
## The following object is masked from 'package:tidyr':
## 
##     extract
## 
## 
## 
## Attaching package: 'sentimentr'
## 
## 
## The following object is masked from 'package:syuzhet':
## 
##     get_sentences
## 
## 
## here() starts at /Users/xy/Downloads/tutorial

Read the RDS data and write as .csv file

## Read the RDS data and write as .csv file
data <- readRDS("reddit.rds")

# Loop through each row of the dataframe
results <- character(nrow(data)) 
for (i in seq_len(nrow(data))) {
  # Extract the string that follows '/r/' and precede the next '/'
  results[i] <- str_extract(data$permalink[i], "(?<=/r/)[^/]+")
}
# You can add the results back to your data frame if you want
data$subreddit <- results
data$extracted_bs<- NULL

write.csv(data, "sample_reddit1.csv", row.names = FALSE)

Sentiment Analysis

Import the data processed from the Colab to R.

reddit_sentiment <- read_csv('sample_reddit_bert2.csv') %>% drop_na('bert_label')
## Rows: 1574 Columns: 23
## ── Column specification ────────────────────────────────────────────────────────
## Delimiter: ","
## chr  (12): tag, title, author, comments, id, permalink, url, comments_feed, ...
## dbl   (9): ...1, num_comments, score, upvote_ratio, created_utc, time, bert_...
## dttm  (1): date
## date  (1): year
## 
## ℹ Use `spec()` to retrieve the full column specification for this data.
## ℹ Specify the column types or set `show_col_types = FALSE` to quiet this message.

Group the bubbles

reddit_sentiment <- reddit_sentiment %>%
  mutate(group = case_when(
    subreddit == "Atlanta" | tag == "beltline" ~ "Atlanta",
    subreddit == "chicago" | subreddit == "chibike" | tag == "606" | tag == "bloomingdale trail" ~ "Chicago",
    subreddit == "nyc" | subreddit == "NYCbike" | subreddit == "MicromobilityNYC" | tag == "highline" ~ "NYC",
    subreddit == "Detroit" | subreddit == "BikeDetroit" | tag == "Joe Louis Greenway" | tag == "dequindre cut" ~ "Detroit",
    TRUE ~ "Other"  # Assign "Other" to rows that do not match any of the above conditions
  ))
reddit_sentiment %<>% mutate(bert_label_numeric = str_sub(bert_label, 1, 1) %>% as.numeric())

3. Visualization

We’ve extracted thread texts related to rail to trails from Reddit for our dataset, and processed this data through the BERT model.Below is the visualization for the findings.

3-1: Sentiment distribution

Number of threads by sentiment category.

reddit_sentiment %>%
  ggplot(aes(x = bert_label)) + 
  geom_bar(fill = "gray") +
  geom_text(stat = 'count', aes(label = ..count..), vjust = -0.5, size = 2.5, color = "black") +  # Adjust the 'size' for smaller text
  labs(title = "Distribution of BERT Labels", x = "BERT Label", y = "Count") +  # Add plot title and axis labels
  theme_light() +
  theme(
    plot.title = element_text(size = 14, face = "bold", hjust = 0.5),  # Adjust the plot title size
    axis.title.x = element_text(size = 12, face = "bold"),  # Adjust the x axis label size
    axis.title.y = element_text(size = 12, face = "bold"),  # Adjust the y axis label size
    axis.text.x = element_text(size = 10),  # Adjust the x axis text size
    axis.text.y = element_text(size = 10)   # Adjust the y axis text size
  )
## Warning: The dot-dot notation (`..count..`) was deprecated in ggplot2 3.4.0.
## ℹ Please use `after_stat(count)` instead.
## This warning is displayed once every 8 hours.
## Call `lifecycle::last_lifecycle_warnings()` to see where this warning was
## generated.

  • Finding:The ‘Distribution of BERT Labels’ chart illustrates the sentiment scores. It’s interesting to see a significant number of 4-star and 5-star ratings, pointing to positive sentiment, and a notable amount of 1-star ratings, indicating extreme negative sentiment.There are fewer 2-star and 3-star ratings, suggesting that moderate or neutral sentiments are less frequently expressed.The users tend to post more extreme sentiments, either very satisfied or unsatisfied, rather than a moderate stance.

Sentiment Analysis by BERT Label (Percentage) and Cities

# Define colors for each group
group_colors <- c("Other" = "gray", 
                  "Atlanta" = "pink", 
                  "NYC" = "plum", 
                  "Detroit" = "lightblue", 
                  "Chicago" = "lightgreen")

# Ensure your data has the 'group' column as a factor and levels exactly match the names in group_colors
reddit_sentiment$group1 <- factor(reddit_sentiment$group, levels = names(group_colors))

# Calculate the percentage for each group within each bert_label
reddit_sentiment1 <- reddit_sentiment %>%
  group_by(bert_label, group) %>%
  summarise(count = n(), .groups = 'drop') %>%
  mutate(percentage = count / sum(count))

# Create the plot
gg <- reddit_sentiment1 %>%
  ggplot(aes(x = bert_label, y = percentage, fill = group)) +
  geom_bar(position = "dodge", stat = "identity", color = "black") +
  scale_fill_manual(values = group_colors) +  # Manually set colors
  theme_light() +  # Apply a light theme
  labs(title = "Sentiment Analysis by BERT Label and Cities",
       y = "Percentage", fill = "Group")  # Update y-axis label and legend label

# Print the plot
print(gg)

  • Findings:This graph presents our sentiment analysis across different cities using BERT.Four and five-star ratings are prominent, indicating high satisfaction, particularly in Chicago and Atlanta.One-star ratings, signifying dissatisfaction, are also significant, especially in Detroit and NYC.Contrastingly, two and three-star ratings are scarce across all cities, suggesting users are less inclined to share moderate views.This disparity in sentiment distribution could reflect cultural or regional differences in communication styles or experiences.

Relationship Bewteen Sentiment Level and other Variables

In this section，we explore whether there is a relationship between sentiment level and other variables,like word counts for the thread and numbers of comments.

01 Word counts by sentiment category.

reddit_sentiment %>%
  ggplot(aes(x = bert_label, y = word_count)) +
  geom_jitter(height = 0, width = 0.05, color = "gray") +  # Set dots to gray
  stat_summary(fun = mean, geom = "crossbar", width = 0.4, color = "red") +
  theme_light() +
  labs(title = "Word Count by BERT Label")  # Add a title to the plot

  • Findings: This graph shows the word count in posts, segmented by BERT sentiment labels from one to five stars.Each dot represents a post, with the word count on the vertical axis.The red lines indicate the average word count for each sentiment rating.Notably, the spread of word counts is quite similar across different sentiments, showing no strong correlation between the length of a post and its sentiment.However, we might observe that the highest word counts appear in the 1-star category, suggesting that negative posts can sometimes be more detailed in their feedback.
#data in group:
library(ggplot2)
library(dplyr)

# Filter out the 'Other' group
reddit_sentiment_filtered <- reddit_sentiment %>%
  filter(group != "Other")

# Create the plot
gg <- reddit_sentiment_filtered %>%
  ggplot(aes(x = bert_label, y = word_count, color = group)) +
  geom_jitter(height = 0, width = 0.05, size = 1) +  # Smaller points
  stat_summary(fun = mean, geom = "crossbar", width = 0.4, color = "black") +
  theme_light()  # Light theme

# Print the plot
print(gg)

* Findings: The spread of word counts in the negative sentiment categories (1 and 2 stars) shows a wider range compared to positive sentiments (4 and 5 stars). This could suggest that users are more verbose when expressing dissatisfaction.Cities like Detroit and Chicago have a notable presence in the higher word count range for 1-star ratings, which might imply that users from these cities provide more elaborate negative feedback.Atlanta shows a consistent pattern across all sentiment categories, with most word counts clustering around the median, indicating uniformity in the length of feedback irrespective of sentiment.

02 Mumber of Comments by sentiment category.

reddit_sentiment %>%
  ggplot(aes(x = bert_label, y = num_comments)) +
  geom_jitter(height = 0, width = 0.05, size=0.6, color = "darkgray") +  # Set dots to gray
  stat_summary(fun = mean, geom = "crossbar", width = 0.4,linewidth=0.2, color = "red") +
  theme_light() +
  labs(title = "Numbers of comments by BERT Label")  # Add a title to the plot

  • Findings: The volume of comments associated with each sentiment rating, from one to five stars.Each point represents the number of comments on a thread, categorized by the BERT-determined sentiment of the initial post.The average number of comments for each sentiment category is marked by the red line. Interestingly, the spread of comments is relatively even across all sentiment ratings, suggesting that both positive and negative posts generate engagement.Extreme sentiments, such as 1-star and 5-star ratings, don’t necessarily indicate more discussion than neutral sentiments.

03 cor test

  • Association between a thread’s sentiment and the number of comments on the thread.
cor.test(reddit_sentiment$num_comments, reddit_sentiment$bert_label_numeric, method = "spearman")
## Warning in cor.test.default(reddit_sentiment$num_comments,
## reddit_sentiment$bert_label_numeric, : Cannot compute exact p-value with ties
## 
##  Spearman's rank correlation rho
## 
## data:  reddit_sentiment$num_comments and reddit_sentiment$bert_label_numeric
## S = 687653045, p-value = 0.02127
## alternative hypothesis: true rho is not equal to 0
## sample estimates:
##         rho 
## -0.05805102
cor.test(reddit_sentiment$word_count, reddit_sentiment$bert_label_numeric, method = "spearman")
## Warning in cor.test.default(reddit_sentiment$word_count,
## reddit_sentiment$bert_label_numeric, : Cannot compute exact p-value with ties
## 
##  Spearman's rank correlation rho
## 
## data:  reddit_sentiment$word_count and reddit_sentiment$bert_label_numeric
## S = 684535791, p-value = 0.03463
## alternative hypothesis: true rho is not equal to 0
## sample estimates:
##         rho 
## -0.05325469
  • We applied Spearman’s rank correlation to assess the relationships in our data.First, the number of comments and sentiment ratings: a p-value of about 0.021 indicates a statistically significant but very weak inverse relationship between sentiment ratings and comment volume. Next, for word count and sentiment ratings:Rho is approximately -0.053, with a p-value of roughly 0.035.This also suggests a significant, yet weak, inverse correlation.The negative rho values suggest that as sentiment ratings increase (becoming more positive), both comments and word count tend to decrease slightly.However, the weak correlation means that while there is a trend, it is not strong and other factors likely play a role in the number of comments and the length of posts.

3-2: Word clouds

Using word clouds, we can visualize words that are frequently seen in either positive or negative threads. Using the same code from the previous Rmd file, we tokenize the texts and remove stop words.

data("stop_words")
replace_reg <- "http[s]?://[A-Za-z\\d/\\.]+|&amp;|&lt;|&gt;"

reddit_sentiment_clean <- reddit_sentiment %>% 
  mutate(title = str_replace_all(title, replace_reg, "")) %>%
  unnest_tokens(word, title, token = "words") %>% 
  anti_join(stop_words, by = "word") %>% 
  filter(str_detect(word, "[a-z]")) %>% 
  filter(!word %in% c('greenway'))

We are not interested in words that are commonly seen in both positive and negative threads. We can identify words that are uniquely seen in either positive or negative threads using anti_join.

reddit_sentiment_clean_negative <- reddit_sentiment_clean %>% 
  filter(bert_label_numeric %in% c(1,2))
reddit_sentiment_clean_positive <- reddit_sentiment_clean %>% 
  filter(bert_label_numeric %in% c(4,5))

reddit_sentiment_clean_negative_unique <- reddit_sentiment_clean_negative %>% 
  anti_join(reddit_sentiment_clean_positive, by = 'word')
reddit_sentiment_clean_positive_unique <- reddit_sentiment_clean_positive %>%
  anti_join(reddit_sentiment_clean_negative, by = 'word')
  • Words appearing in negative threads
n <- 20
h <- runif(n, 0, 1) # any color
s <- runif(n, 0.6, 1) # vivid
v <- runif(n, 0.3, 0.7) # neither too dark or bright

df_hsv <- data.frame(h = h, s = s, v = v)
pal <- apply(df_hsv, 1, function(x) hsv(x['h'], x['s'], x['v']))
pal <- c(pal, rep("grey", 10000))

reddit_sentiment_clean_negative_unique %>% 
  count(word, sort = TRUE) %>%
  wordcloud2(color = pal, 
             minRotation = 0, 
             maxRotation = 0, 
             ellipticity = 0.8)
  • Words appearing in positive threads
reddit_sentiment_clean_positive_unique %>% 
  count(word, sort = TRUE) %>%
  wordcloud2(color = pal, 
             minRotation = 0, 
             maxRotation = 0, 
             ellipticity = 0.8)
  • Findings:We’ve categorized the sentiments into negative (1 and 2 stars) and positive (4 and 5 stars) for clearer analysis.For each sentiment group, we created word clouds.In the positive word cloud, we see recurring terms like ‘free,’ ‘design,’ and ‘connect,’ which reflect the aspects users are happy about.The negative word cloud shows terms like ‘cyclist’ and ‘collision’ more often, which may point to the specific issues or negative experiences discussed.

3-3: Temporal analysis

Sentiment by year.

reddit_sentiment %>%
  ggplot(aes(x = date, y = bert_label_numeric)) +
  geom_jitter(width = 0, height = 0.05, color = "darkgray") +  # Set the dots to gray
  scale_x_datetime(date_labels = "%y",
                   breaks = seq(min(reddit_sentiment$date, na.rm = TRUE), 
                                max(reddit_sentiment$date, na.rm = TRUE), 
                                by = "1 year")) +
  theme_light() +  # Use the light theme
  labs(title = "BERT Label (stars) From 2008-2023", 
       x = "Year", 
       y = "BERT Label Numeric") +  # Add a title and axis labels
  theme(
    plot.title = element_text(size = 14, face = "bold", hjust = 0.5),  # Make the title bigger and bold
    axis.title.x = element_text(size = 12),  # Make the x-axis label bigger
    axis.title.y = element_text(size = 12)   # Make the y-axis label bigger
  )

* Findings: The stacked bar chart illustrates sentiment trends from 2008 to 2023, with each color representing a different sentiment rating from one to five stars.There’s an observable increase in sentiment expression over the years, with a particular upsurge in 4-star and 5-star ratings in recent years, suggesting a trend toward more positive feedback.The number of 1-star ratings also shows growth, but not as pronounced as the positive ones, possibly indicating a general trend toward more extreme expressions of sentiment, either positive or negative.The relatively stable but lower count of 2-star and 3-star ratings over time could suggest that users are less inclined to express nuanced or moderate sentiments.

# Assuming the 'year' column is in a four-digit year format
reddit_sentiment <- reddit_sentiment %>%
  mutate(two_digit_year = substr(as.character(year), 3, 4))

# Now plot with the shortened year labels
reddit_sentiment %>%
  ggplot(aes(x = two_digit_year, fill = bert_label)) +
  geom_bar(position = 'stack') +
  scale_x_discrete(breaks = unique(reddit_sentiment$two_digit_year)) +
  scale_fill_brewer(palette = 'PuRd', direction = -1) +
  dark_theme_grey() # Switching to a light theme for better visibility
## Inverted geom defaults of fill and color/colour.
## To change them back, use invert_geom_defaults().

  • Findings: This scatter plot displays the distribution of sentiment ratings from 2008 to 2023.We see a consistent presence of all sentiment ratings throughout the years, indicating a steady usage of the full range of sentiment expression.There is no apparent trend of increasing or decreasing sentiment ratings, suggesting that the overall sentiment has remained stable over time.The scatter of points across each sentiment rating level is quite dense, reflecting the high volume of data and the widespread use of each sentiment category.
library(ggplot2)
library(dplyr)

# Filter out the 'Other' group
reddit_sentiment_filtered <- reddit_sentiment %>%
  filter(group != "Other") %>%
  mutate(two_digit_year = format(as.Date(paste0(year, "-01-01")), "%y")) # Ensure you have a two_digit_year column

# Define custom colors
custom_colors <- c( "#DE2D26", "#FC9272", "yellow","#9ECAE1","#6BAED6")

# Plot with the filtered data
gg <- reddit_sentiment_filtered %>%
  ggplot(aes(x = two_digit_year, fill = bert_label)) +
  geom_bar(position = 'stack') +
  scale_x_discrete(breaks = unique(reddit_sentiment_filtered$two_digit_year)) +
  scale_fill_manual(values = custom_colors) +  # Use custom colors
  facet_wrap(~group) +  # Facet by 'group'
  theme_light() +  # Apply a light theme
  labs(title = "Sentiment by Year", x = "Year", y="%",fill = "BERT Label")

# Print the plot
print(gg)

  • Findings:These four bar charts depict sentiment trends from 2009 to 2023 across Atlanta, NYC, Detroit, and Chicago.In Atlanta, there’s been a noticeable growth in sentiment expression, especially in positive sentiments (4 and 5 stars) in recent years.NYC’s sentiment distribution appears more varied, with an increase in both 1-star and 5-star ratings, suggesting a polarization in user sentiment.Detroit’s chart reveals a modest rise in sentiment expression over time, with a recent uptick in 1-star ratings, possibly reflecting an increase in negative sentiment or specific local issues.Chicago shows a similar pattern to Atlanta, with a significant increase in positive sentiment expression, particularly in 5-star ratings.Across all cities, there’s a trend toward more extreme sentiments (1 star and 5 stars) being expressed, while moderate sentiments (2 and 3 stars) remain less common.The data could indicate an overall increase in engagement on the platform and may reflect broader societal trends or local events influencing public sentiment.

Sentiment by day of week.

# sentiment by day
reddit_sentiment %>% 
  ggplot(aes(x = day_of_week, fill = bert_label)) +
  geom_bar(position = 'fill') +
  scale_fill_brewer(palette = 'PuRd', direction = -1) +
  dark_theme_grey()

* Findings: The data reveals that weekends witness a lower incidence of negative sentiments and a higher occurrence of positive sentiments. Conversely, weekdays see the opposite effect, with more negative sentiments. This suggests that the mood of Reddit threads shifts in a interesting way, aligning positively with leisure time and more negatively with the typical workweek.

Sentiment by time of day.

reddit_sentiment %>% 
  ggplot(aes(x = time, fill = bert_label)) +
  geom_histogram(bins = 24, position = 'stack', color = 'lightblue', lwd = 0.2) +
  scale_x_continuous(breaks = seq(0, 24, by=1)) + 
  scale_fill_manual(values = c('#bc5090', '#bc5090', 'thistle', 'lavender', 'lavender')) + 
  dark_theme_grey()

Findings: The chart displays the distribution of sentiment ratings throughout a 24-hour period.There’s a clear pattern where sentiment expression peaks during certain hours, likely corresponding to high user activity times.9-12am show a significant increase in sentiment expression, particularly for positive sentiments (4 and 5 stars), which could suggest that users are more likely to share positive experiences during leisure hours, especially in time close to noon.1-star ratings, which indicate negative sentiments, are less frequent overall but are still present during the peak hours, possibly reflecting the times when negative experiences are shared or discussed.