Download

I first searched the list of subreddits having “PhD” as a keyword. Among 73 subreddits, I chose three subreddits, namely, PhD, AskAcademia, and GradSchool because each of their subreddit name seems to have high relevance to the keyworkd and has a large number of subscribers.

keyword = "PhD"
subreddit_list <- RedditExtractoR::find_subreddits(keywords = keyword)
subreddit_list %>%
  select('subreddit','title','subscribers') %>% 
  arrange(desc(subscribers)) %>%
  head(5) %>%
  knitr::kable()

I downloaded threads from each subreddit.

threads_1 <- find_thread_urls(subreddit = "PhD", 
                              sort_by = 'top', 
                              period = 'year') %>% 
  drop_na()
rownames(threads_1) <- NULL

threads_2 <- find_thread_urls(subreddit = "AskAcademia", 
                              sort_by = 'top', 
                              period = 'year') %>% 
  drop_na()
rownames(threads_2) <- NULL

threads_3 <- find_thread_urls(subreddit = "GradSchool", 
                              sort_by = 'top', 
                              period = 'year') %>% 
  drop_na()
rownames(threads_3) <- NULL

I filtered out threads that do not contain the keyword (considering slight variations in the way it is expressed) within either title or text.

#phd_criterion <- c("PhD", "phd", "Phd", "Ph.D.", "PhD.", "Ph.D", "doctorate", "Doctorate", "Doctoral", #"doctoral")
phd_criterion <- c("PhD|phd|Phd|Ph.D.|PhD.|Ph.D|doctorate|Doctorate|Doctoral|doctoral")

print(nrow(threads_1))
threads_1 <- threads_1 
for (phd in phd_criterion){
  threads_1 <- threads_1 %>%
    filter(!(
      str_detect(title, phd_criterion, negate=TRUE) & str_detect(text, phd_criterion, negate=TRUE)
      )
    )
      }
print(nrow(threads_1))

print(nrow(threads_2))
threads_2 <- threads_2 
for (phd in phd_criterion){
  threads_2 <- threads_2 %>%
    filter(!(
      str_detect(title, phd_criterion, negate=TRUE) & str_detect(text, phd_criterion, negate=TRUE)
      )
    )
      }
print(nrow(threads_2))

print(nrow(threads_3))
threads_3 <- threads_3 
for (phd in phd_criterion){
  threads_3 <- threads_3 %>%
    filter(!(
      str_detect(title, phd_criterion, negate=TRUE) & str_detect(text, phd_criterion, negate=TRUE)
      )
    )
      }
print(nrow(threads_3))

threads <- threads_1 %>%
  bind_rows(threads_2) %>%
  bind_rows(threads_3)
threads

# saving on my local env
#threads %>%
#  write_csv("asset/threads.csv")

Import pre-processed data

# saving on my local env
threads <- read_csv("asset/threads.csv")

## Rows: 1324 Columns: 7
## ── Column specification ────────────────────────────────────────────────────────
## Delimiter: ","
## chr  (4): title, text, subreddit, url
## dbl  (2): timestamp, comments
## date (1): date_utc
## 
## 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.

threads

## # A tibble: 1,324 × 7
##    date_utc    timestamp title                    text  subreddit comments url  
##    <date>          <dbl> <chr>                    <chr> <chr>        <dbl> <chr>
##  1 2024-10-30 1730301090 "It's finally done - di… "I w… PhD             18 http…
##  2 2024-10-12 1728769540 "I often feel like I'm … "I o… PhD             17 http…
##  3 2024-10-02 1727863168 "Am I finished my PhD"    <NA> PhD              5 http…
##  4 2024-02-20 1708388079 "Turned down potentiall… "Lik… PhD             32 http…
##  5 2023-12-16 1702768786 "What\u0019s your field… "I\u… PhD            466 http…
##  6 2023-12-14 1702566888 "Did you become friends… "I a… PhD             94 http…
##  7 2024-08-10 1723270716 "F1 as PhD students =\u…  <NA> PhD              4 http…
##  8 2024-07-31 1722406410 "Failed my PhD Proposal… "Hey… PhD             29 http…
##  9 2024-04-16 1713267474 "Absolutely hate the ex… "I p… PhD             56 http…
## 10 2024-09-09 1725841208 "I feel like the lazies… "I h… PhD             26 http…
## # ℹ 1,314 more rows

#rm(threads_1, threads_2, threads_3)

threads %<>% 
  mutate(date = as.POSIXct(date_utc)) %>%
  filter(!is.na(date))

The dataset covers from Nov. 2023 to Nov. 2024. It seems that the number of threads peaks up in March and September, each of which corresponds to the middle of Fall and Spring semesters.

# number of threads by week
threads %>% 
  ggplot(aes(x = date)) +
  geom_histogram(color="black", position = 'stack', binwidth = 604800) +
  scale_x_datetime(date_labels = "%b %y",
                   breaks = seq(min(threads$date, na.rm = T), 
                                max(threads$date, na.rm = T), 
                                by = "1 month")) +
  theme_minimal()

Tokenization and stop words

I tokenized texts into words prior to semantic analysis, which is the main part of this study.

# Word tokenization
words <- threads %>% 
  tidytext::unnest_tokens(output = word, input = text, token = "words") %>%
  select(date, subreddit, title, comments, word)
words

## # A tibble: 276,736 × 5
##    date                subreddit title                            comments word 
##    <dttm>              <chr>     <chr>                               <dbl> <chr>
##  1 2024-10-30 00:00:00 PhD       It's finally done - dissertatio…       18 i    
##  2 2024-10-30 00:00:00 PhD       It's finally done - dissertatio…       18 want…
##  3 2024-10-30 00:00:00 PhD       It's finally done - dissertatio…       18 to   
##  4 2024-10-30 00:00:00 PhD       It's finally done - dissertatio…       18 share
##  5 2024-10-30 00:00:00 PhD       It's finally done - dissertatio…       18 some 
##  6 2024-10-30 00:00:00 PhD       It's finally done - dissertatio…       18 happy
##  7 2024-10-30 00:00:00 PhD       It's finally done - dissertatio…       18 news 
##  8 2024-10-30 00:00:00 PhD       It's finally done - dissertatio…       18 i    
##  9 2024-10-30 00:00:00 PhD       It's finally done - dissertatio…       18 subm…
## 10 2024-10-30 00:00:00 PhD       It's finally done - dissertatio…       18 my   
## # ℹ 276,726 more rows

words %>%
  count(word, sort = TRUE) %>%
  top_n(20) %>%
  mutate(word = reorder(word, n)) %>%
  ggplot(aes(x = word, y = n)) +
  geom_col() +
  xlab(NULL) +
  coord_flip() +
  labs(x = "words",
       y = "counts",
       title = "Unique wordcounts")

## Selecting by n

I refered to a list of “stop words” to sort them out from the word dictionary.

# load list of stop words - from the tidytext package
data("stop_words")

stop_words %>% head()

## # A tibble: 6 × 2
##   word      lexicon
##   <chr>     <chr>  
## 1 a         SMART  
## 2 a's       SMART  
## 3 able      SMART  
## 4 about     SMART  
## 5 above     SMART  
## 6 according SMART

# Regex that matches URL-type string
replace_reg <- "http[s]?://[A-Za-z\\d/\\.]+|&amp;|&lt;|&gt;"

words_clean <- threads %>% 
  # drop URLs
  mutate(text = str_replace_all(text, replace_reg, "")) %>%
  # Tokenization (word tokens)
  unnest_tokens(word, text, token = "words") %>% 
  # drop stop words
  anti_join(stop_words, by = "word") %>% 
  # drop non-alphabet-only strings
  filter(str_detect(word, "[a-z]"))

# Check the number of rows after removal of the stop words. There should be fewer words now
print(
  glue::glue("Before: {nrow(words)}, After: {nrow(words_clean)}")
)

## Before: 276736, After: 87441

Obviously, the keyword “phd” has the highest frequency. It would be helpful to sort out the keyword before drawing a world cloud.

words_clean %>%
  count(word, sort = TRUE) %>%
  top_n(20, n) %>%
  mutate(word = reorder(word, n)) %>%
  ggplot(aes(x = word, y = n)) +
  geom_col() +
  xlab(NULL) +
  coord_flip() +
  labs(x = "words",
       y = "counts",
       title = "Unique wordcounts")

Extract tri-grams from your text data.

I extracted tri-grams using the unnest_tokens function.

# Get n=3 grams
replace_reg <- "http[s]?://[A-Za-z\\d/\\.]+|&amp;|&lt;|&gt;"

words_trigram <- threads %>%
  mutate(text = str_replace_all(text, replace_reg, "")) %>%
  select(text) %>%
  unnest_tokens(output = paired_words,
                input = text,
                token = "ngrams",
                n = 3)
words_trigram

## # A tibble: 273,675 × 1
##    paired_words            
##    <chr>                   
##  1 i wanted to             
##  2 wanted to share         
##  3 to share some           
##  4 share some happy        
##  5 some happy news         
##  6 happy news i            
##  7 news i submitted        
##  8 i submitted my          
##  9 submitted my doctoral   
## 10 my doctoral dissertation
## # ℹ 273,665 more rows

Remove tri-grams containing stop words or non-alphabetic terms.

#separate the paired words into two columns
words_trigram_stretched <- words_trigram %>%
  separate(paired_words, c("word1", "word2", "word3"), sep = " ")

# filter rows where there are stop words under word 1 column and word 2 column
words_trigram_stretched <- words_trigram_stretched %>%
  # drop stop words
  filter(!word1 %in% stop_words$word &
         !word2 %in% stop_words$word &
         !word3 %in% stop_words$word) %>% 
  # drop non-alphabet-only strings
  filter(str_detect(word1, "[a-z]") &
         str_detect(word2, "[a-z]") &
         str_detect(word3, "[a-z]"))

# Filter out words that are not encoded in ASCII
# To see what's ASCII, google 'ASCII table'
library(stringi)
words_trigram_stretched %<>% 
  filter(stri_enc_isascii(word1) &
         stri_enc_isascii(word2) &
         stri_enc_isascii(word3))

Present the frequency of tri-grams in a table.

# Sort the new tri-gram (n=3) counts:
words_counts <- words_trigram_stretched %>%
  count(word1, word2, word3) %>%
  arrange(desc(n))

head(words_counts, 20) %>% 
  head(10) %>%
  knitr::kable()

Discuss any noteworthy tri-grams you come across.

First, several tri-grams indicate how the participants of the selected subreddits are concerned with career path after graduation. (academic job market, post doc position, tenure track job).

Second, PhD programs in certain fields (social sciences, humanities) are mentioned more often, despite the fact that there are a larger number of engineering PhD students at most of universities in the US. Does this indicate that PhD students in certain fields are more likely to engage in (online) discussions?

Third, several words that indicate which year a student is in, such as “fellow”, “senior”, or “incoming”. I guess the social life among the PhD student commnity is very important.

How does the overall sentiment change over time?

To see how the overall sentiment change over time, I plotted monthly average sentiment scores from Nov. 2023 to Nov. 2024.

threads_senti_4analysis <- threads_clean %>%
  bind_cols(threads_clean_sentiment) %>%
  rename(sentiment = ave_sentiment) %>%
  mutate(sentiment_abs = abs(sentiment),
         sentiment_binary = case_when(sentiment > 0 ~ 'positive',
                                       TRUE ~ 'negative')) %>%
  group_by(sentiment_binary) %>%
  arrange(desc(sentiment_abs))  %>%
  select(date, title, text, sentiment_binary, sentiment_abs)

threads_senti_4analysis %>%
  ungroup() %>%
  select(-text) %>%
  mutate(yearmonth = tsibble::yearmonth(date)) %>%
  group_by(yearmonth, sentiment_binary) %>%
  mutate(idx = str_c(yearmonth, sentiment_binary)) %>%
  mutate(sentiment = ifelse(sentiment_binary=="negative" , sentiment_abs*(-1), sentiment_abs)) %>%
  mutate(yearmonth = tsibble::yearmonth(date)) %>%
  mutate(sentiment_binary = factor(sentiment_binary)) %>%
  ggplot(aes(x=yearmonth, y=sentiment, group=idx, fill=sentiment_binary)) +
  geom_boxplot(position = position_dodge(width=0.75)) +
  tsibble::scale_x_yearmonth(date_breaks = "1 month", date_labels = "%b %Y") +
  theme_minimal() +
  labs(
    title = "Average Sentiment Over Time",
    x = "Year-Month",
    y = "Average Sentiment",
    color = "Sentiment"
  )

There are two interesting points to note in the above graph. First, the negative sentiment scores seem to be smallest in absolute value in the beginning of a semester Jan 2024 and Aug 2024 and to be largest in absolute value in the end of a semester Nov 2023 and Apr 2024. I guess that the stress of coursework, dissertation proposals and defenses may have impact on PhD students’ well being and mental health. Second, the positive and negative sides of sentiment seem to change in opposite directions, shaping wider and narrower gaps in each month. Notably, for each month between Mar 2024 and Nov 2024, the absolute values on both sides increase and decrease simultaneously, showing repeated patterns of polarization in sentiments. This indicates that many PhD students feel differently within the same temporal periods or at least those who write on Reddits exhibit their feelings quite differently.

What are the words used when the writers exhibit polarized feelings?

To dive into how the writers expressed their thoughts when they exhibited polarized feelings, I took the texts written in Apr 2024 as an example.

threads_senti_4analysis_polarized <- threads_senti_4analysis %>%
  ungroup() %>%
  mutate(yearmonth = tsibble::yearmonth(date)) %>%
  filter(as.character(yearmonth) == "2024 Apr") %>%
  select(-yearmonth) %>%
  select(-date)

threads_senti_4analysis_polarized %>%
  group_by(sentiment_binary) %>%
  count()

## # A tibble: 2 × 2
## # Groups:   sentiment_binary [2]
##   sentiment_binary     n
##   <chr>            <int>
## 1 negative            37
## 2 positive            81

I will use the same number of examples for each side of sentiment.

# Stop word removal and tokenization
replace_reg <- "http[s]?://[A-Za-z\\d/\\.]+|&amp;|&lt;|&gt;"
phd_criterion <- c("PhD", "phd", "Phd", "Ph.D.", "PhD.")

threads_senti_4analysis_polarized_clean <- threads_senti_4analysis_polarized %>%
  ungroup() %>%
  mutate(title_text = str_c(title, text, sep = "  ")) %>%
  mutate(title_text = str_replace_all(title_text, replace_reg, "")) %>%
  select(-title) %>%
  select(-text) %>%
  # tokenize
  unnest_tokens(word, title_text, token = "words") %>%
  # remove stop words
  anti_join(stop_words, by = "word") %>%
  filter(str_detect(word, "[a-z]")) %>%
  filter(!word %in% phd_criterion) # You need to replace this with your keyword
threads_senti_4analysis_polarized_clean

## # A tibble: 7,069 × 3
##    sentiment_binary sentiment_abs word      
##    <chr>                    <dbl> <chr>     
##  1 positive                 0.450 absolutely
##  2 positive                 0.450 hate      
##  3 positive                 0.450 experience
##  4 positive                 0.450 posted    
##  5 positive                 0.450 similar   
##  6 positive                 0.450 months    
##  7 positive                 0.450 ago       
##  8 positive                 0.450 student   
##  9 positive                 0.450 london    
## 10 positive                 0.450 housing   
## # ℹ 7,059 more rows

Using the anti_join function, I sorted out words that are commonly used in both sides.

# negative text
threads_senti_4analysis_polarized_clean_negative <- threads_senti_4analysis_polarized_clean %>%
  filter(sentiment_binary == "negative") 
# positive text
threads_senti_4analysis_polarized_clean_positive <- threads_senti_4analysis_polarized_clean %>%
  filter(sentiment_binary == "positive")

# Remove words that are commonly seen in both negative and positive threads
threads_senti_4analysis_polarized_clean_negative <- threads_senti_4analysis_polarized_clean_negative %>%
  anti_join(threads_senti_4analysis_polarized_clean_positive, by = 'word')
threads_senti_4analysis_polarized_clean_positive <- threads_senti_4analysis_polarized_clean_positive %>%
  anti_join(threads_senti_4analysis_polarized_clean_negative, by = 'word')

The following word cloud shows words appearing in negative threads. From this, we can see that some of the writers feel pressure from attaining tenure positions and grants. I guess most of them are junior professors who still joing PhD subreddit groups after graduation. Besides, I see words like “decline” or “fired” which indicate that some of the writers have failed in achieving their goals recently.

# Wordcloud with a custom color palette
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 nor too 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))

threads_senti_4analysis_polarized_clean_negative %>%
  count(word, sort = TRUE) %>%
  wordcloud2(color = pal,
       minRotation = -pi/6,
       maxRotation = -pi/6,
       rotateRatio = 1)

Meanwhile, from the word cloud showing words frequently used in texts with positive sentiment, I notice that the writers usually have good feelings with the core elements that constitute what they do as PhD students, namely research, lab, project, etc. Interestingly, I also see time-related words, such as time, month and day. While I may need to explore each text in more detail, I can draw a rough conclusion that some of the writers find joy in their routine school life.

# Wordcloud with a custom color palette
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 nor too 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))

threads_senti_4analysis_polarized_clean_positive %>%
  count(word, sort = TRUE) %>%
  wordcloud2(color = pal,
       minRotation = pi/6,
       maxRotation = pi/6,
       rotateRatio = 1)