How do I count thee? Let me count the ways?

The Mathematics of Taylor Swift

      I confess to being fascinated by Taylor Swift for far more than her music. I think she is an extraordinary person for her philanth...

Showing posts with label text analysis. Show all posts
Showing posts with label text analysis. Show all posts

Thursday, January 30, 2025

The Mathematics of Taylor Swift

      I confess to being fascinated by Taylor Swift for far more than her music. I think she is an extraordinary person for her philanthropy, her speaking out for victims of sexual assault, her advocacy of artists' ownership rights, and her urging her fans to vote.

      But of course there's much more, and so let's look at the mathematics of Taylor Swift.

      Let's start with her net worth. She is estimared to have $1.6 billion in assets. Her largest asset is her music catalog. She did not own the masters (first recording) to her older music, but she re-recorded them and owns the masters to the re-recorded versions and to her future recordings.

      She donates large sums to a variety of organizations. Most of these donations are not public, but here is a sample of some recent donations:

      There are three parts to this mathematical analysis:
         1. her attractiveness,
         2. her lipstick shades, and
         3. her lyrics.
Each part contains a considerable amount of imprecision in measuremenys and judgments. This is more of a work in progress than the final word. Nevertheless, I think it is a non-routine use of math, and it was a lot of fun.

1. How pretty is she?

      You can have your opinion. But mine is based on math.

      The ancient Greeks discovered a particular number called the Golden Ratio, denoted by Greek letter Φ (phi), that has many interesting mathematical properties, apppears in some patterns of nature, and is considered by many to be asthetically pleasing. The Golden Ratio results from finding the point on a line segment that splits the segement into two smaller segments with lengths a and b, such that (a + b)/a = a/b.

      That ratio a/b is the Golden Ratio, Φ. With a little algebra, Φ = (1 + √5)/2 , which is an irrational number so it has an infinite non-repeating decimal, and rounded to three decimal places is 1.618.

      Renaissance artists, plastic surgeons, and makeup artists are among those who use Golden Ratios in various ways with faces to create ideally proportioned faces. Gary Meisner has wriiren extensively on the Golden Ratio, and he believes there are over 20 different ways that the Golden Ratio shows up in human faces and that “the Golden Ratio is also found very commonly in beautiful models of today across all ethnic groups". Biostatistican professor Dr. Kendra Schmid and her colleagues performed various measures of many faces. They began with 17 potential Golden Ratios, and they decided only six of these ratios were predictors of facial attractiveness. See Schmid.

      This takes us to Taylor. I attempted to measure these six ratios on a picture of Taylor. There are many pictures of her, she does enjoy experimenting with different hairstyles, and I had to find one with a hairstyle that gave me the best chance of measuring her from her hairline and also between her ears. The measurement is not exact for many reasons, and because we are using a two-dimensional photo of a three-dimensional object there is certainly some loss of accuracy. Nevertheless, here are the results:

Face length / Face Width1.4722
Mouth width / Interocular distance1.4000
Mouth width / Nose width1.5556
Lips to chin / Interocular1.7000
Lips to chin / Nose width1.8889
Ear length / Nose width1.5556
Average1.5954
% Deviation from Φ - 1.4%

      On average, Taylor's measurements are quite close to phi, the ideal measurement of facial attractiveness.

      (Note that I did the measurements manually, and they are inexact. I understand there are Python libraries such as dlib that can detect facial ladmarks, and this might eliminate some of the imprecision.)

2. What lipstick shade is she wearing?

      A common question on the Internet is what are her favorite lipsticks. I think more interesting mathematically is: given a particular photo of Taylor, can we get the computer to mathematically identify her lipstick?

      First, a little background about image files on the computer. A computer image is a collection of tiny dots called pixels. Each dot is about 1/96 inch. Each pixel contains a color code, as a 6-digit base 16 (hex) number, or as a 6 digit base 10 red-green-blue ordered triple. For example 861A23 in base 16 equals (134, 26, 35) in rgb base 10. The first two hex digits, 86, is the red. 86 in base 16 equals 8*161 + 6*160 = 134 in base ten. The second two hex digits, 1A, is the green, etc.

      We can ask how similar two colors are by plotting them in 3D and calculating their distance.

D = √ [ (134 – 174)2 + (26 – 31)2 + (35 – 61)2 ] = 48.0

      I started with five different photos of Taylor, where it appears to me she is wearing a different shade in each. I tried to crop each photo as a rectangular area as close as I could get to just her lips, and I saved each cropped rectangle as a file. I used R packsge colouR to find the most frequent colors. This was unsatisfactory, possibly because there was too much background color noise. I found a website https://www.remove.bg/ that removes background, and I retried colouR with the background removed files.

      I did the same process with online swatches of ten lipsticks that the Internet says are among Taylor's favorites.

      To my surprise, there were more than 50 different colors, most of them reasonably close, on both the Taylor photos and on the lipstick swatches.

Why are there so many colors?

  • there are 166 = 16,777,216 different computer colors
  • women tell me lipstick never looks same on a person as in a tube
  • my dermatologist offered that lips have lines and grooves that create uneven coloring
  • it is hard to isolate the image of just lips in a rectangle
  • lighting creates distortions
  • there are image resolution and quality issues
      Nevertheless, I decided to do the following. For each of Taylor's ten favorite lipsticks, I found the ten most frequent colors. For each of the five Taylor lip photos, I found their ten most frequent colors. Then for each photo, I calculated the average color distance from the photo to each lipstick. The lipstick with the minimum distance from the photo was then the computer forecast matching a photo to a lipstick. Get it?

      Here are the five Taylor photos with the computer's match:

      In summary, the computer thinks photo 1 has the minimum distance, 320, to its match. Do you agree this is the best match?

PhotoShadeDistance
1Ruby Woo320
2Elson 4625
3Blood Lust388
4Flame392
5Flame399

3. Which of her songs are emotional?

      The R package ‘taylor’ contains lyrics for 11 albums and 240 songs. I will examine TS songs for their emotional content, as follows:

  • % emotional words per song
  • most frequent words per song
  • trend in emotions over time
  • examine theory that 5th song in every album is most emotional
      What makes a song emotional? Some possibilities are: emotional words, volume, tempo. pitch, rhythm, and instrumentation. A few musicians I spoke to have a longer list. But I don't have the data to measure any of these, except emotional words.

      Linguists have created lists of emotional words. Mohammad and Turney created the NRC Emotion Lexicon list of over 14,000 English words and their associations with ten basic emotions (anger, anticipation, disgust, fear, joy, negative, positive, sadness, surprise, and trust). A word may have more than one emotion. For example, faith has the emotions anticipation, joy, positive, and trust.

      Like most datasets, Taylor Swift songs required a considerable amount of data cleaning. In a text analysis like this, data cleaning includes removing punctuation and capitals, deleting stop words (a, oh, the, etc.) that add no value, and fixing informal speech and colloquialisms (dyin’, gonna, etc.). Taylor uses a large number of colloquialisms, often by dropping an ending g in an "ing" word. Also part of the cleaning is that words are stemmed so that love, lovable, loving, and lovely all reduce to "lov"; and then the stemmed words are lemmatized and returned to their root word love. The R packages SnowballC and textstem perform stemming and lemmatizing respectively. However, this combination did not always produce a valid lemmatized word, and I manually adjusted about 500 words which I placed into a file lem_replacements_df.csv. The final result is over 23,000 words from her 240 songs, which I am calling processed words.

      For all 240 songs combined, I calculated Taylor's most frequent words, her most frequent emotional words, her number of words for each of the 10 emotions, and the ratio r = number of emotional words / number of processed) words. The results are contained in the following three plots:

      The above is a baseline for all songs combined. The r value is .364. I asked a friend whose teen-aged daughter is a Swiftie, to suggest a few Taylor songs that she considers especially emotional, and I created the same types of plots for some of these songs. I'll share the plots for the song "My Tears Ricochet", which had a greater than average r value:
      Note that in the above song, the counts are pretty small. Perhaps this is one explanation for why a number of songs that we thought were emotional did not have a higher than average r value.

      Another question I was curious about is whether Taylor's songs have become more emotional over time. I am using original release date of album as the time variable. For songs that Taylor re-recorded, I am using the original release date under the assumption that her re-release may have changed the music but not the lyrics.

      That her songs are becoming more emotional over time appears not to be true. To the contrary, the slope of a poor fitting trend line is negative, but its F-statistic says the linear model is not significant.

      Finally, there is a theory that every fifth song on an album is intentionally more emotional than the others.

      This theory is discussed on the Internet and even Taylor gives it some credence. Albums by release year provide 10 data points (there are 11 albums, but two were released in the same year). In 6 years out of 10 the track 5 r ratio exceeded the all other tracks, and in 4 years out of 10 they did not. A two-sample t-test with one tail concludes that the track 5 mean is not significantly greater than the all other tracks mean, p = .4591. Note that track 5 has an outlier, “All you had to do was stay,” which sounds like it should have a high r, but its r was .134.

4. Bonus: &$!#%

      To the Taylor Tots parents: Our Taylor has been known to use an R-rated word in her songs, or two. Here is the frequency of some of her most frequent such words (and there are others), mostly from her Tortured Poets Department album:

      As I mentioned earlier, each of the three parts of this analysis - her attractiveness, her lipstick shades, and her lyrics - contains various imprecisions and judgments. Perhaps some day I will go back to this project and make some improvements. Comments are welcome. But I think it was a non-routine use of math. And it was fun!

      This project uses three R script files and a number of image files. The R code is shown below, but it is long. Both the R code files and the image files may be found at https://github.com/fcas80/The-Taylor-Swift-Project/tree/main


# file 1 of 3: lovely.txt

library(png)
library(ggplot2)
library(grid)

image_path <- "my_path/ts1_phi_removebg.png"
img <- readPNG(image_path, native=TRUE)   # width 584, height 457
height <- nrow(img)   # 383
width <- ncol(img)    # 331
raster_img <- grid::rasterGrob(img, interpolate = TRUE)
df <- data.frame(xpos = c(0, width), ypos = c(0, height))

# 1. plot photo on minimum grid
ggplot(data = df,
       aes(xpos, ypos)) +
  xlim(0, width) + ylim(0, height) +
  geom_blank() +
  annotation_custom(raster_img, xmin=0, xmax=width, ymin=0, ymax=height) + 
  theme(axis.title.x = element_blank(), 
      axis.title.y = element_blank())

# 2. plot photo on more detailed grid; measurements are manual and imprecise
ggplot(data = df,
       aes(xpos, ypos)) +
  xlim(0, width) + ylim(0, height) +
  geom_blank() +
  annotation_custom(raster_img, xmin=0, xmax=width, ymin=0, ymax=height) +
  geom_hline(yintercept = seq(0, height, by = 10), color = "gray", linwidth = 0.5) +
  geom_vline(xintercept = seq(0, width, by = 10), color = "gray", linwidth = 0.5) +
  scale_x_continuous(breaks = seq(0, width, by = 20)) +
  scale_y_continuous(breaks = seq(0, height, by = 20)) +
  annotate("segment", x = 90, y = 265, xend = 90, yend = 0, color = "red", linwidth = 3) +
  annotate("segment", x = 70, y = 90, xend = 250, yend = 90, color = "red", linwidth = 3) +
  annotate("segment", x = 130, y = 170, xend = 180, yend = 170, color = "red", linwidth = 3) +
  annotate("segment", x = 135, y = 105, xend = 180, yend = 105, color = "red", linwidth = 3) +
  annotate("segment", x = 125, y = 70, xend = 195, yend = 70, color = "blue", linwidth = 3) +
  annotate("segment", x = 160, y = 85, xend = 160, yend = 0, color = "blue", linwidth = 3) +
  annotate("segment", x = 50, y = 160, xend = 50, yend = 90, color = "red", linwidth = 3) +
  theme(axis.title.x = element_blank(),
        axis.title.y = element_blank())

segments_df <- data.frame(
  segment_name = c("face_length", "face_width", "dist_bet_eyes", 
                   "nose_width", "mouth_width", "lips_to_chin", "ear_length"),
  x = c(90, 70, 130, 135, 125, 160, 50),
  y = c(265, 90, 170, 105, 70, 85, 160),
  xend = c(90, 250, 180, 180, 195, 160, 50),
  yend = c(0, 90, 170, 105, 70, 0, 90))
segments_df$dist <- sqrt((segments_df$x - segments_df$xend)^2 + (segments_df$y - segments_df$yend)^2)
segments_df

ratios_df <- data.frame(
  ratio_name = c("face length / width", "mouth width / interocular", "mouth width / nose width", 
       "lips to chin / interocular", "lips to chin / nose width", "ear length / nose width"),
  ratio = rep(0, times=6))

ratios_df$ratio[1] <- round(segments_df$dist[1] / segments_df$dist[2], 4)   # face length / width
ratios_df$ratio[2] <- round(segments_df$dist[5] / segments_df$dist[3], 4)   # mouth width / interocular
ratios_df$ratio[3] <- round(segments_df$dist[5] / segments_df$dist[4], 4)   # mouth width / nose width
ratios_df$ratio[4] <- round(segments_df$dist[6] / segments_df$dist[3], 4)   # lips to chin / interocular
ratios_df$ratio[5] <- round(segments_df$dist[6] / segments_df$dist[4], 4)   # lips to chin / nose width
ratios_df$ratio[6] <- round(segments_df$dist[7] / segments_df$dist[4], 4)   # ear length / nose width 
ratios_df
m <- round(mean(ratios_df$ratio),3)
m
error <- round((m - (1+sqrt(5))/2)/m,3)
error
# END


# file 2 of 3: lipsticks.txt


setwd("my_path")
library(colouR)
library(ggplot2)

# These shades were determined from various Internet articles
shade <- c("Ruby Woo", "Morocco", "Dragon Girl", "Elson 4", "Always Red", "Kyoto Red", 
           "Eadie Scarlet", "Blood Lust", "Flame", "Red to Go")

# These swatches of shades were copied from various Internet stores
df_lipsticks <- data.frame(shade,
        file = c("Mac_Ruby_Woo.png", "NARS_morocco.png", "NARS_Dragon_Girl.png", "PatMcGrath_Elson4.png", 
           "Sephora_always_red.png", "Tatcha_Kyoto_Red.png", "Gucci_Velvet_Eadie_Scarlet.png", 
           "PatMcGrath_Blood_Lust.png", "TomFord_flame.png", "Armani_red_to_go.png" ))

# These files were created by copying various photos of TS, cropping rectangle of lips, then
# using https://www.remove.bg/ to remove backgrounds  
file = c("ts1_lips-removebg-preview.png", "ts2_lips-removebg-preview.png",
       "ts3_lips-removebg-preview.png", "ts4_lips-removebg-preview.png","ts5_lips-removebg-preview.png")
name = c("TS photo 1", "TS photo 2", "TS photo 3", "TS photo 4" ,"TS photo 5")
df_photos <- data.frame(file, name)

############

# p10 for ten lipstick shades, but some only have 1 color
p10 <- function(image, name) {
  # Get primary color
  primary_color <- colouR::getTopCol(image, n = 1, avgCols = FALSE, exclude = TRUE)
  
  top10 <- tryCatch(
    {
      colouR::getTopCol(image, n = 10, avgCols = FALSE, exclude = TRUE)
    },
    error = function(e) {
      primary_color
    }
  )
  
  if (nrow(top10) < 10) {
    top10 <- rbind(top10, top10[rep(1, 10 - nrow(top10)),])
  }
  
  plot <- ggplot(top10, aes(x = hex, y = freq, fill = hex)) +
    geom_bar(stat = 'identity') +
    scale_fill_manual(values = top10$hex) +
    labs(title = paste(name, "Top 10 colors by frequency")) +
    xlab("HEX color code") + ylab("Frequency") +
    theme(
      legend.position = "none",
      plot.title = element_text(size = 15, face = "bold"),
      axis.title = element_text(size = 15, face = "bold"),
      axis.text.x = element_text(angle = 45, hjust = 1, size = 12, face = "bold")
    )
  print(plot)
  return(top10)
}   # close p function

shade_dataframe <- data.frame(matrix(ncol = 10, nrow = 0))

for (i in 1:10) {
  result <- p10(df_lipsticks$file[i], df_lipsticks$shade[i])
  
  # Repeat the first color if there are fewer than 10 colors
  if (nrow(result) < 10) {
    result <- rbind(result, result[rep(1, 10 - nrow(result)),])
  }
  
  shade_dataframe <- rbind(shade_dataframe, t(result$hex))
}

colnames(shade_dataframe) <- paste0("Color", 1:10)
print(shade_dataframe)

#############

p5 <- function(image, name) {
   top10 <- colouR::getTopCol(image, n = 10, avgCols = FALSE, exclude = TRUE)
   return(top10)
}

df_ts_colors <- data.frame()
for(i in 1:nrow(df_photos)){
   top10_colors <- p5(df_photos$file[i], df_photos$name[i])
   top10_colors <- top10_colors$hex
   df_ts_colors <- if(i == 1) {
     df_ts_colors <- top10_colors
    } else {rbind(df_ts_colors, top10_colors)
   }
}
colnames(df_ts_colors) <- c("color1","color2","color3","color4","color5",
                         "color6","color7","color8","color9","color10")
rownames(df_ts_colors) <- df_photos$name
df_ts_colors <- t(df_ts_colors)
print(df_ts_colors)

# begin total RGB distance of a TS photo to centroid of lipstick shade

# sample 3d plot
library(scatterplot3d)
m <- data.frame(
     x = c(174,134),
     y = c(31,26),
     z = c(61,35))

s3d <- scatterplot3d(m$x, m$y, m$z, pch = 19, color = "blue", 
    xlim = c(100, 200), ylim = c(0, 50), zlim = c(0, 100), 
    xlab = expression(bold("R-axis")), ylab = expression(bold("G-axis")), zlab = expression(bold("B-axis")), 
    main = "Plot of two colors in RGB space", cex.main = 2)

s3d$points3d(m$x, m$y, m$z, type = "l", lty = 1, lwd = 2, col = "red")

# Use 'text' for adding labels with bold font
text(s3d$xyz.convert(m$x, m$y, m$z), labels = paste0("(", m$x, ",", m$y, ",", m$z, ")"), 
     pos = 3, cex = 2, col = "#00008B", font = 2)

A <- shade_dataframe$Color1
B <- shade_dataframe$Color2
C <- shade_dataframe$Color3
D <- shade_dataframe$Color4
E <- shade_dataframe$Color5
F <- shade_dataframe$Color6
G <- shade_dataframe$Color7
H <- shade_dataframe$Color8
I <- shade_dataframe$Color9
J <- shade_dataframe$Color10


hex_to_rgb <- function(hex) {
  rgb <- t(col2rgb(hex))
  return(data.frame(r=rgb[1,], g=rgb[2,], b=rgb[3,]))
}

# Function to calculate Euclidean distance between two RGB vectors
euclidean_distance <- function(rgb1, rgb2) {
  sqrt(sum((rgb1 - rgb2)^2))
}

# Function to calculate centroid of a vector of hex colors
calculate_centroid <- function(hex_colors) {
  rgb_values <- hex_to_rgb(hex_colors)
  centroid <- colMeans(rgb_values)
  return(centroid)
}

# Calculate centroids of shades, e.g., centroid_A <- calculate_centroid(A)
  centroids <- lapply(list(A, B, C, D, E, F, G, H, I, J), calculate_centroid)

for (j in 1:nrow(df_photos)) {
   x <- df_ts_colors[,j]
   centroid_x <- calculate_centroid(x)
# Calculate total distance of x to each centroid
   total_distance_to_centroid <- function(x, centroid) {
     x_rgb <- hex_to_rgb(x)
     total <- 0
     for (i in 1:nrow(x_rgb)) {
       total <- total + euclidean_distance(x_rgb[i, ], centroid)
     }
     return(total)
   }

  distances <- sapply(centroids, total_distance_to_centroid, x = x)
  w <- which(distances == min(distances))
  most_similar <- shade[w]
  cat("The lipstick most similar to", df_photos$name[j], "is:", most_similar, 
      "with distance of ", min(distances), "\n")
 }   #  end for loop on photos
  
# END


file 3 of 3:  lyrics.txt

library(taylor)
library(dplyr)
library(tidyverse)
library(tidytext)
data(stop_words)
library(textclean)
library(SnowballC) # For stemming
library(textstem) # For lemmatization

######################################
# functions, preliminaries

# function to print more rows
more_rows <- function(filename){
  options(tibble.print_max = nrow(filename), tibble.print_min = nrow(filename))
  print(filename)
}

# function to fix informal speech (colloquialisms); also contractions, punctuation, lower case
colloq <- function(df) {
  pattern <- "ahh|Ah-uh|ahah|eh|haah|haha|iiii|Oh|oh|Oh-oh|oh-oh|ooh-oh|oohah|Uh|Uh-oh|uh-oh-oh-oh|ya|La-la-la|lala|la-la-la|la-la-la-la|Mm-mm|Mmm-mm|mm-mm|mm-mm-mm-mm|Ha-ha-ha"
  df <- df %>%
    mutate(lyric = str_remove_all(lyric, paste0("\\b(", pattern, ")\\b"))) %>%
    mutate(lyric = str_replace_all(lyric, 
               c("ain't" = "is not", 
                 "Beggin'" = "begging", "beggin'" = "begging", "birth right" = "birthright", "blood-soaked" = "blood soaked",
                 "'bout" = "about", "burnin'" = "burning", "callin'" = "calling", "'Cause" = "because", "'cause" = "because", "Cept" = "Except", "cursee" = "cure",
                 "convers" = "conversation","crashin'" = "crashing", "doin'" = "doing", "driftin'" = "drifting", "dyin'" = "dying", "'em" = "them", "feelin'" = "feeling", "flyin'" = "flying","feverishlying" = "feverishly", "'fore" = "before", "'Fore" = "before", "foreyesgn" = "foreign",
                 "everythi-i-ing" = "everything","fuckin'"="fuck", "gettin'" = "getting", "gonna" = "going to", "gotta" = "got to", "happ'nin'" = "happening", "haven't" = "have not", "Holdin'" = "holding", "hero's" = "hero",
                 "Hopin'" = "hoping", "hopin'" = "hoping","I'ma" = "I am going to", "kinda"="kind of", "king's" = "king", "keepin" = "keeping", "Laughin'" = "laughing", "lookin'" = "looking", "losin" = "losing", "losingg" = "losing", "lovin'" = "loving", "lucki" = "lucky", "no-one" = "no one", "Nothin'" = "nothing", "nothin'" = "nothing","one-hand" = "one hand",
                 "mornin'" = "morning", "'nother"="another", "nothin'" = "nothing", "pickin'" = "picking", "post-mortem" = "postmortem", "prayin'" = "praying", "Prayin'" = "praying", "pretti" = "pretty", "ridin'" = "riding", "Sneakin'" = "sneaking",
                 "outta" = "out of", "'round" = "around", "self-implode" = "self implode", "shoulda" = "should have", "standin'" = "standing", "summer's" = "summer is", "There's" = "there is", "Thinkin'" = "thinking", 
                 "thankin'" = "thanking", "thinkin'" = "thinking", "'Til" = "until", "'til" = "until", "tryin'" = "trying", "tryna" = "trying to", "Tryna" = "trying to","twin-sized" = "twin sized", 
                 "waitin'" = "waiting", "white-knuckle" = "white knuckle",
                 "wanna" = "want to", "weepin'" = "weeping", "whatcha" = "what are you", "Where's" = "where is", "Why'd" = "why did",
                 "wide-eyed" = "wide eyed", "wonderin'" = "wondering", "Wonderin'" = "wondering"))) %>%
    mutate(lyric = map_chr(lyric, clean_text))
  
  return(df)
}


custom_stop_words <- bind_rows(stop_words, 
           tibble(word = c("ah", "dadadada", "ha", "haha", "hey", "hi", "huh", "la", "MMM", "mmm", "mmmm", "mm", "na", "oh", "okay","OK", "ok", "ooh", 
           "uh", "whoa", "woah", "yeah"), lexicon = "custom"))


# Function to clean text  (included in colloq function)
clean_text <- function(text) { 
  text %>% 
    replace_contraction() %>% 
    str_remove_all("[[:punct:]]") %>% 
    str_to_lower()
}

replacements_df <- read.csv("my_path/lem_replacements_df.csv", header = TRUE)

# Function to stem and lemmatize
pro <- function(words_df) {
  words_df <- words_df %>% filter(!is.na(word))
  processed_words <- words_df %>%
    mutate(stemmed_word = wordStem(word),
           lemmatized_word = lemmatize_words(stemmed_word)) %>%
    filter(lemmatized_word != "") %>%  # Filter out empty strings
    left_join(replacements_df, by = c("lemmatized_word" = "stemmed_word")) %>%
    mutate(lemmatized_word = ifelse(is.na(lemmatized_word.y), lemmatized_word, lemmatized_word.y)) %>%
    select(-lemmatized_word.y)  # Remove the extra column created during the join
}

nrc_sentiments <- get_sentiments("nrc") %>%
  filter(word != "count" & word != "heel" & word != "truck" & word != "wear" & word != "word")

common_theme <- theme(
  legend.position="NULL",
  plot.title = element_text(size = 15, face = "bold"),
  plot.subtitle = element_text(size = 12.5, face = "bold"),
  axis.title = element_text(size = 15, face = "bold"),
  axis.text = element_text(size = 15, face = "bold"),
  legend.title = element_text(size = 15, face = "bold"),
  legend.text = element_text(size = 15, face = "bold"))

sentiment_colors <- c("anger" = "red", "anticipation" = "green", "disgust" = "brown",
                      "fear" = "purple", "joy" = "yellow", "negative" = "gray", "positive" = "lightblue",
                      "sadness" = "blue", "surprise" = "pink", "trust" = "turquoise")

emotion_plot <- function(sentiment_summary, song, r){
  ggplot(sentiment_summary, aes(x = sentiment, y = word_count, fill = sentiment)) + 
    geom_col(color = "black") + 
    scale_fill_manual(values = sentiment_colors) +
    labs(title = paste(song, "# words by emotion"),
         subtitle = paste("emotional words / all (processed) words =", r), 
         x = "Emotion", 
         y = "Number of Words") + 
    common_theme + 
    theme(axis.text.x = element_text(angle = 45, hjust=1))
}

lemmatized_colors <- c("red", "blue", "green", "orange", 
                       "purple", "cyan", "magenta", "yellow", "brown", "pink")

word_plot <- function(word_summary, song, title){
  ggplot(word_summary, aes(x = lemmatized_word, y = total_count, fill = lemmatized_word)) + 
    geom_col(color = "black") + 
    scale_fill_manual(values = lemmatized_colors) +
    labs(title = paste(song, title),
         x = "", 
         y = "Frequency") + 
    common_theme +
    theme(axis.text.x = element_text(angle = 45, hjust=1))
}

all_albums <- c("Taylor Swift", "Fearless (Taylor's Version)", "Speak Now (Taylor's Version)", "Red (Taylor's Version)", "1989 (Taylor's Version)",
                "reputation", "Lover", "folklore", "evermore",  "Midnights", "THE TORTURED POETS DEPARTMENT")
number_of_tracks <- c(15, 26, 22, 30, 22, 15, 18, 17, 17, 19, 31)
original_release_year <- c(2006, 2008, 2010, 2012, 2014, 2017, 2019, 2020, 2020, 2022, 2024)
# The years are original release years, i.e., not "Taylor's version" years

#######################################

# Find song by word in title - could be lower case!
x <- "ricochet"
df1 <- taylor_album_songs %>%
   select(album_name, track_name) %>%
   filter(grepl(x, track_name))
n <- df1$album_name[1]
a <- which(all_albums == n)
a

# Find track number
x <- df1$track_name[1]
t <- taylor_album_songs %>%
  filter(album_name == n) %>%
  select(album_name, track_number, track_name) %>%
  filter(track_name == x) %>%
  select(track_number) %>%
  pull() %>% as.numeric()
t

Or instead, choose album a & track t   eg: Style: a=11, t=3
# a <- 8
# t <- 5

#######################################
song <- taylor_album_songs %>% 
  # filter(album_name == all_albums[a] ) %>%  # all songs in an album
  filter(album_name == all_albums[a] & track_number == t) %>% 
  pull(track_name[1])   # NOTE the track_name is always 1
if (length(song) == 0) print("No such track number/song") else print(song)
song <- gsub("\\(Taylor's Version\\)|\\[Taylor's Version\\]", '', song)
album <- all_albums[a]

df <- taylor_album_songs %>%
  filter(album_name == all_albums[a] & track_number == t) %>%
  select(lyrics) %>%
  unnest(lyrics) %>%
  select(line, lyric)
more_rows(df)   # Examine for colloquialisms; add to function colloq

df <- colloq(df)
more_rows(df)

words_df <- df %>%
  unnest_tokens(word, lyric) %>%
  anti_join(custom_stop_words, by = "word")
more_rows(words_df)

processed_words <- pro(words_df) 
more_rows(processed_words) # Examine for words incorrectly lemmatized; add to function pro

processed_words_count <- processed_words %>%
  select(lemmatized_word) %>%
  group_by(lemmatized_word) %>%
  summarize(total_count = n()) %>%
  arrange(desc(total_count)) %>%
  head(10)
more_rows(processed_words_count) 

total_processed_words <- nrow(processed_words) 

sentiment_analysis <- processed_words %>%
  inner_join(nrc_sentiments, by = c("lemmatized_word" = "word"),
             relationship = "many-to-many") %>%
  count(lemmatized_word, sentiment, sort = TRUE)

distinct_sentiment_words <- sentiment_analysis %>% 
  pull(lemmatized_word) %>%
  n_distinct()   

total_sentiment_words <- sentiment_analysis %>%
  select(lemmatized_word, n) %>%
  distinct() %>%
  group_by(lemmatized_word) %>%
  summarize(total_count = sum(n)) %>%
  summarize(total_count = sum(total_count)) %>%
  pull(total_count)   

r <- round(total_sentiment_words / total_processed_words, 3)
cat("emotional words = ", total_sentiment_words, ", total processed words = ", total_processed_words, ", ratio = ", r )

sentiment_summary <- sentiment_analysis %>%
  group_by(sentiment) %>%
  summarise(word_count = sum(n))
sentiment_summary

word_summary <- sentiment_analysis %>%
  select(lemmatized_word, n) %>%
  distinct() %>%
  group_by(lemmatized_word) %>%
  summarize(total_count = sum(n)) %>%
  arrange(desc(total_count)) %>%
  head(10)
word_summary   # check for word with unusual sentiment (truck = trust?); delete from NRC 

word_plot(processed_words_count, song, 
          title="10 most frequent processed words")
word_plot(word_summary, song, 
          title = "10 most frequent emotional words")
emotion_plot(sentiment_summary, song, r)

# End individual song analysis;  for all songs combined, in df <- taylor_album_songs %>%
# delete filter(album_name == all_albums[a] & track_number == t) %>%, and also song <- ""

dirty_words <- processed_words %>%
  select(lemmatized_word) %>%
  filter(, lemmatized_word=="fuck" | lemmatized_word=="shit" | lemmatized_word=="slut" | lemmatized_word=="bitch") %>%
  group_by(lemmatized_word) %>%
  mutate(lemmatized_word = str_replace(lemmatized_word, "fuck", "f**k")) %>%
  mutate(lemmatized_word = str_replace(lemmatized_word, "shit", "sh*t")) %>%
  mutate(lemmatized_word = str_replace(lemmatized_word, "slut", "sl*t")) %>%
  mutate(lemmatized_word = str_replace(lemmatized_word, "bitch", "b*tch")) %>%
  summarize(total_count = n())
dirty_words   # limited to these four; there are others :)

dirty_colors <- c("#FF0000","#000000","#800080","#FFA500")
ggplot(dirty_words, aes(x = lemmatized_word, y = total_count, fill = lemmatized_word)) + 
  geom_col(color = "black") + 
  scale_fill_manual(values = dirty_colors) +
  labs(title = "TS # of R-rated Words",
       x = "", 
       y = "Number of Words") + 
  common_theme 

# END


Sunday, August 25, 2024

Text analysis of 2024 US Presidential convention speeches

Text analysis of 2024 US Presidential convention speeches, by Jerry Tuttle

      Can an actuary / mathematician / data analyst say anything objective and data-oriented about the 2024 US presidential campaign?

      Yes, if I confine my remarks to a numerical text analysis of the candidates' words, rather than attempt to comment on the candidates' political and economic views. This analysis is based solely on data that I collected from the two convention speeches.

      I performed a sentiment analysis of the candidates' emotional words. Here is a graphical summary, discussion to follow:

Sentiment analysis

      I also calculated the candidates' most frequent words, excluding stop words like "a", "and", "the" and so on that don't carry much useful information:

Most frequent words

      I calculated statistics of things like average sentence length and average word length:

Summary statistics

      This blog is my attempt at some objective and data-oriented analysis of the two presidential candidates from their convention sppeches. This is a text analysis of their speeches. Text analysis is a branch of data analysis that takes unstructured text documents, breaks them into sentences and words, and attempts to find some meaning. Text analysis is used by marketers to evaluate customer opinions, by police departments to monitor potential or actual criminal activites, by researchers to evaluate things like whether Shakespeare really wrote all the plays attributed to him and which Beatles songs were mostly written by John versus which by Paul, and by many others.

      This is intended as an objective analysis. I am not trying to make either candidate look good, or bad. I have used these same text analysis techniques in other projects such as analyzing Hamlet, analying short stories, and analyzing the Twitter tweets of a radio talk show host.

      For each of Trump and Harris, I started with a transcript of their convention sppeches. I believe the transcripts are their spoken transcripts, not their written transcripts, based on their very first few sentences. I used various computer packages in R such as tm and tidytext to tokenize the documents into individual sentences, words, and characters. I was guided by the works of Silge and Robinson in Text Mining with R and Williams in Data Science Desktop Survival Guide.

      A summary and a comparison of the of the tokenization of the speeches is the following, repeated from before.

      Obviously Trump spoke for much longer than Harris, so he had many more total words. Harris had a larger average sentence length (18.06 words per sentence versus 10.41), but the two were close in average characters per word and average number of syllables per word. Their numerical Flesch scores were about equal. Flesch score is a measure of the grade level required to understand the text, and we would expect a low grade level in a political speech to the public, in contrast to say a student college paper. Harris had a larger percentage of unique or different words in her speech than Trump (22% to 12%). Trump had a larger percentage of negative words than Harris (43% to 37%), which I will discuss below.

      Sentiment analysis ia analyzing text to determine its emotional tone. Linguistics experts have built dictionaries that associate a large list of words with eight emotions (anger, anticipation, disgust, fear, joy, sadness, surprise, and trust) and also negative and positive sentiments. For example, the word "ache" is associated with sadness. Some words have more than one emotion; for example "admirable" is associated with both joy and trust. Further, some words have either negative or positive associations.

      There are some limitations in sentiment analysis. The sentiment dictionary does not recognize sarcasm, and I am limiting my analysis to single words so I am not picking up negation (such as "not expensive") or other instances where the emotion requires several words. A conclusion from the sentiment distribution graph is that the candidates are surprisingly similar in most of these emotions. The biggest differences are that Trump has a greater portion of his words categorized as anger and negative than Harris has.

Most frequent positive and megative words

      Trump's frequent positive words include "love", adjectives "beautiful" and "wonderful", and also "money". His negative words include "tax", "invasion", "inflation", "war", "illegal", and "crime".

      Harris's frequent positive words include "freedom", "love", "opportunity", and "forward". Her negative words include "tax", "enforcement", "abuse", and "abortion". Interestingly "mother" is both a positive word and a negative word, as is "vote". (I rechecked the sentiment dictionary to confirm these.)

      Trump has a larger percentage of negative words (negative divided by positive plus negative) than Harris (43% to 37%). These positive and negative lists seem consistent with my memory of their speeches.

Most frequent words

      Word frequency provides some clue on what is important to each candidate. Not surprisingly, both candidates frequently used words like "America", "Americans", "country", "nation", and "border". Trump used adjectives like "beautiful" and "incredible". Harris spent considerable time talking about Trump and also about her mother. Harris frequently used "middle", "class", "women", and "law".

Distribution of word sizes

Distribution of syllables per word
      The distributions of size of words and of syllables per word are pretty similar for both candidtaes.

Final thoughts

      It is hard to be indifferent about the 2024 US presidential election. You have your opinion, and I have mine. Much of what the candidates say is their opinion, or their plan if elected, and these things are not things we can verify.

      Some things the candidates say as if they are facts, are stated in a way that is open to interpretation. A good example is that "You are better (or worse) off financially today than four years ago." I can choose one measure, collect some data, and show I am better off; or I can choose a very different measure, collect some data, and show I am worse off.

      Some things the candidates say as facts ARE verifiable. I am in no position to do such verifying, but a number of third parties do this. Here are a few links. I can not vouch for their reliability or bias.

      Text analysis is a popular data analysis technique, and I hope I have added some insight into the candidates in an objective manner.

      The following is my R code:

# Trump convention speech, July 19, 2024
# https://www.nytimes.com/2024/07/19/us/politics/trump-rnc-speech-transcript.html

# Harris convention speech, August 23, 2024
# https://singjupost.com/full-transcript-kamala-harriss-2024-dnc-speech/?singlepage=1

library(tidytext)
library(tm)
library(dplyr)
library(nsyllable)
library(SnowballC)
library(ggplot2)
library(forcats)
library(ggpubr)

speaker <- readline(prompt = "Enter Trump, or enter Harris: ")
if (speaker == "Trump" | speaker == "Harris") print(speaker) else print("Invalid input")
trump_file <- "C:/Users/Jerry/Desktop/Harris_Trump/trump_convention_speech.txt"
harris_file <- "C:/Users/Jerry/Desktop/Harris_Trump/harris_convention_speech.txt"
textfile <- ifelse(speaker=="Trump", trump_file, harris_file)
textfile <- readLines(textfile)
text_df <- data.frame(line = 1:length(textfile), text=textfile)
names(text_df)[2] <- "text"

docs <- Corpus(VectorSource(text_df$text))
docs <- tm_map(docs, content_transformer(tolower))
docs <- tm_map(docs, removeNumbers)
docs <- tm_map(docs, removePunctuation, ucp=TRUE)
docs <- tm_map(docs, stripWhitespace)
inspect(docs[1:8])

custom_colors <- c("#1f77b4", "#ff7f0e", "#2ca02c", "#d62728", "#9467bd",
      "#8c564b", "#e377c2", "#7f7f7f", "#bcbd22", "#17becf",
      "#6a3d9a", "#ff9e1b", "#f6c6c7", "#8dd3c7", "#ffffb3",
      "#bebada", "#fb8072", "#80b1d3", "#fdb462", "#b3e2cd",
      "#ccebc5")

common_theme <- theme(
      legend.position="NULL",
      plot.title = element_text(size=15, face="bold"),
      plot.subtitle = element_text(size=12.5, face="bold"),
      axis.title = element_text(size=15, face="bold"),
      axis.text = element_text(size=15, face="bold"),
      legend.title = element_text(size=15, face="bold"),
      legend.text = element_text(size=15, face="bold"))

docs_df <- data.frame(text = sapply(docs, as.character)) # Convert Corpus to data.frame
wordfile <- unnest_tokens(docs_df, word, text, token = "words")
wordfile %>% count(word, sort=TRUE)
wordcountfile <- mutate(wordfile, numbchars = nchar(word)) # characters per word
long1 <- wordcountfile[which(wordcountfile$numbchars == max(wordcountfile$numbchars)),1] # longest word long2 <- wordcountfile[which(wordcountfile$numbchars == max(wordcountfile$numbchars)),2]
numberchars <- sum(wordcountfile$numbchars)
numberwords <- sum(count(wordcountfile, word, sort=TRUE)$n) # no. words
avgcharperword <- round(numberchars / numberwords, digits=2)
sentencefile <- unnest_tokens(text_df, sentence, text, token = "sentences")
sentencecount <- sum(count(sentencefile, sentence, sort=TRUE)$n)
avgwordpersent <- round(numberwords/sentencecount,2)

wordcountdist <- wordcountfile %>% count(numbchars)
wordcountdist$numbchars <- as.factor(wordcountdist$numbchars)
title <- paste(speaker, "- Distribution of Word Size")
subtitle <- paste("Longest word: ", long1, long2, "characters")
ggplot(wordcountdist, aes(numbchars, n, fill=numbchars)) +
      geom_bar(stat="identity", position = "dodge", width=0.5) +
      labs(title=title, subtitle=subtitle) +
      xlab("number of characters per word") + ylab("") +
      scale_fill_manual(values = custom_colors) +
      theme(legend.position = "none") +
      common_theme

syls <- nsyllable(wordfile$word, language = "en")
syls[which(wordfile$word == "jd")] <- 2 # used because nsyllable generated error here
syls[which(wordfile$word == "nd")] <- 1 # used because nsyllable generated error here
syls[which(wordfile$word == "st")] <- 3 # s/b 21st; used because nsyllable generated error here
syls[which(wordfile$word == "gasolinepowered")] <- 5 # used because nsyllable erred here
long2 <- min(syls[(syls == max(syls, na.rm=TRUE))], na.rm=TRUE)
w <- min(which(syls == long2))
long1 <- wordfile$word[w]
avgsylperword <- round(sum(syls)/numberwords, digits = 2)
avgsylperword
syls <- data.frame(syls) %>% count(syls)
syls$syls <- as.factor(syls$syls)
colnames(syls) <- c("syllables", "n")
title <- paste(speaker, "- Distribution of No. Syllables per Word")
subtitle <- paste("Most syllables: ", long1, long2, "syllables")
ggplot(syls, aes(syllables, n, fill = syllables)) +
      geom_bar(stat="identity", position = "dodge", width=0.5) +
      labs(title=title, subtitle=subtitle) +
      xlab("number of syllables per word") + ylab("") +
      scale_fill_manual(values = custom_colors) +
      theme(legend.position = "none") +
      common_theme

# Flesch-Kincaid reading ease formula
flesch <- round(206.835 - 1.015*(numberwords/sentencecount) - 84.6*(sum(syls$n)/numberwords),2) # Flesch reading ease
flesch
flesch_df <- data.frame(score = c(0,30,50,60,70,80,90,100),
      grade = c("College graduate","College","10th - 12th grade","8th - 9th grade",
      "7th grade","6th grade","5th grade","below 5th grade"))

# Function to find the grade based on score; vlookup
find_grade <- function(score, flesch_df) {
  idx <- findInterval(score, flesch_df$score)
  if (idx == 0) {
   return("below 5th grade") # Handle case where score is below the minimum
  } else {
   return(flesch_df$grade[idx])
  }
}

score_to_find <- flesch
flesch_grade <- find_grade(score_to_find, flesch_df)
flesch_grade

# delete stop words
docs_df <- data.frame(text = sapply(docs, as.character)) # Convert Corpus to data.frame
wordfile <- unnest_tokens(docs_df, word, text, token = "words")
stop_words <- data.frame(tidytext::stop_words) # more words than tm
my_stop_words <- data.frame(word = c("theyre", "hes", "dont",
   "didnt","youre","cant", "im","whats", "weve", "theyve", "youve",
   "couldnt", "wont", "youd"))
wordfile <- anti_join(wordfile, stop_words)
wordfile <- anti_join(wordfile, my_stop_words)

wordfreq <- wordfile
wordfreq <- count(wordfreq, word, sort=TRUE) # word frequency excl stop words
wordfreqdf <- data.frame(wordfreq)

unique_words <- nrow(wordfreq)
portion_unique_words <- round(unique_words / numberwords, digits=2)
wordfreqdf20 <- wordfreqdf[1:21,] # Think about threshold
wordfreqdf20

graphtitle <- paste(speaker, "Word Frequency")
wordfreqdf20$word <- fct_reorder(wordfreqdf20$word, wordfreqdf20$n, .desc = FALSE)
ggplot(data=wordfreqdf20, aes(x=word, y=n, fill=word)) +
      geom_bar(stat="identity", position = "dodge", width=0.5) +
      coord_flip() +
      common_theme +
      xlab("") + ylab("Frequency") +
      ggtitle(graphtitle) +
      scale_fill_manual(values = custom_colors) +
      theme(legend.position = "none")

# sentiments; note mother is both positive and negative!
df1 <- data.frame(wordfile)
colnames(df1) <- "word"
df2 <- get_sentiments("nrc")
df3 <- merge(x=df1, y=df2, by="word", all.x=TRUE, stringsAsFactors=FALSE)
df3 <- subset(df3, !is.na(sentiment))
table(df3$sentiment)
w <- data.frame(table(df3$sentiment))
colnames(w) <- c("sentiment", "n")

sentiment_colors <- c(
   "Anger" = "red",
   "Anticipation" = "green",
   "Disgust" = "brown",
   "Fear" = "purple",
   "Joy" = "yellow",
   "Negative" = "gray",
   "Positive" = "lightblue",
   "Sadness" = "blue",
   "Surprise" = "pink",
   "Trust" = "turquoise")

title <- paste(speaker, "- Sentiment Plot")
ggplot(w, aes(sentiment, n)) +
      geom_bar(stat = "identity", position = "dodge", width = 0.5, fill = sentiment_colors) +
      ggtitle(title) +
      ylab("") +
      common_theme +
      theme(axis.text.x = element_text(angle = 45, hjust=1))

df4 <- df3 %>% filter(sentiment == "positive" | sentiment == "negative")
w <- with(df4, table(sentiment))
neg <- w[1]
pos <- w[2]
neg_ratio <- round(w[1] / (w[1] + w[2]), digits=2)
df5 <- df4 %>% group_by(sentiment) %>% count(word, sort=TRUE)
pos_freq <- df5 %>% filter(sentiment=="positive") %>% top_n(10, wt = n) %>% slice_head(n = 10)
neg_freq <- df5 %>% filter(sentiment=="negative") %>% top_n(10, wt = n) %>% slice_head(n = 10) # ties
pos_freq$word <- fct_reorder(pos_freq$word, pos_freq$n, .desc = FALSE)
neg_freq$word <- fct_reorder(neg_freq$word, neg_freq$n, .desc = FALSE)

title <- paste(speaker, "- Most Frequent Positive and Negative Words")
p1 <- ggplot(pos_freq, aes(word, n)) +
      geom_bar(stat="identity", position = "dodge", width=0.5, fill="darkgreen") +
      ggtitle("Positves") +
      common_theme +
      xlab("") +
      coord_flip()
p2 <- ggplot(neg_freq, aes(word, n)) +
      geom_bar(stat="identity", position = "dodge", width=0.5, fill="red") +
      ggtitle("Negatives") +
      common_theme +
      xlab("") +
      coord_flip()
plot <- ggarrange(p1,p2, ncol=2, nrow=1, legend=NULL)
annotate_figure(plot, top = text_grob(title,
    color = "black", face = "bold", size = 14))

if (speaker == "Trump"){
   t <- data_frame(speaker, numberwords, avgwordpersent, avgcharperword, avgsylperword, flesch, flesch_grade, portion_unique_words, neg_ratio)
   print(t)
   } else {h <- data_frame(speaker, numberwords, avgwordpersent, avgcharperword, avgsylperword, flesch, flesch_grade, portion_unique_words, neg_ratio)
    conclusion <- data.frame(rbind(t,h))
    conclusion <- t(conclusion)
    colnames(conclusion) <- c("Trump", "Harris")
    conclusion <- conclusion[-1,]
    print(conclusion)
}

# the results of stemming and lemmatizing were not used in the report
# stemming
wordfile <- wordfile %>%
    mutate(stem = wordStem(word)) %>%
    count(stem, sort = TRUE)

# lemmatize
df1 <- wordfile # df1 has col named stem
url <- "https://raw.githubusercontent.com/michmech/lemmatization-lists/master/lemmatization-en.txt"
df2 <- read.table(url, header = FALSE, sep = "\t", quote = "", stringsAsFactors = FALSE)
names(df2) <- c("stem", "word")
df3 <- merge(x = df1, y = df2, by = "stem", all.x = TRUE, stringsAsFactors=FALSE)

# if word is not in dictionary, then leave word as is; otherwise, use stemmed word.
df3$word <- ifelse(is.na(df3$word), df3$stem, df3$stem)

# End