Initial commit

nrc_denom <- reviews_final$nrc_positive + reviews_final$nrc_negative

reviews_final$sent_nrc <- ifelse(nrc_denom == 0, 0,

round((reviews_final$nrc_positive - reviews_final$nrc_negative) / nrc_denom, 2))

emotion_cols <- c("anger","anticipation","disgust","fear","joy","sadness","surprise","trust")

for (ec in emotion_cols) reviews_final[[paste0("nrc_", ec)]] <- nrc_mat[[ec]]

# =============================================================================

# 7) Summary stats (needed for report questions)

# =============================================================================

overall_positive_share <- as.numeric(revsummary$total_positive) / as.numeric(revsummary$total_reviews)

avg_sent_manual  <- mean(reviews_final$sent_manual, na.rm = TRUE)

avg_sent_syuzhet <- mean(reviews_final$sent_syuzhet_bing, na.rm = TRUE)

avg_sent_nrc     <- mean(reviews_final$sent_nrc, na.rm = TRUE)

avg_sent_llm     <- mean(reviews_final$sent_llm, na.rm = TRUE)

corr_manual_vs_nrc   <- cor(reviews_final$sent_manual, reviews_final$sent_nrc, use = "complete.obs")

corr_syuzhet_vs_nrc  <- cor(reviews_final$sent_syuzhet_bing, reviews_final$sent_nrc, use = "complete.obs")

corr_llm_vs_manual   <- cor(reviews_final$sent_llm, reviews_final$sent_manual, use = "complete.obs")

# =============================================================================

# 7) Summary stats (needed for report questions)

# =============================================================================

overall_positive_share <- as.numeric(revsummary$total_positive) / as.numeric(revsummary$total_reviews)

avg_sent_manual  <- mean(reviews_final$sent_manual, na.rm = TRUE)

avg_sent_syuzhet <- mean(reviews_final$sent_syuzhet_bing, na.rm = TRUE)

avg_sent_nrc     <- mean(reviews_final$sent_nrc, na.rm = TRUE)

avg_sent_llm     <- mean(reviews_final$sent_llm, na.rm = TRUE)

corr_manual_vs_nrc   <- cor(reviews_final$sent_manual, reviews_final$sent_nrc, use = "complete.obs")

corr_syuzhet_vs_nrc  <- cor(reviews_final$sent_syuzhet_bing, reviews_final$sent_nrc, use = "complete.obs")

corr_llm_vs_manual   <- cor(reviews_final$sent_llm, reviews_final$sent_manual, use = "complete.obs")

if (!"sent_llm" %in% names(reviews_final)) {

reviews_final$sent_llm <- NA_real_

}

overall_positive_share <- as.numeric(revsummary$total_positive) /

as.numeric(revsummary$total_reviews)

avg_sent_manual  <- mean(reviews_final$sent_manual, na.rm = TRUE)

avg_sent_syuzhet <- mean(reviews_final$sent_syuzhet_bing, na.rm = TRUE)

avg_sent_nrc     <- mean(reviews_final$sent_nrc, na.rm = TRUE)

avg_sent_llm     <- mean(reviews_final$sent_llm, na.rm = TRUE)

corr_manual_vs_syuzhet <- cor(reviews_final$sent_manual, reviews_final$sent_syuzhet_bing,use = "complete.obs")

corr_manual_vs_nrc <- cor(reviews_final$sent_manual,

reviews_final$sent_nrc,

use = "complete.obs")

corr_syuzhet_vs_nrc <- cor(reviews_final$sent_syuzhet_bing,

reviews_final$sent_nrc,

use = "complete.obs")

corr_llm_vs_manual <- cor(reviews_final$sent_llm,

reviews_final$sent_manual,

use = "complete.obs")

# -------------------------------

# Helper function: safe correlation

# -------------------------------

# cor() fails if there are no rows where BOTH variables are non-missing.

# This function checks first and returns NA if correlation is impossible.

safe_cor <- function(x, y) {

ok <- complete.cases(x, y)   # rows where both x and y are available

if (sum(ok) < 2) return(NA_real_)  # need at least 2 observations

cor(x[ok], y[ok])

}

# ------------------------------------------------

# Ensure LLM sentiment column exists

# ------------------------------------------------

# The LLM merge may be disabled. To avoid errors in

# later calculations, we explicitly create sent_llm

# and fill it with NA if it does not exist.

if (!"sent_llm" %in% names(reviews_final)) {

reviews_final$sent_llm <- NA_real_

}

# ------------------------------------------------

# Overall positivity share from review summary

# ------------------------------------------------

# This uses the aggregated review statistics (revsummary)

# and is independent of text-based sentiment analysis.

overall_positive_share <- as.numeric(revsummary$total_positive) /

as.numeric(revsummary$total_reviews)

# ------------------------------------------------

# Average sentiment scores by method

# ------------------------------------------------

# na.rm = TRUE ensures missing values do not affect means.

avg_sent_manual  <- mean(reviews_final$sent_manual, na.rm = TRUE)

avg_sent_syuzhet <- mean(reviews_final$sent_syuzhet_bing, na.rm = TRUE)

avg_sent_nrc     <- mean(reviews_final$sent_nrc, na.rm = TRUE)

avg_sent_llm     <- mean(reviews_final$sent_llm, na.rm = TRUE)

# ------------------------------------------------

# Correlations between lexicon-based methods

# ------------------------------------------------

# These correlations are computed using only complete

# observation pairs.

corr_manual_vs_syuzhet <- cor(reviews_final$sent_manual,

reviews_final$sent_syuzhet_bing,

use = "complete.obs")

corr_manual_vs_nrc <- cor(reviews_final$sent_manual,

reviews_final$sent_nrc,

use = "complete.obs")

corr_syuzhet_vs_nrc <- cor(reviews_final$sent_syuzhet_bing,

reviews_final$sent_nrc,

use = "complete.obs")

# ------------------------------------------------

# Correlations involving LLM sentiment

# ------------------------------------------------

# Since LLM sentiment may be missing for all reviews,

# we use the safe_cor() helper to avoid runtime errors.

corr_llm_vs_manual <- safe_cor(reviews_final$sent_llm,

reviews_final$sent_manual)

corr_llm_vs_nrc <- safe_cor(reviews_final$sent_llm,

reviews_final$sent_nrc)

# Initialize result as NA (default if voted_up is missing)

align_votedup <- NA_real_

# Proceed only if the voted_up column exists

if ("voted_up" %in% names(reviews_final)) {

# Convert manual sentiment into a binary prediction:

# TRUE  = positive sentiment

# FALSE = neutral or negative sentiment

pred_pos <- reviews_final$sent_manual > 0

# Compare predicted sentiment sign with actual voted_up value

# mean(TRUE) gives the share of correct alignments

align_votedup <- mean(pred_pos == as.logical(reviews_final$voted_up),

na.rm = TRUE)

}

# Initialize outputs (default if votes_up is missing)

extreme_help_corr <- NA_real_

extreme_help_bins <- NULL

# Proceed only if helpfulness votes are available

if ("votes_up" %in% names(reviews_final)) {

# ------------------------------------------------

# Correlation between sentiment extremeness and votes_up

# ------------------------------------------------

# Extremeness is defined as the absolute value of sentiment,

# regardless of being positive or negative.

extreme_help_corr <- cor(abs(reviews_final$sent_manual),

reviews_final$votes_up,

use = "complete.obs")

# ------------------------------------------------

# Group reviews into low / mid / high extremeness

# ------------------------------------------------

# Quantile-based bins ensure roughly balanced groups.

q <- quantile(abs(reviews_final$sent_manual),

probs = c(0.33, 0.66),

na.rm = TRUE)

bin <- cut(abs(reviews_final$sent_manual),

breaks = c(-Inf, q[1], q[2], Inf),

labels = c("low", "mid", "high"))

# ------------------------------------------------

# Average helpful votes per extremeness group

# ------------------------------------------------

extreme_help_bins <- aggregate(reviews_final$votes_up,

by = list(bin = bin),

FUN = mean,

na.rm = TRUE)

}

monthly_sentiment <- NULL

# Proceed only if review timestamps are available

if ("timestamp_created" %in% names(reviews_final)) {

# ------------------------------------------------

# Create date and month variables

# ------------------------------------------------

reviews_final$day_created <- as.Date(reviews_final$timestamp_created)

reviews_final$month_created <- format(reviews_final$day_created, "%Y-%m")

# ------------------------------------------------

# Compute monthly mean sentiment for each method

# ------------------------------------------------

monthly_sentiment <- aggregate(

reviews_final[, c("sent_manual",

"sent_syuzhet_bing",

"sent_nrc",

"sent_llm")],

by = list(month = reviews_final$month_created),

FUN = mean,

na.rm = TRUE

)

# Ensure chronological order for plotting

monthly_sentiment <- monthly_sentiment[order(monthly_sentiment$month), ]

}

# ------------------------------------------------

# Density plot: Manual vs Syuzhet-Bing vs NRC

# ------------------------------------------------

# Density plots are appropriate because sentiment scores are:

# - continuous

# - bounded between -1 and 1

# - often non-normally distributed

pdf(file.path(GRAPH_DIR, "task5_sentiment_density.pdf"),

width = 9, height = 5)

plot(density(reviews_final$sent_manual, na.rm = TRUE),

main = "Sentiment Density (Manual vs Syuzhet-Bing vs NRC)",

xlab = "Sentiment score",

ylab = "Density",

col = "green")

lines(density(reviews_final$sent_syuzhet_bing, na.rm = TRUE),

col = "red")

lines(density(reviews_final$sent_nrc, na.rm = TRUE),

col = "blue")

legend("topright",

legend = c("Manual (green)", "Syuzhet Bing (red)", "NRC (blue)"),

col = c("green", "red", "blue"),

lty = 1,

bty = "n")

dev.off()

# ------------------------------------------------

# NRC emotion distribution (bar plot)

# ------------------------------------------------

# Aggregate total counts of each emotion across all reviews.

# Negative emotions are shown in grey, positive emotions in yellow.

emotion_totals <- colSums(nrc_mat[, emotion_cols, drop = FALSE],

na.rm = TRUE)

neg_emotions <- c("anger", "disgust", "fear", "sadness")

emo_colors <- ifelse(names(emotion_totals) %in% neg_emotions,

"grey70", "yellow")

pdf(file.path(GRAPH_DIR, "task5_nrc_emotions_bar.pdf"),

width = 9, height = 5)

barplot(emotion_totals,

main = "NRC Emotion Counts (Top 1,000 reviews if limited)",

xlab = "Emotion",

ylab = "Total count",

las = 2,

col = emo_colors)

# ------------------------------------------------

# Monthly sentiment trend plot

# ------------------------------------------------

# Only produced if at least two months of data exist.

if (!is.null(monthly_sentiment) && nrow(monthly_sentiment) >= 2) {

xax <- seq_len(nrow(monthly_sentiment))

pdf(file.path(GRAPH_DIR, "task5_monthly_sentiment.pdf"),

width = 9, height = 5)

plot(xax, monthly_sentiment$sent_manual,

type = "l",

main = "Average Sentiment Over Time (Monthly)",

xlab = "Month",

ylab = "Average sentiment",

xaxt = "n",

col = "green")

axis(1, at = xax,

labels = monthly_sentiment$month,

las = 2,

cex.axis = 0.7)

lines(xax, monthly_sentiment$sent_syuzhet_bing, col = "red")

lines(xax, monthly_sentiment$sent_nrc, col = "blue")

legend("topright",

legend = c("Manual (green)", "Syuzhet Bing (red)", "NRC (blue)"),

col = c("green", "red", "blue"),

lty = 1,

bty = "n")

dev.off()

} else {

message("Monthly sentiment plot skipped: < 2 months of data.")

}

# ------------------------------------------------

# Save review-level data with sentiment scores

# ------------------------------------------------

save(reviews_final,

file = file.path(DATA_DIR,

"reviews_final_with_sentiment.RData"))

# ------------------------------------------------

# Collect all key results into a single list

# ------------------------------------------------

task5.summary <- list(

overall_positive_share = overall_positive_share,

avg_sent_manual = avg_sent_manual,

avg_sent_syuzhet = avg_sent_syuzhet,

avg_sent_nrc = avg_sent_nrc,

avg_sent_llm = avg_sent_llm,

corr_manual_vs_syuzhet = corr_manual_vs_syuzhet,

corr_manual_vs_nrc = corr_manual_vs_nrc,

corr_syuzhet_vs_nrc = corr_syuzhet_vs_nrc,

corr_llm_vs_manual = corr_llm_vs_manual,

corr_llm_vs_nrc = corr_llm_vs_nrc,

align_votedup = align_votedup,

extreme_help_corr = extreme_help_corr,

extreme_help_bins = extreme_help_bins,

monthly_sentiment = monthly_sentiment

)

# ------------------------------------------------

# Save summary object for later use

# ------------------------------------------------

save(task5.summary,

file = file.path(DATA_DIR,

"task5_summary.RData"))

# ------------------------------------------------

# NRC emotion distribution (bar plot)

# ------------------------------------------------

# Aggregate total counts of each emotion across all reviews.

# Negative emotions are shown in grey, positive emotions in yellow.

# Make sure emotion data exists

emotion_totals <- colSums(nrc_mat[, emotion_cols, drop = FALSE],

na.rm = TRUE)

# Safety check: skip plot if all zeros or NA

if (sum(emotion_totals, na.rm = TRUE) > 0) {

neg_emotions <- c("anger", "disgust", "fear", "sadness")

emo_colors <- ifelse(names(emotion_totals) %in% neg_emotions,

"grey70", "gold")

pdf(file.path(GRAPH_DIR, "task5_nrc_emotions_bar.pdf"),

width = 10, height = 6)

barplot(emotion_totals,

main = "NRC Emotion Counts",

xlab = "Emotion",

ylab = "Total count",

las = 2,

col = emo_colors,

cex.names = 0.9)

dev.off()

} else {

message("NRC emotion plot skipped: no emotion counts available.")

}

# ------------------------------------------------

# NRC emotion distribution (bar plot)

# ------------------------------------------------

# Aggregate total counts of each emotion across all reviews.

# Negative emotions are shown in grey, positive emotions in yellow.

# Make sure emotion data exists

emotion_totals <- colSums(nrc_mat[, emotion_cols, drop = FALSE],

na.rm = TRUE)

# Safety check: skip plot if all zeros or NA

if (sum(emotion_totals, na.rm = TRUE) > 0) {

neg_emotions <- c("anger", "disgust", "fear", "sadness")

emo_colors <- ifelse(names(emotion_totals) %in% neg_emotions,

"grey")

pdf(file.path(GRAPH_DIR, "task5_nrc_emotions_bar.pdf"),

width = 10, height = 6)

barplot(emotion_totals,

main = "NRC Emotion Counts",

xlab = "Emotion",

ylab = "Total count",

las = 2,

col = emo_colors,

cex.names = 0.9)

dev.off()

} else {

message("NRC emotion plot skipped: no emotion counts available.")

}

# Safety check: skip plot if all zeros or NA

if (sum(emotion_totals, na.rm = TRUE) > 0) {

neg_emotions <- c("anger", "disgust", "fear", "sadness")

emo_colors <- ifelse(names(emotion_totals) %in% neg_emotions,

"red", "green")

pdf(file.path(GRAPH_DIR, "task5_nrc_emotions_bar.pdf"),

width = 10, height = 6)

barplot(emotion_totals,

main = "NRC Emotion Counts",

xlab = "Emotion",

ylab = "Total count",

las = 2,

col = emo_colors,

cex.names = 0.9)

dev.off()

} else {

message("NRC emotion plot skipped: no emotion counts available.")

}

plot(density(reviews_final$sent_manual, na.rm = TRUE),

main = "Sentiment Density (Manual vs Syuzhet-Bing vs NRC)",

xlab = "Sentiment score",

ylab = "Density",

col = "green")

lines(density(reviews_final$sent_syuzhet_bing, na.rm = TRUE),

col = "red")

lines(density(reviews_final$sent_nrc, na.rm = TRUE),

col = "blue")

legend("topright",

legend = c("Manual (green)", "Syuzhet Bing (red)", "NRC (blue)"),

col = c("green", "red", "blue"),

lty = 1,

bty = "n")

pdf(file.path(GRAPH_DIR, "task5_sentiment_density.pdf"),

width = 10, height = 6)

par(mar = c(5, 5, 4, 2) + 0.1)  # increase left margin

d1 <- density(reviews_final$sent_manual, na.rm = TRUE)

d2 <- density(reviews_final$sent_syuzhet_bing, na.rm = TRUE)

d3 <- density(reviews_final$sent_nrc, na.rm = TRUE)

ymax <- max(d1$y, d2$y, d3$y)

plot(d1,

main = "Sentiment Density (Manual vs Syuzhet-Bing vs NRC)",

xlab = "Sentiment score",

ylab = "Density",

col = "green",

ylim = c(0, ymax * 1.1),

lwd = 2)

lines(d2, col = "red", lwd = 2)

lines(d3, col = "blue", lwd = 2)

legend("topright",

legend = c("Manual", "Syuzhet Bing", "NRC"),

col = c("green", "red", "blue"),

lty = 1,

lwd = 2,

bty = "n")

dev.off()

# =============================================================================

# 11) Graphical comparison of sentiment methods

# This section produces publication-ready plots used in the report.

# =============================================================================

# ------------------------------------------------

# Density plot: Manual vs Syuzhet-Bing vs NRC

# ------------------------------------------------

# Density plots are appropriate because sentiment scores are:

# - continuous

# - bounded between -1 and 1

# - often non-normally distributed

pdf(file.path(GRAPH_DIR, "task5_sentiment_density.pdf"),

width = 10, height = 6)

par(mar = c(5, 5, 4, 2) + 0.1)  # increase left margin

d1 <- density(reviews_final$sent_manual, na.rm = TRUE)

d2 <- density(reviews_final$sent_syuzhet_bing, na.rm = TRUE)

d3 <- density(reviews_final$sent_nrc, na.rm = TRUE)

ymax <- max(d1$y, d2$y, d3$y)

plot(d1,

main = "Sentiment Density (Manual vs Syuzhet-Bing vs NRC)",

xlab = "Sentiment score",

ylab = "Density",

col = "green",

ylim = c(0, ymax * 1.1),

lwd = 2)

lines(d2, col = "red", lwd = 2)

lines(d3, col = "blue", lwd = 2)

legend("topright",

legend = c("Manual", "Syuzhet Bing", "NRC"),

col = c("green", "red", "blue"),

lty = 1,

lwd = 2,

bty = "n")

dev.off()

# ------------------------------------------------

# NRC emotion distribution (bar plot)

# ------------------------------------------------

# Aggregate total counts of each emotion across all reviews.

# Negative emotions are shown in grey, positive emotions in yellow.

# Make sure emotion data exists

emotion_totals <- colSums(nrc_mat[, emotion_cols, drop = FALSE],

na.rm = TRUE)

# Safety check: skip plot if all zeros or NA

if (sum(emotion_totals, na.rm = TRUE) > 0) {

neg_emotions <- c("anger", "disgust", "fear", "sadness")

emo_colors <- ifelse(names(emotion_totals) %in% neg_emotions,

"red", "green")

pdf(file.path(GRAPH_DIR, "task5_nrc_emotions_bar.pdf"),

width = 10, height = 6)

barplot(emotion_totals,

main = "NRC Emotion Counts",

xlab = "Emotion",

ylab = "Total count",

las = 2,

col = emo_colors,

cex.names = 0.9,

ylim =  c(0, 1200))

dev.off()

} else {

message("NRC emotion plot skipped: no emotion counts available.")

}

# ------------------------------------------------

# Monthly sentiment trend plot

# ------------------------------------------------

# Only produced if at least two months of data exist.

if (!is.null(monthly_sentiment) && nrow(monthly_sentiment) >= 2) {

xax <- seq_len(nrow(monthly_sentiment))

pdf(file.path(GRAPH_DIR, "task5_monthly_sentiment.pdf"),

width = 9, height = 5)

plot(xax, monthly_sentiment$sent_manual,

type = "l",

main = "Average Sentiment Over Time (Monthly)",

xlab = "Month",

ylab = "Average sentiment",

xaxt = "n",

col = "green")

axis(1, at = xax,

labels = monthly_sentiment$month,

las = 2,

cex.axis = 0.7)

lines(xax, monthly_sentiment$sent_syuzhet_bing, col = "red")

lines(xax, monthly_sentiment$sent_nrc, col = "blue")

legend("topright",

legend = c("Manual (green)", "Syuzhet Bing (red)", "NRC (blue)"),

col = c("green", "red", "blue"),

lty = 1,

bty = "n")

dev.off()

} else {

message("Monthly sentiment plot skipped: < 2 months of data.")

}

# =============================================================================

# 12) Save processed data and summary outputs

# This section stores all results needed for reproducibility and reporting.

# =============================================================================

# ------------------------------------------------

# Save review-level data with sentiment scores

# ------------------------------------------------

save(reviews_final,

file = file.path(DATA_DIR,

"reviews_final_with_sentiment.RData"))

# ------------------------------------------------

# Save review-level data with sentiment scores

# ------------------------------------------------

save(reviews_final,

file = file.path(DATA_DIR,

"reviews_final_with_sentiment.RData"))

# ------------------------------------------------

# Collect all key results into a single list

# ------------------------------------------------

task5.summary <- list(

overall_positive_share = overall_positive_share,

avg_sent_manual = avg_sent_manual,

avg_sent_syuzhet = avg_sent_syuzhet,

avg_sent_nrc = avg_sent_nrc,

avg_sent_llm = avg_sent_llm,

corr_manual_vs_syuzhet = corr_manual_vs_syuzhet,

corr_manual_vs_nrc = corr_manual_vs_nrc,

corr_syuzhet_vs_nrc = corr_syuzhet_vs_nrc,

corr_llm_vs_manual = corr_llm_vs_manual,

corr_llm_vs_nrc = corr_llm_vs_nrc,

align_votedup = align_votedup,

extreme_help_corr = extreme_help_corr,

extreme_help_bins = extreme_help_bins,

monthly_sentiment = monthly_sentiment

)

# ------------------------------------------------

# Save summary object for later use

# ------------------------------------------------

save(task5.summary,

file = file.path(DATA_DIR,

"task5_summary.RData"))