Sentiment Analysis on Customer Reviews
Saul Ventura
Last update: 07 April, 2019
1 Introduction
1.1 Overview
This project presents the results of an analytic process composed by a sentiment and correlation analyses. The process was executed on two datasets created using public data from customer reviews and financial performance available on Internet. The analyses are realized using R, a programming language for data analysis, statistical computing and visualization
1.2 Potential benefits
Analysis of customer reviews and star ratings can enable organizations to leverage the following benefits:
- Implementation of marketing strategy adjustments.
- Perform return of investments (ROI) metrics on marketing campaigns.
- Improve product or service quality.
- Improve customer service.
- Monitor and control any crisis management.
- Attract new customers.
- Increase sales revenue.
1.3 High level approach
2 Data Collection
2.1 Library import
Loading libraries for this project life cycle: data collection, cleaning, processing & visualization…
Sys.setenv(JAVA_HOME='C:\\Program Files (x86)\\Java\\jre1.8.0_45') # for 32-bit version
# General-purpose data wrangling
library(tidyverse)
library(xts)
library(tidytext)
library(tidyverse)
library(stringr)
library(tidyr)
library(scales)
library(broom)
library(purrr)
library(widyr)
library(igraph)
library(ggraph)
library(SnowballC)
library(wordcloud)
library(reshape2)
library(ggpubr)
library(zoo)
library(highcharter)
library(ggridges)
library(radarchart)
library(DT)
# Parsing of HTML/XML files
library(rvest)
# String manipulation
library(stringr)
# Verbose regular expressions
library(rebus)
# Eases DateTime manipulation
library(lubridate)
#Frequency Tables
library(summarytools)2.2 Web Scraping
Gathering data using Web scraping…
setwd("C:\\Saul\\Portfolio\\Sentiment Analysis")
#Specifying the url for desired website to be scrapped
url <- 'https://cruiseline.com/ship/oasis-of-the-seas/reviews'
#Reading the html content from Amazon
webpage <- read_html(url)
get_reviews <- function(html){
# url <- 'https://cruiseline.com/ship/adventure-of-the-seas/reviews'
# #Reading the html content from Amazon
# webpage <- read_html(url)
# html <- read_html(url)
reviews <- html %>%
# The relevant tag
html_nodes('.review-content') %>%
html_text() %>%
# Trim additional white space
str_trim() %>%
# Convert the list into a vector
unlist()
length(reviews)
reviews[1:length(reviews)]
}
get_review_dates <- function(html){
review_dates <- html %>%
# The relevant tag
html_nodes('.review-sail-date') %>%
html_text() %>%
# Trim additional white space
str_trim() %>%
# Convert the list into a vector
unlist()
length(review_dates)
review_dates[1:length(review_dates)]
}
get_star_rating <- function(html){
ratings <- html %>%
# The relevant tag
html_nodes('.rating-score ') %>%
html_text() %>%
# Trim additional white space
str_trim() %>%
# Convert the list into a vector
unlist()
length(ratings)
ratings[2:length(ratings)]
}
get_data_table <- function(html, company_name){
# Extract the Basic information from the HTML
reviews <- get_reviews(html)
dates <- get_review_dates(html)
ratings <- get_star_rating(html)
# Combine into a tibble
combined_data <- tibble(rating = ratings,
date = dates,
review = reviews)
# Tag the individual data with the company name
combined_data %>%
mutate(company = company_name) %>%
select(rating, date, review)
print(combined_data)
}
get_data_from_url <- function(url, company_name){
html <- read_html(url)
get_data_table(html, company_name)
}
scrape_write_table <- function(url, company_name){
# Read first page
first_page <- read_html(url)
# Extract the number of pages that have to be queried
latest_page_number <- round(3026/20,0) - 1
#get_last_page(first_page)
# Generate the target URLs
list_of_pages <- str_c(url, '?page=', 1:latest_page_number)
# Apply the extraction and bind the individual results back into one table,
# which is then written as a tsv file into the working directory
list_of_pages %>%
# Apply to all URLs
map(get_data_from_url, company_name) %>%
# Combine the tibbles into one tibble
bind_rows() %>%
# Write a tab-separated file
write_tsv(str_c(company_name,'.tsv'))
}
#scrape_write_table(url, 'RCC')
rcc <- read_tsv('RCC.tsv')
#summary(rcc)Checking the number of customer reviews gathered….3K rows, 3 columns.
As well as the interval of time used to extract the reviews.
dim(rcc); min(rcc$date); max(rcc$date)## [1] 3000 3## [1] "Sail Date: Apr 01, 2010 /"## [1] "Sail Date: Sep 30, 2018 /"rcc_revenues <- read_csv('RCCRevenues.csv')3 Data Preparation
3.1 Preparing features
Transforming values to numeric such as the rating. Formatting date values. Creating quarter rating information.
rcc$rating <- as.numeric(gsub(' out of 5', '', rcc$rating))
rcc$date <- gsub('Sail Date: ', '', rcc$date)
rcc$date <- as.Date(gsub(' /', '', rcc$date), "%b%d,%Y" )
rcc$quarter <- as.yearqtr(as.Date( rcc$date, "%m/%d/%Y"))
# generate quarter data
rcc_quarter <-
rcc %>%
group_by(quarter) %>%
summarise_each(funs(mean), rating)
rcc_quarter$rating <- round(rcc_quarter$rating,1)
rcc_quarter$quarter <- as.character(rcc_quarter$quarter)3.2 Table Overview
Taking a look of the review and revenues tables.
datatable(rcc[(1:20),], filter = 'top', options = list(
pageLength = 5, scrollX = TRUE, scrollY = "300px", autoWidth = TRUE), caption = 'Table 1: Customer reviews')datatable(rcc_revenues[(1:20),], filter = 'top', options = list(
pageLength = 5, scrollX = TRUE, scrollY = "150px", autoWidth = TRUE), caption = 'Table 2: Company Revenues')4 Data Exploration
4.1 Graphical Overview
# checking revenues
ggplot(rcc_revenues, aes(quarter, revenue, group=1)) +
geom_line(aes(y = revenue), colour= "red") +
ylab("Revenues") +
xlab("Quarter") +
ggtitle("Royal Caribbean revenues from the last 6 years") + #+ theme_bw()+
theme(axis.text.x = element_text(angle = 45, hjust = 1))
# adding ID
rcc <- tibble::rowid_to_column(rcc, "ID")
############# per month ###################
rcc %>%
count(Month = round_date(date, "month")) %>%
ggplot(aes(Month, n)) +
geom_line(aes(y = n), colour= "blue") +
labs(title = "The Number of reviews per month", y = "Frequency", x="Month/Year") + #+ theme_bw()+
theme(legend.position="top")
rcc <- rcc %>%
mutate(date = as.POSIXct(date, origin = "2010-01-01"),month = round_date(date, "month"))
qplot(rcc$rating,
geom="histogram",
binwidth = 0.1,
main = "Distribution of star ratings",
xlab = "Star Rating",
ylab = "Frequency",
fill=I("green"),
col=I("blue"),
alpha=I(.2))
qplot(rcc$rating,
geom="histogram",
binwidth = 1,
main = "Distribution of star ratings (binwidth = 1)",
xlab = "Star Rating",
ylab = "Frequency",
fill=I("yellow"),
col=I("red"),
alpha=I(.2))
4.2 N-Grams
#Create tidy text format: Unnested, Unsummarized, -Undesirables, Stop and Short words
review_words <- rcc %>%
select(ID, rating, review, date) %>%
unnest_tokens(word, review) %>%
filter(!word %in% stop_words$word,
str_detect(word, "^[a-z']+$"))
review_words## # A tibble: 107,127 x 4
## ID rating date word
## <int> <dbl> <dttm> <chr>
## 1 1 4.8 2019-03-23 17:00:00 amazing
## 2 1 4.8 2019-03-23 17:00:00 trip
## 3 1 4.8 2019-03-23 17:00:00 families
## 4 2 5 2019-03-23 17:00:00 ship
## 5 2 5 2019-03-23 17:00:00 mighty
## 6 2 5 2019-03-23 17:00:00 lots
## 7 2 5 2019-03-23 17:00:00 activities
## 8 2 5 2019-03-23 17:00:00 bored
## 9 2 5 2019-03-23 17:00:00 nights
## 10 2 5 2019-03-23 17:00:00 food
## # ... with 107,117 more rowsglimpse(review_words)## Observations: 107,127
## Variables: 4
## $ ID <int> 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3...
## $ rating <dbl> 4.8, 4.8, 4.8, 5.0, 5.0, 5.0, 5.0, 5.0, 5.0, 5.0, 5.0, ...
## $ date <dttm> 2019-03-23 17:00:00, 2019-03-23 17:00:00, 2019-03-23 1...
## $ word <chr> "amazing", "trip", "families", "ship", "mighty", "lots"...# unigrams
review_words <- rcc %>%
distinct(review, .keep_all = TRUE) %>%
unnest_tokens(word, review, drop = FALSE) %>%
distinct(ID, word, .keep_all = TRUE) %>%
anti_join(stop_words, by = "word") %>%
filter(str_detect(word, "[^\\d]")) %>%
group_by(word) %>%
mutate(word_total = n()) %>%
ungroup()
word_counts <- review_words %>%
count(word, sort = TRUE)
word_counts %>%
head(20) %>%
mutate(word = reorder(word, n)) %>%
ggplot(aes(word, n)) +
geom_col(fill = "orange") +
scale_y_continuous(labels = comma_format()) +
coord_flip() +
labs(title = "The most common words in customer reviews from 2010 to date",
y = "Frequency", x="Word")
## bigrams
review_bigrams <- rcc %>%
unnest_tokens(bigram, review, token = "ngrams", n = 2)
bigrams_separated <- review_bigrams %>%
separate(bigram, c("word1", "word2"), sep = " ")
bigrams_filtered <- bigrams_separated %>%
filter(!word1 %in% stop_words$word) %>%
filter(!word2 %in% stop_words$word)
bigram_counts <- bigrams_filtered %>%
count(word1, word2, sort = TRUE)
bigrams_united <- bigrams_filtered %>%
unite(bigram, word1, word2, sep = " ")
bigrams_united %>%
count(bigram, sort = TRUE)## # A tibble: 21,386 x 2
## bigram n
## <chr> <int>
## 1 royal caribbean 487
## 2 central park 311
## 3 main dining 175
## 4 cruise ship 125
## 5 cruise line 114
## 6 beautiful ship 108
## 7 highly recommend 108
## 8 specialty restaurants 108
## 9 time dining 98
## 10 customer service 88
## # ... with 21,376 more rowsbigram_tf_idf <- bigrams_united %>%
count(bigram)
bigram_tf_idf <- bigram_tf_idf %>% filter(n>50)
ggplot(aes(x = reorder(bigram, n), y=n), data=bigram_tf_idf) +
geom_col(fill = "skyblue") +
labs(title = "The Most Common bigrams in customer reviews from 2010 to date",
y = "Frequency", x="Word") +
coord_flip()
review_subject <- rcc %>%
unnest_tokens(word, review) %>%
anti_join(stop_words)
my_stopwords <- data_frame(word = c(as.character(1:10),"royal" ))
review_subject <- review_subject %>%
anti_join(my_stopwords)
title_word_pairs <- review_subject %>%
pairwise_count(word, ID, sort = TRUE, upper = FALSE)
set.seed(1234)
title_word_pairs %>%
filter(n >= 200) %>%
graph_from_data_frame() %>%
ggraph(layout = "fr") +
geom_edge_link(aes(edge_alpha = n, edge_width = n), edge_colour = "pink") +
geom_node_point(size = 5) +
geom_node_text(aes(label = name), repel = TRUE,
point.padding = unit(0.2, "lines")) +
ggtitle("Bigrams as network of words in customer reviews from 2010 to date") +
xlab("") +
ylab("")+
theme(axis.text = element_blank(),
axis.ticks = element_blank(),
panel.grid = element_blank())
## end bigrams5 Sentiment Analysis
5.1 Applying AFINN lexicon
sentiment_messages <- review_words %>%
inner_join(get_sentiments("afinn"), by = "word") %>%
group_by(ID) %>%
summarize(sentiment = mean(rating),
words = n()) %>%
ungroup() %>%
filter(words >= 10)
# sentiment_messages %>% arrange(desc(sentiment))Reviewing a sample of the best reviews
rcc[ which(rcc$ID==392), ]$review[1]## [1] "Everything about this ship was over the top. We loved the beautiful venues from a actual live park complete with nightly classical music to the Boardwalk with a real Merry go round. The street partys on the Promenade were so much fun. Our favorite restaurant was the American Icon Grille which was complimentary and so delicious. The service was excellent. The staff were friendly and accommodating. Our room attendant Andrea was a hardworking and kind man who worked nonstop to keep the rooms clean and with fresh towels. I can't say enough about how wonderful this ship and its staff were. Just a relaxing and awesome cruise."#sentiment_messages %>% arrange(sentiment)Reviewing a sample of the worst reviews
rcc[ which(rcc$ID==557), ]$review[1]## [1] "My favorite part of the cruise is the food at the buffet restaurants. The food at the Silk dining room sucked. The chicken was dry and hard. The spaghetti was uncooked and not enough tomato sauce. I would recommend to you that when a customer requested to be on a share table that you should accommodate her. But I would not recommend this to anyone else. I have been on the ship for eight days and for six days I have been asking to put me on tables with other people. That did not happened. Only twice they place me on a share table with a couple, because I complained every night and that was it. There were 4 empty chairs next to me. They did not even attempt to fill the whole table. So I am very upset about that. If I wanted to eat alone I would have stayed home. Also, because I purchased two excursion with another company, they would not help me directions or anything else about the tours. I missed my tour excursion in Cozumel, Mexico because I did not know where to go and No one would assists me. So The glass bottom boat left me. I wasted my time and my money. I spent a fortune at the Vitality Spa for my teeth whitening except my teeth did not get to be white. The sad thing about that is the lady tried to convince that my teeth were white!!!! To satisfy me, you can send me two free tickets and all will be forgiven."AFINN <- sentiments %>%
filter(lexicon == "AFINN") %>%
select(word, afinn_score = score)
#
# AFINN
reviews_sentiment <- review_words %>%
inner_join(AFINN, by = "word") %>%
group_by(ID, rating) %>%
summarize(sentiment = mean(afinn_score))
reviews_sentiment_integer <- subset(reviews_sentiment,reviews_sentiment$rating %in% c(1,2,3,4,5) )
#ggplot(reviews_sentiment, aes(rating, sentiment, group = rating)) +
ggplot(reviews_sentiment_integer, aes(x = rating, y = sentiment, color = factor(rating), group = rating)) +
geom_boxplot() +
geom_hline(yintercept=0, linetype="dashed", color = "red") +
geom_text(aes(4.7, 0.05, label = "Neutral Sentiment", vjust = 0), size = 4, color = "red") +
guides(color = guide_legend(title="Star Rating")) +
labs(y = "Average Sentiment Score", x = "Review Star Rating",
title = "Sentiment of Royal Caribeean Cruise Reviews, by Star Rating",
subtitle = "Using the AFINN lexicon")+theme_bw()
FCA_sentiment <- review_words %>%
inner_join(get_sentiments("nrc"))
# FCA_sentiment%>%head()
FCA_sentiment_integer <- subset(FCA_sentiment,FCA_sentiment$rating %in% c(1,2,3,4,5) )
#############
# Graph Using the AFINN lexicon by PERIOD
#############
review_words %>%
inner_join(get_sentiments("afinn")) %>%
group_by(date) %>%
summarize(average_sentiment = mean(score), words = n()) %>%
#filter(words >= 10) %>%
ggplot(aes(date, average_sentiment)) +
geom_line(aes(y = average_sentiment), colour= "darkgreen") +
geom_hline(color = "red", lty = 2, yintercept = 0) +
labs(y = "Average sentiment score", x = "Year",
title = "Sentiment of Royal Caribeean Cruise Reviews, by period",
subtitle = "Using the AFINN lexicon")+
geom_smooth(method = "loess")
5.2 Applying NRC lexicon
#############
# Graph Using the NRC lexicon by PERIOD
#############
theme_set(theme_bw())
#Alternatively
#FCA_sentiment%>%group_by(page, page_total, sentiment)%>%count()
FCA_sentiment_integer %>%
count(date, sentiment) %>%
filter(sentiment %in% c("positive", "negative",
"joy", "trust","fear","sadness"))%>%
mutate(sentiment = as.factor(sentiment)) %>%
#ggplot(aes(page, n / page_total, fill = sentiment)) +
ggplot(aes(date, n / sum(n), fill = sentiment)) +
geom_area(position = "identity", alpha = 0.5) +
labs(y = "Relative frequency", x = "Year",
title = "Sentiment of Royal Caribeean Cruise Reviews, by period",
subtitle = "Using the NRC lexicon")+theme_bw()+
scale_fill_manual(values=viridis_pal(option = "D")(6))+
scale_y_continuous(labels = scales::percent) 
#############
# Graph as stock type
#############
ldat=review_words %>%
inner_join(get_sentiments("afinn")) %>%
group_by(date) %>%
summarize(average_sentiment = mean(score), words = n())
highchart(type = "stock") %>%
hc_title(text = "Sentiment of Royal Caribeean Cruise Reviews, by period") %>%
hc_subtitle(text = "") %>%
hc_tooltip(valueDecimals = 2) %>%
hc_add_series_times_values(ldat$date, ldat$average_sentiment,
name = "",color="#440154FF")%>%
hc_add_theme(hc_theme_gridlight())%>%
hc_yAxis(title = list(text = "Average sentiment score"),labels = list(format = "{value}"), max = 4,min=-4,plotLines = list(
list(label = list(text = ""),
color = "red",
width = 2,
value = 0))) # sUMMARY USING nrc
FCA_sentiment %>%
count(sentiment, word) %>%
filter(sentiment %in% c("positive", "negative",
"joy", "trust","fear","sadness")) %>%
group_by(sentiment) %>%
top_n(10) %>%
ungroup %>%
mutate(word = reorder(word, n)) %>%
mutate(sentiment = as.factor(sentiment)) %>%
ggplot(aes(word, n, fill = sentiment)) +
geom_bar(alpha = 0.8, show.legend = FALSE,stat = "identity") +
coord_flip() +
scale_y_continuous(expand = c(0,0)) +
facet_wrap(~sentiment, scales = "free") +
labs(y = "Total number of occurrences", x = "",
title = "Sentiment of Royal Caribeean Cruise Reviews",
subtitle = "Using the NRC lexicon")+theme_bw()+
scale_fill_manual(values=viridis_pal(option = "D")(6))
5.3 Applying BING lexicon
##################### bing ###############
reviews <- rcc %>%
filter(!is.na(review)) %>%
select(ID, review) %>%
group_by(row_number()) %>%
ungroup()
tidy_reviews <- reviews %>%
unnest_tokens(word, review)
tidy_reviews <- tidy_reviews %>%
anti_join(stop_words)
bing_word_counts <- tidy_reviews %>%
inner_join(get_sentiments("bing")) %>%
count(word, sentiment, sort = TRUE) %>%
ungroup()
bing_word_counts %>%
group_by(sentiment) %>%
top_n(10) %>%
ungroup() %>%
mutate(word = reorder(word, n)) %>%
ggplot(aes(word, n, fill = sentiment)) +
geom_col(show.legend = FALSE) +
facet_wrap(~sentiment, scales = "free") +
labs(y = "Contribution to sentiment", x = "",
title = "Sentiment of Royal Caribeean Cruise Reviews",
subtitle = "Using the BING lexicon")+theme_bw()+
coord_flip() +
ggtitle('Words that contribute to positive and negative sentiment in the reviews')
##################### end bing ###############5.4 Using WordClouds
###################### wordclouds
print('A Wordcloud containing the top 100 most frequently occurring words')## [1] "A Wordcloud containing the top 100 most frequently occurring words" rcc %>%
unnest_tokens(word, review) %>%
filter(!word %in% stop_words$word) %>%
dplyr::count(word,sort = TRUE) %>%
ungroup() %>%
head(50) %>%
with(wordcloud(word, n, max.words = 100,colors=brewer.pal(8, "Dark2")))
review_words <- rcc %>%
unnest_tokens(word, review) %>% #Break the dialogue into individual words
filter(!nchar(word) < 3) %>%
anti_join(stop_words)
print('A comparison Wordcloud using BING lexicon')## [1] "A comparison Wordcloud using BING lexicon" review_words %>%
inner_join(get_sentiments("bing")) %>%
dplyr::count(word, sentiment, sort = TRUE) %>%
acast(word ~ sentiment, value.var = "n", fill = 0) %>%
comparison.cloud(colors = c("Red", "Steel Blue"),
max.words = 50)
5.5 Using Radar chart
###################### ## radar
review_words <- rcc %>%
unnest_tokens(word, review) %>% #Break the dialogue into individual words
filter(!nchar(word) < 3) %>%
anti_join(stop_words)
SW_nrc <- review_words %>%
inner_join(get_sentiments("nrc"))
print("Overall NRC Sentiment")## [1] "Overall NRC Sentiment" # head(SW_nrc)
#Get the count of words per sentiment per year
sentiment_nrc <- SW_nrc %>%
group_by( sentiment) %>%
dplyr::count( sentiment) %>%
mutate(percent_count=n/sum(n)*100)
radar_chart <- sentiment_nrc %>%
chartJSRadar(showToolTipLabel = TRUE,
main = "Radar Charts of NRC Sentiments")
radar_chart6 Testing Hypotheses
Using the Shapiro function to confirm if the variables are normalized or not.
######################################
## hypothese test
#####################################
# ggscatter(reviews_sentiment, x = "rating", y = "sentiment",
# add = "reg.line", conf.int = TRUE,
# cor.coef = TRUE, cor.method = "pearson",
# xlab = "Rating", ylab = "Sentiment")
shapiro.test(reviews_sentiment$rating)##
## Shapiro-Wilk normality test
##
## data: reviews_sentiment$rating
## W = 0.79447, p-value < 2.2e-16shapiro.test(reviews_sentiment$sentiment)##
## Shapiro-Wilk normality test
##
## data: reviews_sentiment$sentiment
## W = 0.9575, p-value < 2.2e-16shapiro.test(rcc_revenues$revenue)##
## Shapiro-Wilk normality test
##
## data: rcc_revenues$revenue
## W = 0.88637, p-value = 0.01119ggqqplot(reviews_sentiment$rating, ylab = "Rating")
ggqqplot(reviews_sentiment$sentiment, ylab = "Sentiment")
ggqqplot(rcc_revenues$revenue, ylab = "Revenues")
The Shapiro function confirmed that only the revenue variable was normalized. Hence, We will use the Kendall rank correlation coefficient that can evaluate variables which are not necessarily normalized.
res<-cor.test(reviews_sentiment$rating,reviews_sentiment$sentiment, method="kendall")
res##
## Kendall's rank correlation tau
##
## data: reviews_sentiment$rating and reviews_sentiment$sentiment
## z = 15.811, p-value < 2.2e-16
## alternative hypothesis: true tau is not equal to 0
## sample estimates:
## tau
## 0.2270537rcc_rating_revenue <- merge(rcc_quarter, rcc_revenues, by = "quarter")
# head(rcc_quarter)
# head(rcc_revenues)
res<-cor.test(rcc_rating_revenue$rating,rcc_rating_revenue$revenue, method="kendall")
res##
## Kendall's rank correlation tau
##
## data: rcc_rating_revenue$rating and rcc_rating_revenue$revenue
## z = 4.0865, p-value = 4.379e-05
## alternative hypothesis: true tau is not equal to 0
## sample estimates:
## tau
## 0.619293For both cases, the p-value of the tests were less than the significance level alpha = 0.05.
7 Findings
Based on the results of the sentiment analysis, below findings were identified:
There is an important increasing of the number of reviews starting 2014. In contrast, the sentiment score has had an increase but really slightly.
A support to explain the number of reviews as a result of more people taking the service, is the evolution of the revenues. It is clear an increasing of the revenues after 2014.
Although after testing the hypotheses, We can confirm that the Sentiment score is correlated to the Star rating, this relationship is low and can be evidenced by the boxplot chart.
As a result of using two lexicons: NRC and BING, We can mention how important is the food in this kind of service. Customers do not only consider the quality but also the waiting list for the reservations and for a direct access to the restaurants.
The prior point can be also supported by the fact that the most common words and pairs of words within the customer reviews are: “food”, “time”, “dining”, “main dinning”, “specialty restaurants”, “time dining”, “drink package”.