tf-idf-wine-reviews.R

library(tidyverse)
library(ggplot2)

dataset <- read.csv("winemag-data.csv", 
                          header=TRUE, stringsAsFactors=FALSE) 
dataset %>% glimpse(102)

# Quick view and descriptive stats
min(dataset$points)
max(dataset$points)
mean(dataset$points)
sd(dataset$points)
quantile(dataset$points, na.rm = TRUE)

min(dataset$price, na.rm = TRUE)
max(dataset$price, na.rm = TRUE)
mean(dataset$price, na.rm = TRUE)
sd(dataset$price, na.rm = TRUE)
quantile(dataset$price, na.rm = TRUE) 

# Histogram of points
ggplot(dataset, aes(points)) + geom_histogram(stat = "count") 

# Histogram of prices below $100
ggplot(subset(dataset, price <= 100), aes(price)) + 
  geom_histogram(stat = "count") 

# Scatter plot
ggplot(subset(dataset, price <= 100), aes(x=points, y = price)) +
  geom_point() +
  geom_count() + 
  geom_smooth(method='lm') +
  theme(legend.position = "none") +
  facet_wrap(~country)

# Filter by keyword
dataset_filtered <- dataset %>%
  filter(country == "France") 

# Basic descriptive stats of subset
mean(dataset_filtered$points)
mean(dataset_filtered$price, na.rm = TRUE)

# Count most commonly used words in reviews
library(tidytext)

tokenized_comments <- dataset_filtered %>%
  select(description, designation, points, price, province, region_1, variety, winery) %>%
  unnest_tokens(word, description) %>%
  anti_join(stop_words) %>%
  filter(word != "wine") %>% # need to take out the word "wine"
  group_by(province, word) %>%
  tally()

tokenized_comments %>% glimpse()

# Plot top review words by province

tokenized_comments %>%
  group_by(province) %>%
  top_n(15) %>%
  arrange(desc(n)) %>%
  ggplot(aes(x = reorder(word, n), y = n, fill = factor(province))) +
  geom_bar(stat = "identity") +
  theme(legend.position = "none") +
  facet_wrap(~ province, scales = "free") + 
  coord_flip() +
  labs(x = "Top words",
       y = "Frequency",
       title = "Top words used in Wine Enthusiast reviews of French wines by province",
       subtitle = "")

# Plot top 15 tf_idf words

library(stringr)
tf_idf_words <- tokenized_comments %>%
  bind_tf_idf(word, province, n) %>%
  arrange(desc(tf))
tf_idf_words

tf_idf_words %>% 
  filter(!str_detect(word, "^8")) %>% 
  filter(!str_detect(word, "^9")) %>%
  filter(word != "sample") %>%
  filter(word != "barrel") %>%
  filter(word != "bordeaux") %>%
  filter(word != "loire") %>%
  filter(word != "beaujolais") %>%
  filter(word != "champagne") %>%
  filter(word != "provence") %>%
  top_n(25) %>%
  arrange(desc(tf_idf)) %>%
  ggplot(aes(x = reorder(word, -tf_idf), y = tf_idf, fill = province)) +
  geom_col() +
  labs(x = NULL, y = "tf-idf") +
  coord_flip() +
  theme(legend.position = "none") +
  facet_wrap(~ province, scales = "free") +
  labs(x = "Top words",
       y = "TF-IDF score",
       title = "Words characteristic of French wines by province based on 150,000 Wine Enthusiast reviews",
       subtitle = "")

# Map 

library(maps)
map('france', col = 1:10)

francemap <- map_data('france')
francewines <- read.csv("wines-france.csv", 
                        header=TRUE, stringsAsFactors=FALSE) 
francewines %>% glimpse(102)

ggplot() + geom_polygon(data=francemap, 
                        aes(x=long, y=lat, group = group), 
                        color="white", fill="grey92") 
+ theme_void()

counts <- francewines %>% 
  group_by(province) %>% 
  summarise(lat = lat, lon = lon, count = n(),
            mean_price = mean(price, na.rm = TRUE),
            mean_points = mean(points)) 
counts %>% glimpse()

ggplot() + geom_polygon(data=francemap, aes(x=long, y=lat, group = group),color="white", fill="grey92" ) + 
  geom_point(data=counts, aes(x=lon, y=lat, size = mean_points), color="lightblue") + 
  scale_size(name="Average points", range = c(1, 15)) +
  geom_text(data = counts, aes(x = lon, y = lat, label = province), vjust = 0.3, hjust = -0.3, check_overlap = TRUE) +
  theme_void()

ggplot() + geom_polygon(data=francemap, aes(x=long, y=lat, group = group),color="white", fill="grey92" ) + 
  geom_point(data=counts, aes(x=lon, y=lat, size = mean_price), color="firebrick3") + 
  scale_size(name="Average price", range = c(1, 15)) +
  geom_text(data = counts, aes(x = lon, y = lat, label = province), vjust = 0.3, hjust = -0.3, check_overlap = TRUE) +
  theme_void()