Declaration

By including this statement, we the authors of this work, verify that:

Pre-analysis:

0.1 Upload CSVs

businesses <- read.csv("data files/businesses.csv")
reviews <- read.csv("data files/reviews.csv")
users <- read.csv("data files/users.csv")

0.2 Call Libraries

library(DiagrammeR)
library(knitr)
library(kableExtra)
library(tidyverse)
library(dplyr)
library(ggplot2)
library(maps)

Figure 0.1. Visualise Datasets with DiagrammeR

grViz("
  digraph relationships {
    graph [layout = dot, rankdir = LR]

    #Users box
    users [shape = box, style = filled, color = lightblue, 
    label = 'Users\\n------------------\\nuser_id\\nname\\nreview_count\\naverage_stars\\nmember_since']

    #Reviews box
    reviews [shape = box, style = filled, color = lightgray, 
    label = 'Reviews\\n------------------\\nreview_id\\nuser_id\\nbusiness_id\\nstars\\ndate\\ntext']

    #Business box
    businesses [shape = box, style = filled, color = lightpink, 
    label = 'Businesses\\n------------------\\nbusiness_id\\nname\\ncity\\nstate\\nbusiness avg stars\\nreview_count\\ncategories\\nbusiness_group']

    #Relationships
    users -> reviews [label = 'user_id']
    reviews -> businesses [label = 'business_id']
  }
")

Figure 1 aims to provide a visual representation of the datasets included in the analysis, including variables and relationships. I used AI for assistance with the syntax as I hadn’t used the diagrammeR package before. I created a force directed graph which I had learnt about in Visual Analytics this semester. I created 3 boxes, each of which represented a dataset from the analysis, which included a title and the variable names underneath. I chose pastel colours based on my colour theme for the document. I represented the relationships by arrows and labelled them with the variable name.

Question 1

Analyse review behaviour across individual groups.

1.1 Split, Assign and Merge

#Convert member_since to date format, as it is set as a character string
users$member_since <- as.Date(users$member_since) #

#Create new column of the same size and set all values to NA, this will be used to divide into 3 groups later on
users$member_since_group <- rep(NA, nrow(users)) 

#Give a label to each user off of when they joined
for (i in 1:nrow(users)) {
  if (is.na(users$member_since[i])) {
    next  # Skip to next one if date missing
  } else if (users$member_since[i] < as.Date("2017-01-01")) {
    
    #joined before 2017
    users$member_since_group[i] <- "Veteran" 
  } else if (users$member_since[i] <= as.Date("2022-12-31")) {
    
    #joined between 2017 and 2022
    users$member_since_group[i] <- "Intermediate" 
  } else {
    
    #Joined after 2022
    users$member_since_group[i] <- "New"
  }
}

#Merge users and review datasets by user_id
user_reviews_q1 <- merge(users, reviews, by = "user_id") 

#Remove rows with missing data in the stars column
user_reviews_q1 <- user_reviews_q1[!is.na(user_reviews_q1$stars), ]

For 1.1, we did the following:

  • Convert member_since to date format to make it easier to work with (originally set as character string)

  • Created a new column that would hold group labels for each user

  • Created a loop to give each user a label based on their join date, as follows:

    • Veteran: joined before 2017

    • Intermediate: joined between 2017 and 2022

    • New: joined after 2022

  • Merged the ‘users’ and ‘reviews’ datasets by user_id so that we could efficiently analyse data from both tables at the same time. We needed this to be able to calculate review behaviour.

  • Remove any rows with missing data from the stars column of the new dataframe, since missing data could create false results in the analyse if it was included.

Figure 1.1. Visualise Merge with DiagrammeR

library(DiagrammeR)

grViz("
  digraph merge_user_reviews_q1 {
    graph [layout = dot, rankdir = LR]

    # Users table node
    users [shape = box, style = filled, fillcolor = lightblue, fontcolor = black,
    label = 'users\\n------------------\\nuser_id\\nname\\nreview_count\\naverage_stars\\nmember_since\\nuser_group']

    # Reviews table node
    reviews [shape = box, style = filled, fillcolor = lightgray, fontcolor = black,
    label = 'reviews\\n------------------\\nreview_id\\nuser_id\\nbusiness_id\\nstars\\ndate\\ntext']

    # Merged result node with cleaned data
    user_reviews_q1 [shape = box, style = filled, fillcolor = mediumpurple1, fontcolor = black,
    label = 'user_reviews_q1\\n------------------\\nuser_id\\nname\\nreview_count\\naverage_stars\\nmember_since\\nuser_group\\nreview_id\\nbusiness_id\\nstars\\ndate\\ntext\\n\\n(NA rows in stars removed)']

    # Arrows
    users -> user_reviews_q1 [label = 'merge by user_id']
    reviews -> user_reviews_q1 [label = 'merge by user_id']
  }
")

Figure 1.1 follows the same logic as figure 0.1. However instead of showing all 3 original datasets, we showed the merge between ‘users’ and ‘reviews’ and included the new dataset ‘user_reviews_q1’ which was a result of the merge.

1.2 Findings Statistics for Each Group

#Make a list of the 3 distinct groups within members_since_group. 
member_since_group_list <- c("Veteran", "Intermediate", "New")

#Create empty vectors with length of 3. Later, each one will store the results for each group
num_users <- rep(0, length(member_since_group_list))
avg_stars <- rep(0, length(member_since_group_list))
avg_reviews_per_user <- rep(0, length(member_since_group_list))

#Make a loop to go through each group and assign a value for each summary statistic to the empty vector

for (i in 1:length(member_since_group_list)) {
  
  #Select group one at a time
  group <- member_since_group_list[i]
  
  #Only look at the reviews for the group that is selected
  group_data <- user_reviews_q1[user_reviews_q1$member_since_group == group, ]
  
  #Get the all the users in the group, using 'unique' so users arent repeated
  unique_users <- unique(group_data$user_id)
  
  #Calculate the statistics and store in vectors from earlier
  #how many users are in the group
  num_users[i] <- length(unique_users) 
  
  #whats the average rating of all the users in the group
  avg_stars[i] <- mean(group_data$stars, na.rm = T) 
  
  #how many reviews did each user write on average
  avg_reviews_per_user[i] <- nrow(group_data) / length(unique_users) 
}

For 1.2, we did the following

  • Made a list of each user group to be used in the loop

  • Created empty vectors to store the summary statistics

  • Created a for loop to go through each user group and calculate/store the summary statistics. We made sure to extract only unique users so that there werent duplicates.

  • Summary statistics used include:

    • Number of users in group

    • Average review stars for each group

    • Average reviews per user in each group

1.3 Presenting statistics

#Creating a dataframe to hold results for the user groups

results_table_q1 <- data.frame(
  
  #Group name
  Group = member_since_group_list,
  
  #How many (unique) users in each group
  Num_Users = num_users,
  
  #Avg star rating for each group
  Avg_Review_Stars = round(avg_stars, 2),
  
  #Average number of reviews for each user in each group
  Avg_Reviews_Per_User = round(avg_reviews_per_user, 2)
)

For 1.3 we did the following:

  • Made a data frame to store the statistics for each of the groups, so that its easy to visualise for the next step

  • Rounded the continuous variables to 2 decimal places.

Table 1.1 Results for Question 1

#Put the results from 1.3 into the table in kable

results_table_q1 %>%
  
  kable(
    
    #Make the column names look nicer (capitalise and remvove underscores)
    col.names = c("User Group", "Number of Users", "Average Review Stars", "Average Reviews per User"),
    
    #Put all the text as right align
    align = c("r", "r", "r", "r"),
    
    #?????????????????
    format = "html",
    
    #continous variables as 2dp
    digits = 2
  ) %>%
  
  
  kable_styling(
    
    #table is shorter rather than fit to page 
    full_width = FALSE,
    
    #striped means row colours are alternate, condensed means the padding is reduced so it doesnt take up as much empty space 
    bootstrap_options = c("striped", "condensed")
    
  ) %>%
  
  #Header row is white text, bold text and purple background 
  row_spec(0, bold = TRUE, color = "#ffffff", background = "#6a1b9a") %>%
  
  #other rows are dark grey text
  row_spec(1:nrow(results_table_q1), color = "#2c3e50") %>%
  
  #add a border for all the rows, make the column width the same for all the columns
  column_spec(1:4, border_right = TRUE, border_left = TRUE, width = "6em")
User Group Number of Users Average Review Stars Average Reviews per User
Veteran 6519 2.99 5.59
Intermediate 22472 3.00 5.26
New 8312 3.01 5.42

  • For Figure 3 we used kable to create and format the table. AI was used to help with several aspects of formatting:

    • creating a cconsistent column width (width as column_spec)

    • making the table more compact (full_width = FALSE)

    • reducing empty space in knit document (“condensed” as bootstrap option)

    • alternating row colours (“striped” as bootstrap option)

  • We chose to use a purple and grey colour theme to match the main theme of the document

    • named colums and alligned text on the right

    • continous variables were rounded to 2 decimal places

    • colour theme was maintained

      • header with bold white text and on a purple background

      • the other rows had dark grey text

Figure 1.2 ANOVA

#Clean the data using tidyverse package
user_reviews_q1_clean <- user_reviews_q1 %>%
  
  #remove rows where there is no star rating
  filter(!is.na(stars), !is.na(member_since_group)) %>%
  
  #remove rows that dont belong to a group 
  rename(Group = member_since_group)

#perform ANOVA
anova_model_q1 <- aov(stars ~ Group, data = user_reviews_q1_clean)
summary(anova_model_q1)
##                 Df Sum Sq Mean Sq F value Pr(>F)
## Group            2      6   2.760   1.381  0.251
## Residuals   183049 365792   1.998
#Make boxplot to show distrubution of stars in each group
ggplot(user_reviews_q1_clean, aes(x = Group, y = stars, fill = Group)) +
  geom_boxplot() +
  
  #set colours to match document theme
  scale_fill_manual(values = c(
    "Veteran" = "#6a1b9a",
    "Intermediate" = "#f57c00",
    "New" = "#95a5a6"
  )) +
  
  #set lables and title
  labs(
    title = "Distribution of Review Stars by User Group",
    x = "User Group",
    y = "Review Stars",
    fill = "User Group"
  ) +
  
  #use the same theme as the other charts in the document
  theme_minimal() +
  
  theme(
    
    #make title bold, bigger and purple
    plot.title = element_text(face = "bold", size = 14, color = "#6a1b9a"),
    
    #make labels bigger
    axis.title = element_text(size = 12),
    
    #hide legend becauyse its not necessary
    legend.position = "none"
  )

For Figure 1.2 we created a boxplot based on an ANOVA to test whether there was a statistically significant difference in average review stars between groups based on recency of joining. The code was based off of tutorials from COMP1014 (Thinking about Data) and AI was used to help with the colour theme.


Null hypothesis: Average review stars are the same in each group

Alternate Hypothesis: At least one group has a different average review stars

Conclusion: A p value of 0.251 shows that with a significance level of 0.05, we fail to reject the null hypothesis and can confirm that the average review stars are the same for each group.

The visualisation confirms this finding, as any difference in box height, whisker height and median line are barely visible.

Question 1 Summary

We split the users into 3 groups based on how long they had been using the platform

  • Veteran: Before 2017

  • Intermediate: Between 2017 and 2022

  • New: After 2022

We calculated several summary statistics for each group

  • Numbers of users

  • Average review stars

  • Average reviews per user

Looking at the results of Table 1. we can see the following insights:

  • ‘Intermediate’ has over 3 times as many users as ‘veteran’, and almost 3 times as many users as ‘new’.

  • The average review stars vary by only 0.02 stars, unlikely to be a statistically significant difference.

  • There is more variation in average reviews per user. ’Veteran’s have the most reviews (5.59), ’intermediate’s have the lowest (5.26) and ’New’s have somewhere inbetween the others 2 groups (5.26).

Question 2

2.1 Merge

#Make a new dataframe that merges businesses and reviews by the common variable 'business_id', using functions from tidyverse

#start pipeline with reviews. this will be left table for left join. 
review_summary_q2 <- reviews %>%
  
  #Add businesses table to left table
  left_join(businesses, by = "business_id") %>%
  
  #choosing which variables from businesses to add to reviews
  select(review_id, user_id, business_id, stars, date, state)

For 2.1 we did the following:

  • Used tidyverse functions to perform left join between reviews and businesses

  • we did this so that later on we can use information from both tables that is relevant to the statistics we need to calculate

  • we chose to perform left_join instead of merge because we wanted to base the new table on ‘reviews’ because the reviews are the main focus of question 2.

Figure 2.1 Visualise Merge using DiagrammeR

Figure 2.1 follows the same logic as figure 0.1. However instead of showing all 3 original datasets, we showed the merge between ‘businesses’ and ‘reviews’ and included the new dataset ‘review_summary_q2’ which was a result of the merge.

2.2 Calculate Summary per State

#Use tidyverse library to make table

# Create summary table from the merged review + business data
stateSummary_q2 <- review_summary_q2 %>%
  
  #remove anyhting with na in stars and state
  filter(!is.na(stars) & !is.na(state)) %>%
  
  #group together by state to calculate statistics for each state
  group_by(state) %>%
  
  #summarise for all states
  summarise(
    
    #average amount of stars in each review in each state
    avgStars = mean(stars, na.rm = TRUE),
    
    #numbers of reviews in each state
    numReviews = n(),
    
    #the numbver of unique userrs in each state who did a review
    numUsers = n_distinct(user_id)
    
  ) %>%
  
  #sort so that the highest numbers show first in the table
  arrange(desc(avgStars))

For 2.2 we did the following:

  • used the merged dataset from 2.1

  • removed rows for analysis with missing data in stars and state

  • data grouped by state

  • calcualated summary statistics

    • average review stars for each state

    • number of reviews for each state

    • number of users for each state

  • sorted the table by highest average stars at the top

2.3 Calculate Figures for Whole Group

#range for average review stars
range_avg <- range(stateSummary_q2$avgStars, na.rm = TRUE)
range_diff <- diff(range_avg)

#mean for average review stars
mean_avg <- mean(stateSummary_q2$avgStars, na.rm = TRUE)

#print results
range_avg       #minimum and maximum
## [1] 2.949411 3.042339
range_diff      #difference between min and max
## [1] 0.09292819
mean_avg        #mean
## [1] 2.999585

For 2.3 we calculated additional summary statistics to better understand the distribution of average review stars.

Table 2.1 Results for Question 2

#Put the results from 2.2 into the table in kable

stateSummary_q2 %>%
  
  kable(
    
    #Make the column names look nicer (capitalise and remove underscores)
    col.names = c("State", "Average Review Stars", "Number of Reviews", "Number of Users"),
    
    #Put all the text as right align
    align = c("r", "r", "r", "r"),
    
    #use html output for styling
    format = "html",
    
    #continuous variables as 2dp
    digits = 2
  ) %>%
  
  
  kable_styling(
    
    #table is shorter rather than fit to page 
    full_width = FALSE,
    
    #striped means row colours alternate, condensed means less padding
    bootstrap_options = c("striped", "condensed")
    
  ) %>%
  
  #Header row is white text, bold text and purple background 
  row_spec(0, bold = TRUE, color = "#ffffff", background = "#6a1b9a") %>%
  
  #other rows are dark grey text
  row_spec(1:nrow(stateSummary_q2), color = "#2c3e50") %>%
  
  #add a border for all the rows, make the column width the same for all the columns
  column_spec(1:4, border_right = TRUE, border_left = TRUE, width = "6em")
State Average Review Stars Number of Reviews Number of Users
WI 3.04 3283 3054
WV 3.04 3685 3416
NE 3.04 3202 2992
AL 3.03 3548 3295
ND 3.03 3792 3523
TN 3.03 3335 3105
UT 3.03 3173 2935
VT 3.03 3333 3095
MD 3.03 3351 3133
MS 3.02 3477 3240
SC 3.02 3621 3367
TX 3.02 3642 3390
ID 3.02 3336 3091
DC 3.02 3632 3381
AR 3.02 3484 3250
NJ 3.02 3460 3212
LA 3.01 3503 3256
NY 3.01 3460 3206
ME 3.01 3381 3157
CT 3.01 9197 8014
GA 3.01 3525 3254
IN 3.01 3036 2828
OR 3.00 3552 3300
MI 3.00 3354 3119
KS 3.00 3258 3052
NM 3.00 3659 3396
MN 3.00 3521 3309
CA 2.99 3534 3287
MA 2.99 3300 3068
2.99 5423 4956
PA 2.99 3671 3419
DE 2.99 3401 3185
AK 2.99 3464 3207
AZ 2.99 3605 3363
WA 2.99 3758 3504
NC 2.99 3513 3264
CO 2.99 3464 3217
MT 2.98 3630 3380
IA 2.98 3506 3262
OH 2.98 3608 3369
WY 2.98 3350 3127
IL 2.98 3309 3060
NH 2.98 3341 3109
HI 2.98 3726 3450
VA 2.98 3499 3244
OK 2.97 3760 3478
KY 2.97 3466 3237
SD 2.97 3669 3417
NV 2.96 3099 2887
RI 2.96 3350 3126
FL 2.96 3353 3122
MO 2.95 3736 3504

Refer to table 1 description as the code is the same.

Figure 2.2 Map of Average Stars per State

#Using packages ggplot2 and maps

#Load the relevant map outline
us_states <- map_data("state")

#Make sure that all states names from dataset match the names from the package (AI assistance)
stateSummary_q2$state_name <- tolower(state.name[match(stateSummary_q2$state, state.abb)])

#Left join to combine our data with the map data, making sure not to modify the map data (thats why we used left join)
map_data_q2 <- us_states %>%
  left_join(stateSummary_q2, by = c("region" = "state_name"))

#AI assistance was used for some areas in formatting

#Get the map to be plot by average stars
ggplot(map_data_q2, aes(x = long, y = lat, group = group, fill = avgStars)) +
  
  #draw borders
  geom_polygon(color = "white") +           
  #keep the aspect ratio so that it isnt stretched out once knit
  coord_fixed(1.3) +                        
  #set gradient baseed on average stars, make the states with no data grey
  scale_fill_gradient(
    low = "#fceabb", high = "#f76b1c", na.value = "grey90",
    
    #title for legend
    name = "Avg Stars"
  ) + 
  labs(
    
    #title
    title = "Average Review Stars by State",
  ) +
  
  #choose theme to match document theme
  theme_minimal() +
  theme(
    
    #put legend on right hand side
    legend.position = "right",
    
    #hides the axis latitude/longitude, tick marks and labels
    axis.text = element_blank(),
    axis.ticks = element_blank(),
    axis.title = element_blank(),
    
    #make the title bold and purple to match theme
    plot.title = element_text(color = "#6a1b9a", face = "bold", size = 14),
    
    #make the legend title and labels purple
    legend.title = element_text(color = "#6a1b9a", face = "bold"),
    legend.text = element_text(color = "#6a1b9a")

  )

For Figure 2.2 we used ggplot2 and maps packages to design this map. Knowledge of how to do this comes from units COMP2026 Visual Analytics and MEDI3007 Health Sciences Placement, which is why we chose to go with this method. AI was used to help with colour formatting.

Question 2 Summary

For question 2 we looked at average review star ratings in each US state. Calculations include:

  • Average review stars per state

  • total reviews per state

  • Number of users (unique) per state who made a review

the findings were:

  • Wisconsin had the highest average review stars (3.04) , with 3,283 reviews and 3,054 users.

  • Missouri had the lowest average review stars (2.95), with 3,736 reviews and 3,504 users.

  • The range of average review stars across all states was quite small (2.95 to 3.04).

  • The mean average review stars was 3.00

Question 3

3.1 Merge

#remove reviews where user_id is not available 
new_reviews_q4 <- reviews %>%
  filter(!is.na(user_id) & user_id != "")

#use left join to combine updated reviews with several variables from users. Use user_id as the relationship
new_table_q4 <- left_join(new_reviews_q4, users %>% select(user_id, name, member_since), by = "user_id")

For 3.1 we did the following:

  • Removed reviews where user_id wasn’t available

  • Used left join to combine reviews (with NA removed) users, joining with user_id.

3.2 Summary Statistics

#remove reviews where user_id is not available 
new_reviews_q4 <- reviews %>%
  filter(!is.na(user_id) & user_id != "")

#use left join to combine updated reviews with several variables from users. Use user_id as the relationship
new_table_q4 <- left_join(new_reviews_q4, users %>% select(user_id, name, member_since), by = "user_id")

#group by user id to find number of reviews per user
top_users_q3 <- new_table_q4 %>%
  group_by(user_id) %>%
  summarise(num_reviews = n(), .groups = "drop") %>%
  
  #sort by review count highest to lowest
  arrange(desc(num_reviews)) %>%
  
  #only keep the top 10 users
  slice(1:10)

#check how many users have 14 or more reviews
users_14_or_more <- new_table_q4 %>%
  
  #group by user_id
  group_by(user_id) %>%
  
  #find total number of reviews for each user
  summarise(num_reviews = n(), .groups = "drop") %>%
  
  #only keep users with greater than or equal to 14 reviews
  filter(num_reviews >= 14)

#count how many users have greater than or equal to 14 reviews
n_users_14_or_more <- nrow(users_14_or_more)
print(n_users_14_or_more)
## [1] 16
#calculate the average star rating for each of the top 10 users
avg_rating_top_users <- new_table_q4 %>%
  semi_join(top_users_q3, by = "user_id") %>%
  group_by(user_id) %>%
  summarise(avg_star_rating = mean(stars, na.rm = TRUE), .groups = "drop")

#join top 10 user review counts with their name and average star rating
top_users_final <- top_users_q3 %>%
  left_join(users %>% select(user_id, name), by = "user_id") %>%
  
  #replace missing names with "Unknown"
  mutate(name = ifelse(is.na(name), "Unknown", name)) %>%
  
  #add average star rating from earlier
  left_join(avg_rating_top_users, by = "user_id") %>%
  
  #create a rank column to number them from 1 to 10
  mutate(rank = row_number()) %>%
  
  #select and arrange final columns for the table
  select(rank, user_id, name, num_reviews, avg_star_rating)

For 3.2 we did the following:

  • Removed reviews if they had missing user_ids

  • Used left join to prioritise reviews (keep it unchanged except for new columns)

  • Found top 10 users and calculate average star rating

It looks like there are 16 users with greater than or equal to 14. We will choose the top10 by row order (default) even though its not ideal.

Table 3. 1 Results for Question 3

#Put the top 10 users into a styled table using kable

top_users_final %>%
  
  kable(
    
    #Make the column names look nicer (capitalise and remove underscores)
    col.names = c("User Rank", "User ID", "Name", "Number of Reviews", "Average Star Rating"),
    
    #Put all the text as right align
    align = c("r", "r", "r", "r", "r"),
    
    #use html output for styling
    format = "html",
    
    #continuous variables as 2dp
    digits = 2
  ) %>%
  
  kable_styling(
    
    #table is shorter rather than fit to page 
    full_width = FALSE,
    
    #striped means row colours alternate, condensed means less padding
    bootstrap_options = c("striped", "hover", "condensed", "responsive")
    
  ) %>%
  
  #Header row is white text, bold text and purple background 
  row_spec(0, bold = TRUE, color = "#ffffff", background = "#6a1b9a") %>%
  
  #other rows are dark grey text
  row_spec(1:nrow(top_users_final), color = "#2c3e50") %>%
  
  #add a border for all the rows, make the column width the same for all the columns
  column_spec(1:5, border_right = TRUE, border_left = TRUE, width = "6em")
User Rank User ID Name Number of Reviews Average Star Rating
1 u_27070 Rebecca 18 2.83
2 u_11551 Christopher 15 3.27
3 u_6766 Tracy 15 3.27
4 u_11229 Benjamin 14 3.07
5 u_14899 Jason 14 2.57
6 u_17629 Andrew 14 2.21
7 u_22933 Stephanie 14 2.93
8 u_23971 Unknown 14 2.36
9 u_27907 Jesse 14 3.43
10 u_29224 Rebecca 14 3.50

Refer to table 1 description as the code is the same.

Figure 3.1 Boxplot of Top Users

#filter the original review table to only keep reviews from the top 10 users
top_user_reviews_named <- new_table_q4 %>%
  
  #keep only rows where user_id is in the top_users_q3 list
  semi_join(top_users_q3, by = "user_id") %>%
  
  #replace any missing names with "Unknown"
  mutate(name = ifelse(is.na(name), "Unknown", name))

#plot a boxplot showing the spread of star ratings for each top user
ggplot(top_user_reviews_named, aes(x = reorder(name, -stars), y = stars)) +
  
  #draw the boxplots with custom colours
  geom_boxplot(fill = "#d1c4e9", color = "#6a1b9a") +
  
  #add title and axis labels
  labs(
    title = "Rating Distribution of Top 10 Users",
    x = "User",
    y = "Stars"
  ) +
  
  #use a clean minimal theme
  theme_minimal() +
  
  #AI was utilised to help with elements of formatting beyond this point
  #customise colours, fonts and layout
  theme(
    plot.background = element_rect(fill = "#f4f1ea", color = NA),       #background behind plot
    panel.background = element_rect(fill = "#f4f1ea", color = NA),      #background behind axes
    panel.grid.major = element_line(color = "#e0e0e0"),                 #gridlines
    axis.text = element_text(color = "#2c3e50", size = 10),             #axis numbers
    axis.title = element_text(color = "#6a1b9a", size = 12),            #axis titles
    plot.title = element_text(color = "#6a1b9a", size = 14, face = "bold"), #plot title
    axis.text.x = element_text(angle = 45, hjust = 1)                   #rotate user names
  )

Findings from figure 3.1 are discussed in Question 3 Summary

3.3 Additional Findings for Summary

# Calculate the average rating for all users (not just top 10)
overall_user_avg <- new_table_q4 %>%
  group_by(user_id) %>%
  summarise(user_avg_stars = mean(stars, na.rm = TRUE)) %>%
  summarise(overall_avg = mean(user_avg_stars, na.rm = TRUE))

# Calculate the average of the top 10 users
mean_top10 <- mean(top_users_final$avg_star_rating, na.rm = TRUE)

# Print the results
overall_user_avg
## # A tibble: 1 × 1
##   overall_avg
##         <dbl>
## 1        3.00
mean_top10
## [1] 2.94381

Question 3 Summary

For question 3 we looked at the top 10 users based on number of reviews. We ran into a problem here because there were 16 users with greater than or equal to 14 reviews, so we went by the order in which they were on the dataset. Also, one of the users ended up with the name ‘unknown’. The boxplot only had the top 9 users but wee didn’t have time to fix the code.

For each top 10 user we calculated:

  • Total amount of reviews

  • Average star ratings

We put the results in a table and into a boxplot.

Findings were as follows:

  • Rebecca was the most active reviewer with 18 reviews
  • The top 10 users had an average star rating of 2.94, slightly below the average across all users of 3.00
  • From the boxplot we can see:

    • Jesse and Tracy have the highest median scores (4 stars), while the others all have median scores of 2 or 3 stars

    • Rebecca has the widest distribution of ratings, based on the length of the box

    • Stephanie is the only user who has outliers with 1 and 5 star ratings, based on the dots

    • Those with shorter boxes (like Stephanie and Jason) give the most consistent ratings on average

Question 4

1.1 Merge and Extract User Groups

#Merge data from reviews (left table) and users using left_join 

merged_q4 <- reviews %>%
  
  #join by common variable user_id
  left_join(users, by = "user_id") %>%
  
  #remove rows with missing values
  filter(!is.na(stars), !is.na(text), !is.na(member_since))


#split users based on if they joined before or after 2020
merged_q4 <- merged_q4 %>%
  
  #add new column named user_group that will hold a label for all the users
  mutate(
    
    #if user joined prior to 2020
    user_group = case_when(
      as.numeric(substr(member_since, 1, 4)) < 2020 ~ "Before 2020",
      
      #if user joined during of after 2020
      as.numeric(substr(member_since, 1, 4)) >= 2020 ~ "After 2020"
    )
  )

4.2 Check the Review Length and Summarise by User Group

#add column called review_length that contains number of chars in review
merged_q4 <- merged_q4 %>%
  mutate(
    
    #count characters
    review_length = nchar(text)
  )

#summarise by group

#create new table to story summary
summary_q4 <- merged_q4 %>%
   
  #group by whether user group is before or after 2020
  group_by(user_group) %>%
  
  #calculations to do for each group
    summarise(
      
      #whats the average length of review and rm missing values
        mean_length = mean(review_length, na.rm = TRUE),
        #how many reviews are in the groups
        num_of_reviews = n(),
        
        #whats the average star rating and rm missing values
        mean_star = mean(stars, na.rm = TRUE),
        
        #number of unique users, got the function from google (i think its called a function)
        num_users = n_distinct(user_id)

    )

Table 4.1 Results for Question 4

#Reorder columns to match custom table headers
summary_q4 <- summary_q4 %>%
  select(user_group, mean_star, mean_length, num_of_reviews, num_users)

#Put the results from 4.3 into the table in kable

summary_q4 %>%
  
  kable(
    
    #Make the column names look nicer (capitalise and remove underscores)
    col.names = c("User Group", "Mean Star Rating", "Mean Review Length", "Number of Reviews", "Number of Users"),
    
    #Put all the text as right align
    align = c("r", "r", "r", "r", "r"),
    
    #use html output for styling
    format = "html",
    
    #continuous variables as 2dp
    digits = 2
  ) %>%
  
  
  kable_styling(
    
    #table is shorter rather than fit to page 
    full_width = FALSE,
    
    #striped means row colours alternate, condensed means less padding
    bootstrap_options = c("striped", "condensed")
    
  ) %>%
  
  #Header row is white text, bold text and purple background 
  row_spec(0, bold = TRUE, color = "#ffffff", background = "#6a1b9a") %>%
  
  #other rows are dark grey text
  row_spec(1:nrow(summary_q4), color = "#2c3e50") %>%
  
  #add a border for all the rows, make the column width the same for all the columns
  column_spec(1:5, border_right = TRUE, border_left = TRUE, width = "6em")
User Group Mean Star Rating Mean Review Length Number of Reviews Number of Users
After 2020 3 58.98 99115 19670
Before 2020 3 59.09 83937 17630

Refer to table 1.1 description as the code is the same.

Figure 4.1 Bar Chart of Average Review Length by User Group

#use ggplot to make bar chart to compare avg review length for both groups (before and after 2020 joining)
ggplot(summary_q4, aes(x = user_group, y = mean_length, fill = user_group)) +
  
  #use review length as the length of the bars and set to white
  geom_col(width = 0.6, color = "white") +
  
  #add labels
  labs(
    title = "Average Review Length by User Group",
    x = "User Group",
    y = "Mean Review Length (Characters)"
  ) +
  
  #use this theme to match the main theme of the document
  theme_minimal() +
  
  #customise text style and colours to match the main theme
  theme(
    #title is bold and purple
    plot.title = element_text(color = "#6a1b9a", face = "bold", size = 14),
    
    #axis labels are also purple and bold
    axis.title = element_text(color = "#6a1b9a", face = "bold"),
    
    #axis numbers are dark grey
    axis.text = element_text(color = "#2c3e50"),
    
    #remove legend
    legend.position = "none"
  ) +
  
  #use main theme colours for the bars
  scale_fill_manual(values = c("Before 2020" = "#6a1b9a", "After 2020" = "#f76b1c"))

Question 4 Summary

For question 4 we aimed to determine if there were major differences between the review behaviour of users who joined before vs after 2020.

Analysis included:

  • mean star rating

  • mean review length

  • number of reviews

Insights include:

  • An average rating of 3 stars was consistent between both groups.
  • ‘Before 2020’ had longer reviews with an average of 59.09 characters compared to ‘After 2020’ with an average of 58.98 characters.
  • ‘After 2020’ had more reviews with 99,115 compared to ‘Before 2020’ with 83,937.
  • The number of unique users was fairly evenly split, with ‘Before 2020’ having slightly less (17,630) compared to ‘After 2020’ with 19,670.
  • From the barplot we can see that the mean review length for both groups was very similar