Google analytics content segmentation k means segmentation in R

basic plot.r

plot <- ggplot (data=data,
          aes(x=pageTitle,y=pageviews)) +
            geom_col()
#not so pretty plot
plot 

compare k means.r

library(rstatix)
k3 <- kmeans(d_top50summary, centers = 3, nstart = 25)
k4 <- kmeans(d_top50summary, centers = 4, nstart = 25)
k5 <- kmeans(d_top50summary, centers = 5, nstart = 25)

# plots to compare
p1 <- fviz_cluster(k2, geom = "point", data = df) + ggtitle("k = 2")
p2 <- fviz_cluster(k3, geom = "point",  data = df) + ggtitle("k = 3")
p3 <- fviz_cluster(k4, geom = "point",  data = df) + ggtitle("k = 4")
p4 <- fviz_cluster(k5, geom = "point",  data = df) + ggtitle("k = 5")
library(gridExtra)

grid.arrange(p1, p2, p3, p4, nrow = 2)

convert row names from string to numeric.r

#SO solution to convert the row names for string to 
#numeric values
d_top50summary<- d_top50summary %>% remove_rownames %>% 
  column_to_rownames(var="postType")

create row id.r

#Add row # as an ID so that it makes it easier to read
#a chart for Pareto later
#https://stackoverflow.com/questions/23518605/add-an-index-numeric-id-column-to-large-data-frame
d_subset <- tibble::rowid_to_column(d_subset, "ID")
view(d_subset)

distance matrix.r

d_top50summary <- scale(d_top50summary)
view(d_top50summary)
library(factoextra)
library(ggplot2)
#https://www.analyticsvidhya.com/blog/2020/02/4-types-of-distance-metrics-in-machine-learning/
distance <- get_dist(d_top50summary)
fviz_dist(distance, 
          gradient = list(low = "#00AFBB", mid = "white", high = "#FC4E07"))

dplyr mutate.r

d_top50summary<- d_top50exclHPOthers %>%
  group_by(postType)%>%
  summarize(
    br=sum(bounces)/sum(pageviews)*100,
    timeonpage=sum(timeonpage)/sum(pageviews),
    entrancerate=sum(entrances)/sum(pageviews)*100,
    exitrate=sum(exits)/sum(pageviews)*100,
    pagevaluetotal=sum(pagevalue),
    pageviewstotal=sum(pageviews)
    )%>%
  mutate(br,timeonpage,entrancerate, exitrate,pagevaluetotal,
         pageviewstotal)
view(d_top50summary)
d_top50summary <- relocate(d_top50summary,postType,pageviewstotal)

elbow method k means.r

set.seed(123)
fviz_nbclust(d_top50summary, kmeans, method = "wss")

exclude zero pageviews.r

#0 pageviews doesn't help...let's exclude them
d_subset <- subset(d_sorted,pageviews>0)

full code.r

#install packages
install.packages("googleAnalyticsR")
install.packages("tidyverse")

#load libraries
library(googleAnalyticsR)
library(tidyverse)

#creating service account json key
#http://code.markedmondson.me/googleAnalyticsR/articles/setup.html 
ga_auth(json_file = "C:\\json-key-folder\\service-key.json")
#check View Id's
account_list <- ga_account_list()
account_list$viewId
#Assign View ID to var
#change to yours
ga_id <- 1234567

data <- google_analytics(ga_id,
date_range = c("2020-03-01","2021-02-16"),
metrics=c("pageviews","bounces","timeonpage",
          "entrances","exits","pagevalue"),
dimensions="ga:pageTitle",
anti_sample=TRUE)

plot <- ggplot (data=data,
          aes(x=pageTitle,y=pageviews)) +
            geom_col()
#not so pretty plot
plot          
#sort data by desc pageviews
d_sorted <- arrange(data, desc(pageviews))          
#Min for PV variable shows 0 in Summary...seems like 
summary(d_sorted)
#0 pageviews doesn't help...let's exclude them
d_subset <- subset(d_sorted,pageviews>0)
summary(d_subset)
view(d_subset)
#Add row # as an ID so that it makes it easier to read
#a chart for Pareto later
#https://stackoverflow.com/questions/23518605/add-an-index-numeric-id-column-to-large-data-frame
d_subset <- tibble::rowid_to_column(d_subset, "ID")
view(d_subset)
#geom_col to view pageview counts by pages (ID's)
plot <- ggplot (data=d_subset,
                aes(x=ID,y=pageviews)) +
  geom_col()
plot
#column for cumulative running total of pageviews
#https://nhsrcommunity.com/blog/pareto-chart-in-ggplot2/
d_subset$cumsum <- cumsum(d_subset$pageviews)
view(d_subset)
#let's try out a pareto chart
pareto <- ggplot(d_subset, aes(x=ID,y=pageviews)) +
  geom_col() + 
  geom_path(aes(y=cumsum, group=1),
            colour="slateblue1", lty=3, size=0.9) +
  labs(title = "Pareto Plot",
       x = 'page id', y = 'pageviews')
pareto
summary(d_subset)

install.packages("ggQC")
library(ggQC)
#https://stackoverflow.com/questions/1735540/creating-a-pareto-chart-with-ggplot2-and-r
#SO solution: Package called ggQC for this
ggplot2::ggplot(d_subset, aes(x = ID, y = pageviews)) +
ggQC::stat_pareto(point.color = "red",
                    point.size = 0.2,
                    line.color = "black",
                    bars.fill = c("blue"))
#at appx page ID = 50 (top 50 pages), we can 
#account for appx 80% of the total content
d_top50 <- subset(d_subset,ID<=50)
d_top50
view(d_top50)
#ok, now let's try to add a conditional field
#that creates a blog post category using the keywords
#from the page title. We can then use the blog post
#category as the content group

#https://stackoverflow.com/questions/51872718/how-to-detect-if-a-string-contains-a-regex-pattern-using-dplyr-pipe-mutate/51872769
#SO solution on IF ELSE with str_detect
d_top50content <- d_top50 %>%
  mutate(pageTitle = tolower(pageTitle))%>%
  mutate(postType = 
    ifelse(str_detect
      (pageTitle, "r programming|r studio|geom|ggplot"),"R programming",
      ifelse(str_detect
             (pageTitle, "data studio"),"Data Studio",
      ifelse(str_detect
          (pageTitle, "adobe launch"),"Adobe Launch",
      ifelse(str_detect
                 (pageTitle, "bigquery|big query"),"BQ",
      ifelse(str_detect
            (pageTitle, "facebook|social|fb|reach|matched"),"Facebook Ads",  
      ifelse(str_detect
         (pageTitle, "google cloud"),"GCP",
         ifelse(str_detect
                (pageTitle, "google analytics|ga app"),"GA",
      ifelse(str_detect
        (pageTitle, "google tag manager|gtm"),"GTM",
        ifelse(str_detect
               (pageTitle, "screaming frog|vital"),"SEO",
      ifelse(str_detect
          (pageTitle, "analytics log"),"Homepage",
    ifelse(str_detect
           (pageTitle, "adwords|search term"),"Google Ads",      
          ifelse(str_detect
           (pageTitle, "apriori"),"Algorithm",
                 
      "Other"))))))))))))
    )
view(d_top50content)
#exclude homepage and others
d_top50exclHPOthers <- subset(d_top50content,
                postType!="Homepage")

view(d_top50exclHPOthers)
#remove cumulative sum column from the calculations
d_top50exclHPOthers <- select(d_top50exclHPOthers, -cumsum)
view(d_top50exclHPOthers)
#relocate function to move the postType column to the front
#https://dplyr.tidyverse.org/reference/relocate.html
d_top50exclHPOthers <- relocate(d_top50exclHPOthers,postType,pageTitle)

d_top50summary<- d_top50exclHPOthers %>%
  group_by(postType)%>%
  summarize(
    br=sum(bounces)/sum(pageviews)*100,
    timeonpage=sum(timeonpage)/sum(pageviews),
    entrancerate=sum(entrances)/sum(pageviews)*100,
    exitrate=sum(exits)/sum(pageviews)*100,
    pagevaluetotal=sum(pagevalue),
    pageviewstotal=sum(pageviews)
    )%>%
  mutate(br,timeonpage,entrancerate, exitrate,pagevaluetotal,
         pageviewstotal)
view(d_top50summary)
d_top50summary <- relocate(d_top50summary,postType,pageviewstotal)
view(d_top50summary)
#SO solution to convert the row names for string to 
#numeric values
d_top50summary<- d_top50summary %>% remove_rownames %>% 
  column_to_rownames(var="postType")

d_top50summary <- scale(d_top50summary)
view(d_top50summary)
library(factoextra)
library(ggplot2)
#https://www.analyticsvidhya.com/blog/2020/02/4-types-of-distance-metrics-in-machine-learning/
distance <- get_dist(d_top50summary)
fviz_dist(distance, 
          gradient = list(low = "#00AFBB", mid = "white", high = "#FC4E07"))
#https://www.youtube.com/watch?v=_UVHneBUBW0
k2 <- kmeans(d_top50summary, 
             centers = 2, nstart = 25)
str(k2)
k2
fviz_cluster(k2, data = d_top50summary)
#http://www.sthda.com/english/articles/31-principal-component-methods-in-r-practical-guide/118-principal-component-analysis-in-r-prcomp-vs-princomp/
res.pca <- prcomp(d_top50summary, scale = TRUE)

fviz_eig(res.pca)
res.pca
prcomp(d_top50summary)
summary(res.pca)


library(rstatix)
k3 <- kmeans(d_top50summary, centers = 3, nstart = 25)
k4 <- kmeans(d_top50summary, centers = 4, nstart = 25)
k5 <- kmeans(d_top50summary, centers = 5, nstart = 25)
# plots to compare
p1 <- fviz_cluster(k2, geom = "point", data = df) + ggtitle("k = 2")
p2 <- fviz_cluster(k3, geom = "point",  data = df) + ggtitle("k = 3")
p3 <- fviz_cluster(k4, geom = "point",  data = df) + ggtitle("k = 4")
p4 <- fviz_cluster(k5, geom = "point",  data = df) + ggtitle("k = 5")
library(gridExtra)

grid.arrange(p1, p2, p3, p4, nrow = 2)

set.seed(123)
fviz_nbclust(d_top50summary, 
             kmeans, method = "wss")

geom_col.r

#geom_col to view pageview counts by pages (ID's)
plot <- ggplot (data=d_subset,
                aes(x=ID,y=pageviews)) +
  geom_col()
plot

ggQC pareto.r

install.packages("ggQC")
library(ggQC)
#https://stackoverflow.com/questions/1735540/creating-a-pareto-chart-with-ggplot2-and-r
#SO solution: Package called ggQC for this
ggplot2::ggplot(d_subset, aes(x = ID, y = pageviews)) +
ggQC::stat_pareto(point.color = "red",
                    point.size = 0.2,
                    line.color = "black",
                    bars.fill = c("blue"))
#at appx page ID = 50 (top 50 pages), we can 
#account for appx 80% of the total content
d_top50 <- subset(d_subset,ID<=50)

Google Analytics Content Segmentation K Means Clustering.r

#install packages
install.packages("googleAnalyticsR")
install.packages("tidyverse")

#load libraries
library(googleAnalyticsR)
library(tidyverse)

#creating service account json key
#http://code.markedmondson.me/googleAnalyticsR/articles/setup.html 
ga_auth(json_file = "C:\\Users\\akhan\\Dropbox\\blog content\\2021\\Feb 2021\\K means clustering\\service-key.json")
#check View Id's
account_list <- ga_account_list()
account_list$viewId
#Assign View ID to var
ga_id <- 41377551

data <- google_analytics(ga_id,
date_range = c("2020-03-01","2021-02-16"),
metrics=c("pageviews","bounces","timeonpage",
          "entrances","exits","pagevalue"),
dimensions="ga:pageTitle",
anti_sample=TRUE)

plot <- ggplot (data=data,
          aes(x=pageTitle,y=pageviews)) +
            geom_col()
#not so pretty plot
plot          
#sort data by desc pageviews
d_sorted <- arrange(data, desc(pageviews))          
#Min for PV variable shows 0 in Summary...seems like 
summary(d_sorted)
#0 pageviews doesn't help...let's exclude them
d_subset <- subset(d_sorted,pageviews>0)
summary(d_subset)
view(d_subset)
#Add row # as an ID so that it makes it easier to read
#a chart for Pareto later
#https://stackoverflow.com/questions/23518605/add-an-index-numeric-id-column-to-large-data-frame
d_subset <- tibble::rowid_to_column(d_subset, "ID")
view(d_subset)
#geom_col to view pageview counts by pages (ID's)
plot <- ggplot (data=d_subset,
                aes(x=ID,y=pageviews)) +
  geom_col()
plot
#column for cumulative running total of pageviews
#https://nhsrcommunity.com/blog/pareto-chart-in-ggplot2/
d_subset$cumsum <- cumsum(d_subset$pageviews)
view(d_subset)
#let's try out a pareto chart
pareto <- ggplot(d_subset, aes(x=ID,y=pageviews)) +
  geom_col() + 
  geom_path(aes(y=cumsum, group=1),
            colour="slateblue1", lty=3, size=0.9) +
  labs(title = "Pareto Plot",
       x = 'page id', y = 'pageviews')
pareto
summary(d_subset)

install.packages("ggQC")
library(ggQC)
#https://stackoverflow.com/questions/1735540/creating-a-pareto-chart-with-ggplot2-and-r
#SO solution: Package called ggQC for this
ggplot2::ggplot(d_subset, aes(x = ID, y = pageviews)) +
ggQC::stat_pareto(point.color = "red",
                    point.size = 0.2,
                    line.color = "black",
                    bars.fill = c("blue"))
#at appx page ID = 50 (top 50 pages), we can 
#account for appx 80% of the total content
d_top50 <- subset(d_subset,ID<=50)
d_top50
view(d_top50)
#ok, now let's try to add a conditional field
#that creates a blog post category using the keywords
#from the page title. We can then use the blog post
#category as the content group

#https://stackoverflow.com/questions/51872718/how-to-detect-if-a-string-contains-a-regex-pattern-using-dplyr-pipe-mutate/51872769
#SO solution on IF ELSE with str_detect
d_top50content <- d_top50 %>%
  mutate(pageTitle = tolower(pageTitle))%>%
  mutate(postType = 
    ifelse(str_detect
      (pageTitle, "r programming|r studio|geom|ggplot"),"R programming",
      ifelse(str_detect
             (pageTitle, "data studio"),"Data Studio",
      ifelse(str_detect
          (pageTitle, "adobe launch"),"Adobe Launch",
      ifelse(str_detect
                 (pageTitle, "bigquery|big query"),"BQ",
      ifelse(str_detect
            (pageTitle, "facebook|social|fb|reach|matched"),"Facebook Ads",  
      ifelse(str_detect
         (pageTitle, "google cloud"),"GCP",
         ifelse(str_detect
                (pageTitle, "google analytics|ga app"),"GA",
      ifelse(str_detect
        (pageTitle, "google tag manager|gtm"),"GTM",
        ifelse(str_detect
               (pageTitle, "screaming frog|vital"),"SEO",
      ifelse(str_detect
          (pageTitle, "analytics log"),"Homepage",
    ifelse(str_detect
           (pageTitle, "adwords|search term"),"Google Ads",      
          ifelse(str_detect
           (pageTitle, "apriori"),"Algorithm",
                 
      "Other"))))))))))))
    )
view(d_top50content)
#exclude homepage and others
d_top50exclHPOthers <- subset(d_top50content,
                postType!="Homepage")

view(d_top50exclHPOthers)
#remove cumulative sum column from the calculations
d_top50exclHPOthers <- select(d_top50exclHPOthers, -cumsum)
view(d_top50exclHPOthers)
#relocate function to move the postType column to the front
#https://dplyr.tidyverse.org/reference/relocate.html
d_top50exclHPOthers <- relocate(d_top50exclHPOthers,postType,pageTitle)

d_top50summary<- d_top50exclHPOthers %>%
  group_by(postType)%>%
  summarize(
    br=sum(bounces)/sum(pageviews)*100,
    timeonpage=sum(timeonpage)/sum(pageviews),
    entrancerate=sum(entrances)/sum(pageviews)*100,
    exitrate=sum(exits)/sum(pageviews)*100,
    pagevaluetotal=sum(pagevalue),
    pageviewstotal=sum(pageviews)
    )%>%
  mutate(br,timeonpage,entrancerate, exitrate,pagevaluetotal,
         pageviewstotal)
view(d_top50summary)
d_top50summary <- relocate(d_top50summary,postType,pageviewstotal)
view(d_top50summary)
#SO solution to convert the row names for string to 
#numeric values
d_top50summary<- d_top50summary %>% remove_rownames %>% 
  column_to_rownames(var="postType")

d_top50summary <- scale(d_top50summary)
view(d_top50summary)
library(factoextra)
library(ggplot2)
#https://www.analyticsvidhya.com/blog/2020/02/4-types-of-distance-metrics-in-machine-learning/
distance <- get_dist(d_top50summary)
fviz_dist(distance, 
          gradient = list(low = "#00AFBB", mid = "white", high = "#FC4E07"))
#https://www.youtube.com/watch?v=_UVHneBUBW0
k2 <- kmeans(d_top50summary, 
             centers = 2, nstart = 25)
str(k2)
k2
fviz_cluster(k2, data = d_top50summary)
#http://www.sthda.com/english/articles/31-principal-component-methods-in-r-practical-guide/118-principal-component-analysis-in-r-prcomp-vs-princomp/
res.pca <- prcomp(d_top50summary, scale = TRUE)

fviz_eig(res.pca)
res.pca
prcomp(d_top50summary)
summary(res.pca)


library(rstatix)
k3 <- kmeans(d_top50summary, centers = 3, nstart = 25)
k4 <- kmeans(d_top50summary, centers = 4, nstart = 25)
k5 <- kmeans(d_top50summary, centers = 5, nstart = 25)
# plots to compare
p1 <- fviz_cluster(k2, geom = "point", data = df) + ggtitle("k = 2")
p2 <- fviz_cluster(k3, geom = "point",  data = df) + ggtitle("k = 3")
p3 <- fviz_cluster(k4, geom = "point",  data = df) + ggtitle("k = 4")
p4 <- fviz_cluster(k5, geom = "point",  data = df) + ggtitle("k = 5")
library(gridExtra)

grid.arrange(p1, p2, p3, p4, nrow = 2)

set.seed(123)
fviz_nbclust(d_top50summary, 
             kmeans, method = "wss")

googleanalyticsR api call.r

library(googleAnalyticsR)
library(tidyverse)

#creating service account json key
#http://code.markedmondson.me/googleAnalyticsR/articles/setup.html 
ga_auth(json_file = "C:\\Your-Folder-Containing-JSON-file\\service-key.json")
#check View Id's
account_list <- ga_account_list()
account_list$viewId
#Assign View ID to var
ga_id <- 1234567

data <- google_analytics(ga_id,
date_range = c("2020-03-01","2021-02-16"),
metrics=c("pageviews","bounces","timeonpage",
          "entrances","exits","pagevalue"),
dimensions="ga:pageTitle",
anti_sample=TRUE)

k means cluster.r

k2 <- kmeans(d_top50summary, centers = 2, nstart = 25)
str(k2)
k2
fviz_cluster(k2, data = d_top50summary)

pareto chart.r

#column for cumulative running total of pageviews
#https://nhsrcommunity.com/blog/pareto-chart-in-ggplot2/
d_subset$cumsum <- cumsum(d_subset$pageviews)
view(d_subset)
#let's try out a pareto chart
pareto <- ggplot(d_subset, aes(x=ID,y=pageviews)) +
  geom_col() + 
  geom_path(aes(y=cumsum, group=1),
            colour="slateblue1", lty=3, size=0.9) +
  labs(title = "Pareto Plot",
       x = 'page id', y = 'pageviews')
pareto

scree plot.r

#http://www.sthda.com/english/articles/31-principal-component-methods-in-r-practical-guide/118-principal-component-analysis-in-r-prcomp-vs-princomp/
res.pca <- prcomp(d_top50summary, scale = TRUE)
fviz_eig(res.pca)
res.pca
prcomp(d_top50summary)
summary(res.pca)

sort by desc.r

#sort data by desc pageviews
d_sorted <- arrange(data, desc(pageviews))          
#Min for PV variable shows 0 in Summary...seems like 
summary(d_sorted)

str_detect.r

d_top50content <- d_top50 %>%
  mutate(pageTitle = tolower(pageTitle))%>%
  mutate(postType = 
    ifelse(str_detect
      (pageTitle, "r programming|r studio|geom|ggplot"),"R programming",
      ifelse(str_detect
             (pageTitle, "data studio"),"Data Studio",
      ifelse(str_detect
          (pageTitle, "adobe launch"),"Adobe Launch",
      ifelse(str_detect
                 (pageTitle, "bigquery|big query"),"BQ",
      ifelse(str_detect
            (pageTitle, "facebook|social|fb|reach|matched"),"Facebook Ads",  
      ifelse(str_detect
         (pageTitle, "google cloud"),"GCP",
         ifelse(str_detect
                (pageTitle, "google analytics|ga app"),"GA",
      ifelse(str_detect
        (pageTitle, "google tag manager|gtm"),"GTM",
        ifelse(str_detect
               (pageTitle, "screaming frog|vital"),"SEO",
      ifelse(str_detect
          (pageTitle, "analytics log"),"Homepage",
    ifelse(str_detect
           (pageTitle, "adwords|search term"),"Google Ads",      
          ifelse(str_detect
           (pageTitle, "apriori"),"Algorithm",
                 
      "Other"))))))))))))
    )
view(d_top50content)
#exclude homepage and others
d_top50exclHPOthers <- subset(d_top50content,
                postType!="Homepage")

view(d_top50exclHPOthers)
#remove cumulative sum column from the calculations
d_top50exclHPOthers <- select(d_top50exclHPOthers, -cumsum)
view(d_top50exclHPOthers)
#relocate function to move the postType column to the front
#https://dplyr.tidyverse.org/reference/relocate.html
d_top50exclHPOthers <- relocate(d_top50exclHPOthers,postType,pageTitle)