3.1.1 Understanding the code

# Creating the dataset
ministry <- questions %>% group_by(Ministry) %>% 
    summarise(Count = n()) %>% 
    arrange(desc(Count)) %>%
    mutate(Ministry = factor(Ministry, levels = rev(Ministry)))

First, I create a new dataframe that consists of the number of questions asked to each ministry, and transform the Ministry column into a factor.

The process is:

1. Group the dataset by Ministry - using group_by()
2. summarise to get the number of occurences - using n()
3. Sort the dataframe in descending order of Count - using arrange()
4. Convert Ministry into a factor - using mutate()

Next, we look at the code for creating the plot.

# Creating the plot
ggplot(ministry, aes(x = Ministry, y = Count, fill = Ministry)) + geom_bar(stat = 'identity') + 
    geom_hline(yintercept = 0, size = 1, colour="#333333") +
    scale_fill_manual(values = rev(colorRampPalette(brewer.pal(8, "Dark2"))(length(ministry$Ministry)))) +
    bbc_style() +
    coord_flip() +

# Removing legend, showing axis labels and adding the title
    theme(legend.position = "none", 
            axis.title = element_text(size = 18), 
            panel.grid.major.x = element_line(color="#cbcbcb"), 
            panel.grid.major.y=element_blank()) +

    labs(title = "Questions asked by each ministry", 
          subtitle = "Winter Session of 17th Lok Sabha", 
          y = "Number of questions") +

# Showing the number of questions in the graph
    geom_label(aes(x = Ministry, y = Count, label = Count),
         hjust = 1, 
         vjust = 0.5, 
         colour = "white", 
         fill = NA, 
         label.size = NA, 
         family="Helvetica", 
         size = 6)

The first section is the basic ggplot2 code for creating a horizontal bar chart, with manual colors. Of particular importance here is: bbc_style()

bbc_style() (and subsequently, finalise_plot()) is a function from bbplot that makes the chart components follow BBC style, while allowing room for further manual customization. For more information, visit the bbplot GitHub repo.

Then, I remove the legend, show the axis labels, add a title and subtitle to the plot in the next two sections. Post that, I add the number of questions as a label in the chart to facilitate easy comprehension.

finalise_plot() simply packages the graphic, adds a footnote and resizes it - producing an image - QuestionsByMinistry.jpg

3.1.2 Understanding the graphic

Now that we have understood how to create the above graphic, let us take a moment to interpret it.

Health and Family Welfare leads the race with 332 questions asked to it, with Railways trailing at 288.

It seems that most of the core ministries, such as, Human Resource Development, Environment, Finance, Road Transport and Highways, were asked questions in this session, which does that indicate that the House debated on some pertitnent topics. However, a closer analysis is required on the subjects of such questions.

3.2.1 Understanding the code

# Creating the dataset

shift_data <- questions %>% group_by(Week = floor_date(Date, 'week'), Ministry) %>% 
    summarise(count = n()) %>% 
    top_n(5, wt = count) %>% 
    group_by(Week) %>%
    arrange(Week, desc(count)) %>%
    mutate(rank = row_number()) %>%
    filter(rank <= 5) %>%
    ungroup()

Here, we want to create rankings for each ministry based on the number of questions asked to them in each week. So, our workflow would be to convert the data from daily to weekly, count the number of questions, assign ranks and plot it. To create the dataset, we follow this process:

1. Group the dataset by Week and Ministry - using group_by() and floor_date()
2. summarise to get the count of number of questions - using n()
3. Keep only the top 5 ministries in each week - using top_n()
4. Sort by descending order of number of questions asked to each week - using arrange()
5. Add a rank column for each week and ministry - using mutate() and row_number()
6. Remove any ministry with overlapping ranks - using filter()

Next, we look at the code for creating the plot.

# Creating the plot
ggplot(shift_data, aes(x = Week, y = rank, group = Ministry)) +
    geom_line(aes(color = Ministry), size = 2) +
    geom_point(aes(color = Ministry), size = 5) +
    scale_y_reverse(breaks = 1:5) +
    bbc_style() +
    scale_color_tableau() +
    
# Adding legend, showing axis labels and adding the title
    theme(legend.position = 'right', 
            axis.title = element_text(size = 18), 
            panel.grid.major.y = element_line(color="#cbcbcb"), 
            panel.grid.major.x=element_blank()) +
    
    labs(title = "Top 5 ministries by number of questions", 
            subtitle = "Winter Session of 17th Lok Sabha",
            y = "Rank", 
            x = "Week")

Similar to the previous plot, this one also creates a basic ggplot2 chart, adds bbc_style() and scale_color_tableau() and the labels and titles. Pretty standard stuff!

3.2.2 Understanding the graphic

As before, let us interpret this visualisation as well.

Agriculture and Farmers’ Welfare dominated the questions in the first week of the Session, but disappeared in the next two weeks, only to come back at second place in the last week.

Health and Family Welfare jumped to the top spot after first week and remained there, while Human Resource Development created a plateau at the 4th and 2nd place. Road Transport and Highways, Home Affairs and Jal Shakti made the top charts at least once.

All in all, this visualisation seems to provide a bit more insight into how the focus shifted from one Ministry to another during the session.

3.3.1 Creating a wordcloud

To visualise textual data, let’s create a wordcloud from the subject of questions asked.

# Ministry-wise - Questions wordlcloud

ministry_wordcloud <- function(ministry){
    
    # Filtering selected ministry and creating the dataset
    ministry_q <- questions %>% filter(Ministry == str_to_upper(ministry)) %>%
        mutate(Subject = as.character(Subject)) %>%
        select(Subject) %>%
        unnest_tokens(word, Subject) %>%
        anti_join(get_stopwords(), by = 'word') %>%
        count(word, sort = T) %>%a
        head(40)
        
    # Plotting the dataset as a wordcloud
    (
        ggplot(ministry_q, aes(label = word, color = word, size = n)) + 
            geom_text_wordcloud_area(rm_outside = T, family = 'Helvetica') +
            scale_size_area(max_size = 40) +
            bbc_style() +
            labs(title = paste0("Ministry of ", ministry, " - Subject of questions asked"), 
                 subtitle = "Winter Session of 17th Lok Sabha")

    ) %>% finalise_plot(source_name = "Data: Lok Sabha; Created by Lakshya Agarwal", 
                        save_filepath = paste0('graphs/wordcloud/', ministry, '.jpg'), 
                        width = 960, height = 540)
}

The above code creates a function - ministry_wordcloud() - that takes in a Ministry Name as an input and produces a wordcloud of the subject of questions asked to that ministry. For example, for Health and Family Welfare, we run

ministry_wordcloud("Health and Family Welfare")

which gives

# Creating the dataset
ministry_q <- questions %>% filter(Ministry == str_to_upper(ministry)) %>%
        mutate(Subject = as.character(Subject)) %>%
        select(Subject) %>%
        unnest_tokens(word, Subject) %>%
        anti_join(get_stopwords(), by = 'word') %>%
        count(word, sort = T) %>%a
        head(40)

ggplot(ministry_q, aes(label = word, color = word, size = n)) + 
            geom_text_wordcloud_area(rm_outside = T, family = 'Helvetica') +
            scale_size_area(max_size = 40) +
            bbc_style() +
            labs(title = paste0("Ministry of ", ministry, " - Subject of questions asked"), 
                 subtitle = "Winter Session of 17th Lok Sabha")

3.3.2 Understading the wordcloud

Now that we have understood how the wordcloud was generated, we can turn to interpreting them and gleaning information.

3.3.3 Shiny app

I have also created an application for the above visualisation that can be accessed here.