Plotting & Theme Functions • DisplayR

library(ggplot2)
library(DisplayR)
library(dplyr)
library(tidyr)

Plotting Functions

When working on research projects or Understand phases, there is often a set of charts you will create at the beginning, and then at multiple points throughout the project. These charts will help you perform common exploratory data analysis (EDA) tasks to better understand your data and identify areas of investigation. Some of these charts sit naturally in other packages such as:

{ParseR} - descriptive plots of text distributions e.g. Pairwise Correlations, n-gram plots, Weighted log-odds
{SegmentR} - plots for topic modelling outputs e.g. top terms
{ConnectR} - network visualisations e.g. Retweet network, mention network

Other plots which do not have a clear and obvious location, e.g. sentiment distributions, volume over time etc. will sit here in {DisplayR}. They most likely will not be the centrepiece of a delivery, and they should not be used as a substitute for thinking about the problem at hand and creating bespoke graphics when called for*. They do however represent valuable efficiency gains in the EDA workflow, and prime clients for a Capture Intelligence deck or delivery.

*See the Capture Cookbook and Visualisation Guidelines for more information on creating Bespoke graphics and adhering to good data visualisation principles.

Volume Over Time

When working with the data from various Social Media APIs (Twitter, Facebook, Reddit etc.) there is usually a time stamp or date/time column, this column can be used to visualise when the conversation takes place and to identify when important changes in volume occur.

We’ll use {DisplayR}’s dataset:

df <- DisplayR::disp_example

Single Element - Ungrouped Data Frame

We feed the dataset into the disp_plot_vot function, setting date = date, plot_type to line or bar and time_unit to whatever unit we wish to use (see the function documentation (?dr_plot_vot) for available units).

df %>%
  dr_plot_vot(date_var = date, 
                plot_type = "line", 
                time_unit = "day")

Alternatively, we could generate a bar plot where we count the volume per week and change the plot colour:

df %>%
  dr_plot_vot(date_var = date, 
              plot_type = "bar", 
              time_unit = "week",
              "#d80a83",
              smooth = TRUE)
#> `geom_smooth()` using method = 'loess' and formula = 'y ~ x'

You’ll notice the plot’s title, y-axis label and x-axis text (there is no x axis label currently) change auto-magically.

Multiple Elements - Grouped Data Frame

Often we want - or need - to visualise the trend in volume for multiple things at the same time, where things are similar objects contained in the same variable. For example, we may want to visualise the volume over time per topic, or per sentiment category.

We can plot the volume of each topic over time, though it can be (is nearly always) inadvisable to map 9 colours on a plot:

df %>%
  filter(!is.na(sentiment)) %>%
  dr_plot_vot_group(group_var = topic,
                    date_var = date,
                    time_unit = "week")

As there are only three sentiment categories, we could map sentiment more simply, and use the scale_colour_manual function to re-colour our lines:

sentiment_colours <- c("positive" = "#107C10", "negative" = "#D83B01", "neutral" = "#FFB900")

df %>%
  filter(!is.na(sentiment)) %>%
  dr_plot_vot_group(group_var = sentiment,
                    date_var = date,
                    time_unit = "day") +
  ggplot2::scale_colour_manual(values = sentiment_colours) +
  ggplot2::labs(title = stringr::str_wrap("Neutral and negative sentiment consistently higher than positive barring brief periods in March.", width = 80)) # Use stringr::str_wrap to wrap the title according to a specified with (# of characters)
#> Scale for colour is already present.
#> Adding another scale for colour, which will replace the existing scale.

Sentiment Distributions

We looked briefly at how to plot sentiment over time using the dr_plot_vot_group function, but we can also use some sentiment-specific functions to visualise sentiment.

We have:

dr_plot_sent - a vertical bar chart showing the distribution of sentiment
dr_plot_sent_group - stacked horizontal bar chart for group sentiment
dr_plot_sent_vot - stacked bar chart specifically for sentiment categories

Ungrouped

When we want to plot a quick sentiment distribution chart using bars we can use the dr_plot_sent function, which has limited scope for customisation, but we can manually add things like a title. The bar_label can be set to “percent”, “volume” or “none”.

df %>%
  dr_plot_sent(sentiment_var = sentiment,
               bar_labels = "percent") +
  labs(title = "Negative sentiment ~14% higher than positive")

Grouped

More often than not, we won’t want to create just one sentiment chart, so we’ve created a function for plotting grouped sentiment charts.

The dr_plot_sent_group function comes with the additional argument, group_var =. When setting plot_type = "percent" and bar_labels = "volume", we can add the y-axis (visually, the x-axis in the output chart) ticks back and give a title to make the plot more explicit,

df %>%
  dr_plot_sent_group(group_var = topic, plot_type = "percent", bar_labels = "volume") +
  scale_y_continuous() + # Add the percentage back to the axis ticks
  labs(y = "Percentage of posts")  #Add the label back
#> Scale for y is already present.
#> Adding another scale for y, which will replace the existing scale.

There are other options, we can set the bar’s labels to percent, and titles or change the axis labels, move the legend etc.

df %>%
  dr_plot_sent_group(group_var = topic, plot_type = "percent", bar_labels = "percent") +
  labs(x = "Percentage of posts", #We use x = because the plot has had its axes reversed
       title = "Risks & Challenges drive negative sentiment",
       subtitle = "Plot ordered by overall volume", #Just an example
       caption = "Data Source: Social Media")+#Just an example
  theme(legend.position = "right")

We can also plot the volumes rather than percentages on the bars themselves:

df %>%
  dr_plot_sent_group(group_var = topic, sentiment_var = sentiment, plot_type = "volume", bar_labels = "none") +
  theme(panel.grid.major.x = element_line()) +
  scale_y_continuous()
#> Scale for y is already present.
#> Adding another scale for y, which will replace the existing scale.

Sentiment over time

We showed an example of using the dr_plot_vot_group function to visualise sentiment over time, there is also the dr_plot_sent_vot function, made specifically for sentiment and can provide bars or lines. As we’ve already seen lines for sentiment, we’ll show bars. This function allows the user to enter the sentiment mapping directly, so let’s switch to a more accessibility-friendly colour palette.

df %>%
  dr_plot_sent_vot(sentiment_var = sentiment,
                   date_var = date,
                   plot_type = "bar",
                   time_unit = "day",
                   sentiment_colours = c("positive" = "#007fff",
                                         "negative" = "#D83B01",
                                         "neutral" =  "#a89797"))

Raincloud plot

Visualising the distribution of a variable is an important step in obtaining information about our data.

There are many visualisation techniques to effectively relay this information, such as histograms, boxplots, and density plots. However, each of these comes with their own pitfalls, for example boxplots hide data the number of data points ‘behind’ the boxplot is an unknown.

A very neat approach is known as a raincloud plot, which combines multiple visualisations into an intuitive, modular, and statistically robust format. Effectively it combines a density plot, raw jittered data points, and a boxplot to offer maximum utility whilst ensuring the underlying data is not hidden by summary statistics.

This plot can be achieved through the dr_plot_raincloud()

set.seed(123)
df %>%
  slice_sample(n = 500) %>%
  drop_na(sentiment) %>%
  dr_plot_raincloud(grouping_variable = sentiment,
                    continuous_variable = V1,
                    smoothness = 1)

Theme Functions

There are a number of theme functions included in {DisplayR}, they are to be applied according to the nature of the project and tend to take care of colour palettes and fonts. If you think there is a theme, or a number of themes missing, submit an issue at: (https://github.com/jpcompartir/DisplayR/issues).

The general idea is to add the theme you need after rendering your plot. Let’s get a plot and then style it using our themes.

(
  plot <- df %>%
  dr_plot_vot_group(group_var = topic, time_unit = "day")
 )

We could add a dark theme with dr_theme_black

plot +
  dr_theme_black()

Or we could add a Microsoft theme using dr_theme_microsoft:

plot +
  dr_theme_microsoft(scale_type = "discrete")
#> Scale for colour is already present.
#> Adding another scale for colour, which will replace the existing scale.

A SHARE theme using dr_theme_share:

plot +
  dr_theme_share(scale_type = "discrete")
#> Scale for colour is already present.
#> Adding another scale for colour, which will replace the existing scale.

A SAMY theme using dr_theme_samy:

plot +
  dr_theme_samy(scale_type = "discrete")
#> Scale for colour is already present.
#> Adding another scale for colour, which will replace the existing scale.