Text Visualization

LIS 4/5693: Information Retrieval and Text Mining

Dr. Manika Lamba

A Picture Is Worth a Thousand Words

Our brains have been evolved to be very efficient in visual analysis

Visualization helps us to:

• absorbs information quickly

• connect the dots

• find patterns and outliers

• acquire & share insights

• attract audience

In this module, we will learn more about data visualization and the basics behind it. Visualization is an essential part of exploratory data analysis, enabling faster comprehension, better decision-making, and more impactful storytelling with data.

Text analysis is not just about collecting data, it’s about transforming raw data into meaningful insights, knowledge, and ultimately, impact. However, misinterpretation or excessive manipulation can lead to misleading conclusions.

Data visualization helps the user make sense of large (and often times small) amounts of data using computational and analytic tools. In academic and special libraries data visualization helps researchers and business users recognize and understand the “patterns” in the data to assist them in decision making, research, product development, marketing, and other uses not yet realized.

Along with the emergence of “big data” came the need for more visual means to analyze and make sense of the data. Visualization also allows for re-purposing of datasets beyond their original use.

Information professionals assist in data visualization but also in managing and preparing the data for use.

An academic library may title these professionals as Data analysis librarian, Data manager, Digital Projects librarian/specialist, etc. The titles are emerging as fast as the visualization tools themselves. Special librarians have always conducted data analysis and competitive intelligence for their companies. Data visualization is an extension of this service to their user community.

Information professionals also use data visualization themselves for innovative ways to present information to users, to market their services, and to synthesize data for stakeholders.

The saying “A picture is worth a thousand words” holds true in data analysis because our brains are naturally wired to process visual information quickly and efficiently.

Instead of sifting through raw numbers, visual representations such as graphs, charts, and plots help us absorb information at a glance.

Through visualization, we can easily identify patterns, trends, and outliers through thoughtful visual encoding that might otherwise be overlooked in tabular data. It allows us to connect the dots between different variables, uncover hidden relationships, and gain meaningful insights.

Additionally, visualization is a valuable tool for sharing findings, as it makes complex data more accessible and engaging for others.

Whether presenting data to a technical audience or a general audience, clear and well-designed visuals help communicate insights effectively and capture attention.

In this lecture, we will explore:

• The journey from raw data to actionable insights
• How to structure and interpret data effectively
• Turning insights into meaningful decisions

Please note that throughout this lecture, when I use the term “data,” I am using it as a broad, umbrella term. In this context, it explicitly includes textual data (such as documents, social media posts, articles, transcripts, and any other forms of text) that we analyze using IR, NLP, or text mining techniques.

Why Create Visualizations Generally?

Most information visualizations are acts of interpretation masquerading as presentation. In other words, they are images that act as if they are just showing us what is, but in actuality, they are arguments made in graphical form (Johanna Drucker)

We create visualizations to answer questions and sometimes even to discover new questions. Visualizations help us make decisions, see data in context, expand our memory, support graphical reasoning, identify patterns, present arguments, tell stories, and even inspire action.

Visualizations are not just decorative. They are cognitive tools. They help us process complex information more efficiently than raw tables or text ever could.

Some goals are about attention and beauty — getting people to read, attracting different types of learners, and making the work engaging. Other goals focus on understanding — helping people explain, interpret, and explore data. And finally, some goals move toward implication — proving a point, correcting misconceptions, evoking emotion, or encouraging action.

This brings us to the quote on this slide by Johanna Drucker:

“Most information visualizations are acts of interpretation masquerading as presentation. In other words, they are images that act as if they are just showing us what is, but in actuality, they are arguments made in graphical form.”

This statement reminds us that visualizations are not neutral. They are not simply transparent windows into data. Every choice we make — what data to include, what scale to use, what colors to apply, what to highlight, what to omit — reflects interpretation.

A visualization might appear objective, but it always carries assumptions, framing decisions, and design choices. In that sense, every visualization is an argument. It persuades the viewer to see the data in a certain way.

As creators of visualizations, especially in academic and research settings, we must be mindful of this. We are not just presenting data — we are shaping how others understand it.

So as we move forward in this course, I want you to think critically about both sides: How can visualizations help us understand and communicate complex information? And how might they also influence, frame, or even bias interpretation?

Why Create Visualizations Generally?

Again data visualization is not isolated. It connects rigorous analysis with thoughtful communication, balancing discovery and storytelling depending on your purpose and audience. This slide illustrates the relationship between Data Analysis, Data Visualization, and Information Design, and how they exist along a spectrum from exploratory to explanatory work.

On the left side, we have Data Analysis. This is primarily exploratory. Here, the goal is to investigate the data, test hypotheses, discover patterns, and identify insights. The audience is often the analyst themselves or a small research team. The focus is on depth, accuracy, and discovery rather than presentation polish.

In the center, we see Data Visualization. This is the bridge between analysis and design. Visualization serves both exploratory and explanatory purposes. During exploration, visualizations help analysts see patterns, trends, and anomalies more clearly. During explanation, those same or refined visualizations are used to communicate findings to others.

On the right side, we have Information Design, which leans more toward explanatory work. Here, the focus shifts to communication. The goal is not just to analyze data, but to present it in a clear, engaging, and accessible way for a broader audience. This is where storytelling, layout, typography, and aesthetics play a stronger role.

The overlap between Data Visualization and Information Design is where we often see infographics. Infographics prioritize explanation and communication, sometimes simplifying or summarizing analysis to make it digestible and compelling for a general audience.

The arrow at the top emphasizes that visualization exists on a continuum. Some visualizations are created for exploration — messy, iterative, and analytical. Others are carefully designed for explanation — polished, curated, and persuasive.

Understanding where your work sits on this spectrum is important. Are you exploring data to discover insights? Or are you explaining insights to others? The answer affects your design choices, level of detail, and presentation style.

We are Wired for Visualization

Humans are biologically wired for visual processing where good visualization design works with human perception — using pre-attentive features to guide attention, reveal patterns, and make complex information easier to understand.

Research shows that a substantial portion of the human brain is devoted to processing visual information. Our visual system can process enormous amounts of data extremely quickly — estimates suggest up to 10 million bits per second. That means we are naturally optimized to detect patterns, differences, movement, and contrast almost instantly.

This biological wiring explains why visualizations are so powerful. When we encode data visually, we are tapping into one of the brain’s strongest processing systems.

The second image on the slide highlights something called pre-attentive processing. This refers to the brain’s ability to rapidly detect certain visual features before we consciously focus our attention.

Notice the small grid examples. In each case, one element “pops out” immediately — perhaps because of a different color hue, brightness, orientation, texture, position, size, or shape. You do not have to scan carefully to find it. Your brain detects it almost instantly.

These features — color, orientation, texture, alignment, brightness, saturation, size, and shape — are known as pre-attentive attributes. They are processed automatically and quickly by our visual system.

This is why thoughtful encoding is so powerful. When we use visual attributes strategically, we reduce cognitive effort and guide attention intentionally. Instead of forcing someone to read through numbers or text, we allow the brain to recognize differences instantly.

However, this also means that poor design can mislead attention. If everything is highlighted, nothing stands out. If color is used inconsistently, it creates confusion rather than clarity.

So when we say “we are wired for visualization,” we mean that effective visualizations align with how the brain naturally processes information. They leverage perception instead of fighting against it.

How Encoding Works?

Guess how many ‘7s’ there are in this set

Now, lets explore a very important concept in visualization: encoding.

On this slide, you see a large grid of numbers, and I ask you: How many 7s are there?

At first glance, this feels difficult. The numbers all look the same. They are the same color, same size, same weight. Nothing stands out visually. Your brain has to scan each character individually, which takes time and effort.

How Encoding Works?

Effective visualizations require thoughtful encoding

Design decisions have a big impact on what people will see in the data

Now, try guessing again.

They are the same set of numbers but now the 7’s pop out to us.

Now look at the this image.

It is the exact same set of numbers. Nothing has changed in the data itself. But this time, the 7s are highlighted in a different color.

Suddenly, the task becomes much easier. The 7s “pop out” to us. We can estimate or count them much more quickly.

What changed? The encoding changed.

Encoding refers to how data is visually represented — through color, size, position, shape, contrast, or other visual attributes. In the second image, color was used as a visual encoding to differentiate the 7s from the other numbers.

This simple change demonstrates a powerful principle: Effective visualizations require thoughtful encoding.

Design decisions — such as which colors to use, what to highlight, how to group elements, or what to make prominent — dramatically influence what people notice and how quickly they understand the information.

The data did not change. Only the visual encoding changed. Yet our perception changed significantly.

This example also reinforces an earlier idea: visualizations are not neutral. The way we encode data shapes what viewers see first, what stands out, and what might be overlooked.

So when creating visualizations, we must carefully consider: - What do we want the viewer to notice? - What comparisons should be easy? - What patterns should stand out?

Good encoding reduces cognitive load and guides attention intentionally. Encoding is the bridge between raw data and human perception. The choices we make in encoding determine how effectively our audience can interpret and understand the data.

Some attributes pop out more than others

Now, effective visualizations leverage the most perceptually powerful attributes for the most important information, guiding attention naturally and efficiently. This slide builds on our discussion of pre-attentive processing and highlights an important idea: not all visual attributes are equally powerful.

On the right, you see a table of visual attributes — things like line length, 2D position, orientation, line width, size, shape, hue, and intensity. Some of these attributes are better suited for representing quantitative data, while others are more effective for representing qualitative or categorical differences.

For example, 2D position and line length are very strong for quantitative comparisons. Our visual system is very good at judging position along a common scale and comparing lengths. That’s why bar charts and scatterplots are so effective for numerical data. On the other hand, attributes like shape, hue, enclosure, or added marks are often better for distinguishing categories rather than precise quantities.

Now look at the left image and ask yourself: What is easier to distinguish here, color differences or shape differences?

In most cases, color differences pop out faster than shape differences. Your brain can instantly detect a change in hue without conscious effort. Shape differences, while still noticeable, often require slightly more attention to process.

This demonstrates an important design principle:

• When something is important, use stronger visual attributes

• When something is secondary, use subtler ones

If we misuse attributes, for example using shape to encode subtle quantitative differences, we make interpretation harder than it needs to be.

The key takeaway is that visual attributes have a hierarchy of perceptual strength. Some attributes grab attention immediately; others require more deliberate processing.

As visualization designers, we should choose attributes intentionally based on:

• The type of data (quantitative vs qualitative)

• The importance of the information

• And how quickly we want viewers to notice it

Types of Visualizations

Data visualization comes in various forms, each serving a specific purpose depending on the type of data being analyzed and the insights we seek. There are many types of visualization that you can choose from once you know more about your data after the EDA process.

Some of the most common ones are:

1. Histograms display the distribution of numerical data, helping us understand the frequency and spread of values.

2. Bar charts are useful for comparing categorical data, showing differences between groups or trends over time.

3. Pie charts are commonly used for showing proportions, though they are best for simple comparisons rather than detailed analysis.

4. Line charts are effective for tracking changes and trends, especially in time-series data.

5. Scatter plots help visualize relationships between two numerical variables, making it easier to identify correlations and patterns.

6. More advanced visualizations include heatmaps, which use color intensity to represent data values and highlight patterns across a matrix, and box plots, which summarize data distribution, outliers, and variability.

7. Tree maps and word clouds are great for hierarchical and text-based data, respectively.

Choosing the right visualization depends on the data and the story we want to tell, as each type enhances our ability to interpret information and make informed decisions.

One-Variable Graph

• Histogram shows distribution and concentration of data

• Bar Plot compares the same variable across different groups and shows how the data distributed among the groups

• Pie Chart shows the groups of what your data is made, allows you to see the most and least as well as everything in-between, and shows the distribution of the data among groups

One-variable visualization, also known as univariate visualization, focuses on analyzing a single variable to understand its distribution, central tendency, and spread. A common method for visualizing one-variable data is a histogram, which groups numerical data into bins and displays the frequency of values within each bin. This helps us see patterns such as whether the data is normally distributed, skewed, or contains outliers.

Bar charts are another option when dealing with categorical data, showing the frequency of different categories using bars of varying heights.

Using these visualizations, we can quickly interpret key characteristics of a dataset, such as its shape, spread, and any anomalies, making them essential tools for exploratory data analysis.

Two-Variable Graph

• Scatter Plot shows all the data on the plot, you can see the distribution, correlations, and the spread of the data between two variables

• Line Plot similar to scatter plot but the points are connected, make it easier to see the trends and evolution

• 2D Histogram shows distribution of two variables related to each other

• Box or Whisker Plot shows spread of data and statistical information

Two-variable visualization, also known as bivariate visualization, is used to explore the relationship between two different variables. The choice of visualization depends on the type of data—whether both variables are numerical, categorical, or a mix of both.

One of the most common ways to visualize two numerical variables is a scatter plot, which displays individual data points on a graph with one variable on the x-axis and the other on the y-axis. This helps identify correlations, trends, or clusters in the data. For example, a scatter plot can show whether an increase in one variable leads to an increase or decrease in another.

Another effective visualization is a box plot (or box-and-whisker plot), which provides a summary of the data’s median, quartiles, and potential outliers. It is particularly useful for detecting variability and spotting extreme values.

By using these techniques, we can better understand how two variables interact, uncover trends, and gain deeper insights from the data.

Three Variable Graph

• Heatmap shows two variables and other quantity (amount, intensity, height) with a colormap

• Multiple-variable bar plot shows multiple variables for multiple groups

• 3D Plot shows the 3rd variable on surface in 3D plot

For high-dimensional datasets, applying the t-Distributed Stochastic Neighbor Embedding (t-SNE) algorithm and alike to reduce the dimension.

When we analyze data, sometimes we need to explore the relationship between three different variables simultaneously. This is called multivariate visualization, and it helps us uncover more complex patterns and interactions that might not be visible with just one or two variables.

One common method for visualizing three variables is a bubble chart. A bubble chart is similar to a scatter plot, where two numerical variables are plotted on the x and y axes, but a third variable is represented by the size of the bubbles. For example, if we were analyzing cities, the x-axis could represent population, the y-axis could represent average income, and the bubble size could represent the crime rate.

Another useful visualization is the 3D scatter plot, which extends a regular scatter plot by adding a third axis (z-axis). This is helpful when we want to see how three numerical variables interact, but it can sometimes be difficult to interpret without interactive tools.

When both variables are categorical, a stacked bar chart or a heatmap is useful. These visualizations display how frequently different categories appear together, helping identify patterns and relationships in the data. For a combination of categorical and numerical variables, a stacked bar chart can be effective. Here, the x-axis might represent categories (e.g., different product types), the y-axis could show total sales, and the bars can be divided into segments representing a third categorical variable, such as different regions.

Using three-variable visualizations allows us to gain deeper insights into data relationships, helping us make more informed decisions and detect more complex trends.

Some Examples

Visualization is flexible and adaptable. The form should follow the data and the analytical question. Different types of data require different visual strategies. This slide shows a range of examples to highlight the diversity of data visualizations and the many ways data can be represented visually.

Starting in the top left, we see a visualization of a large image collection arranged in a spatial layout. This type of visualization allows users to explore patterns within a collection, perhaps by similarity, metadata, or thematic clustering. It emphasizes exploration and interaction, helping users navigate complex datasets visually.

In the top right, we see a more abstract network-style visualization. This type of visualization often represents relationships—between people, ideas, documents, or concepts. The flowing lines and clustered forms suggest connections and structure that might not be obvious in tabular data. These visualizations are particularly useful for understanding relational data and structural patterns.

In the bottom left, we see a social network visualization of tweets. Nodes represent users, and connections represent interactions, such as mentions or retweets. The size and color of nodes may encode frequency or influence. This type of visualization helps reveal communities, central actors, and communication dynamics within social media networks.

In the bottom right, we see a geographic network map. This visualization combines spatial mapping with relational data, showing connections across regions. It demonstrates how location can be encoded along with relationships and temporal trends. Geographic visualizations are especially powerful when analyzing movement, diffusion, or communication across space.

Together, these examples show that data visualization is not limited to bar charts or line graphs. Depending on the type of data—images, networks, text, geography, time—we can use different visual forms to reveal patterns, relationships, clusters, and trends.

Each visualization makes specific design choices about encoding: position, color, size, shape, density, and connectivity. These choices determine what stands out and how the viewer interprets the information.

The readings for this week, the supplementary presentation, and the additional links included in this week’s module provide many more examples of different types of visualizations particularly for text visualization. Please take some time to explore these resources at your own pace to broaden your understanding of the different types of visualizations and how they can be applied effectively.

As you move forward in this course, think critically about which visualization type best fits your data and your analytical goals.

Ecosystem of Data Visualization Tools

Visualization ecosystem is diverse. Understanding the trade-off between flexibility and ease of use will help you make informed decisions in your own projects. This slide gives us an overview of the ecosystem of data visualization tools and how they compare along two important dimensions: ease of learning and flexibility.

On the horizontal axis, we move from easy to learn on the left to harder to learn on the right. On the vertical axis, we move from not flexible at the bottom to highly flexible at the top.

The slide also distinguishes between two broad categories:

• Apps (shown in red)

• Charting libraries (shown in blue)

Let’s start with apps. Tools like Google Sheets, Excel, Tableau, Plotly, and Illustrator are generally easier to learn. Many of them have graphical interfaces and require little or no coding. They are excellent for quick visualizations, dashboards, and business reporting. However, they may be less flexible when you need highly customized or complex visualizations.

Now, looking at charting libraries. Libraries such as Matplotlib, Seaborn, Bokeh, ggplot2, Vega, and D3 require programming knowledge, so they tend to be harder to learn. However, they are much more flexible. With these tools, you can customize nearly every aspect of a visualization, build interactive systems, and integrate visualizations into larger applications. At the extreme end, tools like D3 are highly flexible but also require significant effort and technical expertise.

Some tools fall in the middle. For example, Vega-Lite and Highcharts attempt to balance flexibility and usability. They provide more structure than low-level libraries but more control than simple apps.

The key takeaway is that there is no single “best” visualization tool. The right tool depends on:

• Your technical background,

• The complexity of your project,

• The level of customization required,

• And the intended audience.

If you need quick exploratory visuals, an app may be sufficient. If you are building interactive, research-grade, or web-based visualizations, a programming library may be more appropriate.

Python Visualization Landscape

This slides shows the popular visualization libraries names in Python. We will use Pandas and Altair packages in this week’s lab assignment.

Altair is a wrapper around a Javascript library. Vega-Lite, which itself is a wrapper around another library called Vega. If you have heard of D3.js, then Vega is an alternative library for making a visualization grammar. If you are curious about Vega, you can look at their simple bar chart tutorial here: https://vega.github.io/vega/tutorials/bar-chart/.

The biggest benefit of Altair being a wrapper is that you can download the compiled vega and drop it into any html file to publish your graph to the web (more info on that is available here https://vega.github.io/vega-lite/tutorials/getting_started.html#embed). This is super exciting because D3 has a very steep learning curve, and other visualization libraries like ggplot2 are not great for embedding interactive versions.

Data Visualization Resources

This slide and the next highlight several important resources to help you get started with data visualization. I encourage you to explore them at your own pace.

Data Visualization Resources