Data Segmentation

Data Segmentation

In this guide, you’ll learn about an important part of advancing from social listening to reaping the benefits of digital consumer intelligence. Without effective data segmentation, bringing audience insights into the heart of your decision making just isn’t going to happen. We’ll walk you through the different ways you can segment your data using Brandwatch tools to generate better insights for initiatives across your organization.

Below you’ll find examples and how-to guides on:

• Approaching data segmentation
• Trend analysis: Diving into topics of interest, both known and emerging
• The different ways of segmenting your data, and when to use them (from keyword or filter-based rules to Custom Classifiers and machine learning)
• How to analyze and monitor standalone topics within multi-layered conversations 
• The different ways to segment an audience

Throughout this guide, Caitlin Jamali, Senior Solutions Strategist at Brandwatch will share best practices for market research (MR) and insights professionals on the different ways data can be segmented.

Caitlin is part of Brandwatch’s Strategy and Insights team, which is focused on understanding clients’ goals and objectives and synthesizing the right solutions to meet them. Whether it is putting together dashboards that address specific questions or scoping out a series of reports that will help clients get the information on an ongoing basis, the Brandwatch Strategy and Insights team works on mapping out that plan for clients.

Let’s start with an example. The social conversation related to the pharmaceutical industry is huge, and that’s why segmentation is extremely important. A quick glance at the data in Brandwatch Consumer Research showed that in the period between April 9 and May 10 2021 alone there were over a million mentions and 546k unique authors who published pharma-related content online.

“There’s so much data that it can be overwhelming,” Caitlin says. This is where data segmentation comes in – it’s about making sense of all the information.

For instance, by segmenting the data by sentiment, you can learn about the brands generating the most positive and negative conversation.

By segmenting the data by topics, you could start to understand the major drivers of interest.

(And, by combining sentiment analysis with topic analysis, whether they’re positive or negative.)

In fact, there are officially over 50,000 ways to segment and visualize your data using Brandwatch Consumer Research. In reality, the combinations are infinite when you can create your own categories!

So, how should an analyst get started with segmentation? Whether you are looking to get a landscape view of your industry or to discover what people are saying about a niche product or ingredient, the first step is always to have a clear question in mind.

We’ll review this approach in practice in the next section, using an example of a brand in the anti-aging market. 

Trend analysis: Diving into topics of interest, both known and emerging

Let’s say a beauty company’s marketing analysts are interested in looking at the anti-aging market specifically to help inform new campaigns. Their research question might be: What topics are trending in the anti-aging conversation?

To approach this, first, the team would create a query to gather mentions about the industry. This is essentially where data segmentation starts – it defines the segment of all of the data available that you want to analyze further. 

To get a sense of a broader landscape, the team would look into the volume of mentions over time. Brandwatch’s AI-assistant Iris makes it easier to help pick out significant peaks in conversations and identifies what drove them, allowing analysts to find trends faster.

As you can see from this example above, Iris detected three major spikes in conversations that would be worth exploring. The researcher could click on each peak to view the mentions or key drivers and then use this information to segment data further going forward. This would enable them to track new trends or even get alerted when those trends see an influx of mentions or positive or negative sentiment.

Another way to identify what’s trending in the conversation would be to look at the topic cloud with the most-mentioned topics within a given period of time. As shown in the word cloud below, the word Botox is prominently displayed as one of the frequently used words in conversations about anti-aging products and solutions.

In this case, the team could create a rule to segment the data and look further into the Botox conversation, both historically and as it develops. 

An additional way to explore insights around your brand, topic, or industry is by doing sentiment analysis and digging into the positive and negative discussion drivers. Any recurring themes discovered through this broad market analysis are worth monitoring on a regular basis to help protect your brand from a misstep or to address pain points. 

Of course, there will be some trends that analysts will be particularly interested to explore, regardless of whether they’re already popping up in conversations. Using their own expertise in the industry, they could easily build rules to segment the anti-aging conversation by their predefined topics of interest. By combining this method with the above, researchers can segment out both expected and unexpected trends in the data. 

Essentially, you can take an exploratory approach to data segmentation or a much more focused one when you know exactly what you are looking for. The best research is a healthy combination of these two approaches – segmenting the data by topics you’d expect to see the conversation around, but also allowing for segmentation around what’s emerging in the data set organically.

The example above is for a general look at trends within an industry, but regardless of what you’re looking at (no matter how niche), defined research questions will help guide you and segment the data in the most appropriate way for your research objectives.

Practical application: The different ways of segmenting your data and when to use them

Within the Brandwatch Consumer Research platform, a wide range of features can help you segment your research. 

Rules and filters

For example, you can choose to filter by a specific channel like forums and learn about what people are saying regarding your topic in a smaller setting. This is great if you want to understand how different site users discuss different topics, and you could do this with Twitter data too, but at the end of the day, there will likely still be a lot of data to go through.

There are lots of other ways to filter your data that could lead to more interesting insights. One might be to segment by sub-topics and to use rules to do so.

Use case:

Let’s return to the Botox trend we mentioned above. If researchers wanted to find out what was driving people to decide to get Botox, they could look into the existing dashboards, read through the content, look at word clouds, and review mentions bubbling up in the data set that could potentially become emerging trends. If something is coming up frequently enough in the data, the team could build out rules specifically around those findings to segment the data even further and discover the conversation drivers. 

The example below analyzes the common adjectives associated with both Botox and fillers (another injectable anti-aging treatment). 

When you have an example like this of neatly defined categories that are easy to search for with keywords, rules are a great way to go. Creating a category for body mentions, for example, an analyst could simply write rules around foreheads, lips, etc and bucket the conversation to be analyzed in a dashboard. These rules can be applied historically as well as going forward.

Custom Classifiers and Machine Learning

Creating rules is one way to segment data, but there’s another way to go when you’re trying to segment by a topic that isn’t easily defined using keywords. 

Brand sentiment is one of the most valuable types of analysis a brand can conduct to discover exactly how the public or a target audience feels about something at any moment in time. 

Let’s think about sentiment for a second. Someone might say on social media, “Ahhh, I have such a headache. I’m gonna grab some Ibuprofen.” Automatic sentiment segmentation would probably mark this comment as negative, but it’s actually not negative about Ibuprofen. It’s only negative at a post level, but it’s not negative about the brand. 

Custom Classifiers introduce an opportunity for researchers and analysts to go in and train the platform to recognize mentions they would consider to be positive or negative based on the language people use around their specific product, brand, or feature. This differs from rules-based segmentation in that the analyst doesn’t need to write out any strings of keywords. Instead, they can drag and drop relevant posts into their chosen category.

For drug companies, this kind of analysis could help them break down brand mentions by:

• Positive or negative comments on brand/marketing
• Positive or negative comments on experiences with products (taste/texture etc)
• Positive or negative comments about corporate brand ('evil drug company')
• Positive or negative comments about price

Analysts could then understand more about the sentiment around each part of the conversation and make useful distinctions when sharing insights with the organization. Negativity may be increasing, for example, and being able to pinpoint exactly what is driving that negativity will be hugely useful in a crisis. It might also be useful to filter out mentions from a particular segment to get to deeper insights in other categories.

Powered by machine learning, Custom Classifiers can help segment data into categories without the need to write rules around them. It’s particularly useful when a desired category is difficult to define with keywords, such as purchase intent or the different aspects of your organization as a whole.

The opportunities with Custom Classifiers are virtually limitless: intent, purchase phase, post-purchase perceptions, how attitudes and needs are changing over time, brand health, and many other metrics or issues can be measured with this method.

How can analysts monitor standalone topics within conversations? 

Now that you’ve learned how to segment the data, what can you do with it? Let’s continue with our example around anti-aging.

As we reviewed the topic cloud for the anti-aging conversation this year, we noticed that one of the themes seen in discussions was ‘fillers’. There are many different longer-lasting anti-aging treatments, and fillers are frequently viewed as an alternative solution to Botox. 

If researchers wanted to learn more about fillers in conversations related to the anti-aging market, they could zoom in on this topic by creating a tag and building a rule around it. 

Once the tag is created, a researcher could review the data using the topic cluster component filtered by fillers and watch consumer stories emerge. As you can see from the screenshot below, fillers are connected with several different topics such as ‘40s’, ‘worry’, and ‘obsessed’. Each of these connections could potentially enrich the ongoing research with fascinating details regarding the use of fillers and perceptions around them. 

If the team wanted to get really granular about people discussing fillers, they could select additional filters such as the location of the users, author gender, sentiment, etc when building out the rule. Brandwatch Consumer Research allows a great level of customization here.

Audience segmentation: Getting deeper insights about your audience

Understanding the lens through which your customers see you is essential if you want to bring consumer needs to the heart of your decision making.

What’s a better way to learn about your customers than learning it firsthand, by understanding conversations happening among the specific groups of the population who talk about the topics your business is interested in?

A deep dive into audiences can help brands understand who their target audience is, what they are saying, and how they are feeling about a particular issue.

Segmenting audience through self-identification 

There are many different ways you can drill into the audience’s data using Brandwatch Consumer Research. For example, we could apply rules that would segment data by medical professionals by looking for people saying “I’m a healthcare worker”, “I’m a nurse”, or “I’m a doctor”.

While you might see a few people say: “As a healthcare worker I’ve noticed...” in different forums, not everyone self-identifies in posts with “I’m a healthcare worker and I think…”.

So while the data may be accurate, this approach to audience segmentation will likely deliver a smaller sample size. 

Segmenting audience using Social Panels

A better way to segment your audience is by using multiple factors, such as what people put in their profiles as well as what hashtags they are using or accounts they are following. This is where Social Panels come into play. 

Once we know this information about our audience, we are able to find them easier. It’s hugely impactful to be able to see a broader conversation as opposed to just looking at authors who plainly self-identify as part of a certain group. Social Panels can bring more volume and richer details to the data for MR experts to analyze.

The volume of conversation around Covid-19 in relation to the beauty industry is enormous. For example, one of the recent reports published by the Economist suggested that an increase in Zoom time triggered by the pandemic is fueling a boom in cosmetic surgery worldwide. The report talks about how video-conferencing and people staring at their own faces for hours made consumers very self-conscious about their appearances on the screen and in real-life, ultimately pushing them towards getting procedures. In fact, the pandemic has led to a 10% increase in cosmetic surgery in the US, and in France, cosmetic surgery is up by close to 20%.

This could be a great opportunity for MR experts working on analyzing the beauty sector to study new conversations around anti-aging solutions like injections and surgeries. There could be whole new audiences out there considering these different options, but they may not even be on the radar of researchers if no one is looking for them.

Social Panels is a very handy tool to help you slice the data and gain an understanding of exactly who your audience is, what they are posting about, and what they are saying they are concerned about. It provides an excellent way to pull out valuable insights the team can act upon.

Identifying influencers by segmenting your audience

Let’s imagine that a pharma company is looking to identify Twitter influencers to find someone they might want to partner with that would be interested to speak about or advocate on behalf of the brand. 

In this case, the research team would first define the target audience, then set up conditions in Social Panels to filter the audience according to the marketing team’s specific influencer criteria.

For example, if you were to look for Twitter users who are involved in the healthcare and beauty conversation, your filters might look like what’s shown below. Here, the user bio must contain one of the listed keywords.

This search could return a rather large volume of users. To narrow it down further, the team could also select “followers of” and list their company and its subsidiaries (if applicable), and perhaps even competitor accounts.You could also specifically be looking for people who talk about a particular product in a specific region.

You could do this by adding several additional filters such as “Tweets about” and “Author location”. In the case of Botox, the filters might look like this: