Using Pattern Recognition For Powerful Insights

Strategies for Using Custom Categories

When working with unstructured data, it’s crucial that analysts have a high degree of flexibility when creating categories because every business is different; each business has a unique market and defines, say, “purchase intent” differently. In broad-scope landscapes like social media platforms, the data has to be categorized in a way that’s relevant to the researcher’s question in hand. This is where custom categories come into play in social media analytics.

What are custom categories?

In our previous post about mining data from unstructured data, we defined custom categories as a Gmail filter.
“Fundamentally, filtering in Gmail is adding a label or tag to emails, so that it can count and group emails of the same kind together. If properly trained, Gmail’s Inbox classifies emails into user-built Topics like Promotions, Downloads, Invoices, or whatever you need to categorize. This model looks for patterns in the content of every email – such as keywords, phrases, authors – and assigns it to the most pertinent category; it doesn’t follow pre-defined parameters”
Through the creation of custom categories, machine-learning analysis provides a degree of flexibility and nuance that leads to greater relevance and depth of results. From industry analysis to branding analysis, machine learning can provide key social media insights to organizations.  

Two Approaches to Social Data Analysis

Rules-based vs. Pattern-based

A keyword-based approach to data analysis, an example of a rules-based method, only analyzes a limited scope of data because of its rules and constraints. When analyzing unstructured social data, a keyword-based approach isn’t enough to gain actionable business insights.
An example of a keyword-based approach is the Google Search Engine:

When you’re searching for information about a certain topic in Google, such as “what is social media analytics”, the results shown in the search pages usually contain the majority of the inserted keywords (and don’t provide information that doesn’t, but could still be relevant to the question), and this is why the traditional keyword-based approach can sometimes be limiting.
Unlike keyword-based tools, machine learning classifies examples of data by looking at patterns to measure and group the social conversation into relevant categories.
When we talk about a pattern-based approach, we’re referring to a method where a social media analytics tool is not told to categorize data through fixed, specific keywords in each category; instead, it is about training the tool by giving it a few examples of what you’re looking for in each custom category.
In the example above, we’ve analyzed public reaction on social media to a company’s logo redesign. The different categories shown, such as “Love the New Logo” and “Lack of Brand Recognition”, are created and customized by the researcher, and they describe the components behind the positive and negative sentiment.
In this case, the tool would be trained by picking a few posts such as “I really hate their new logo” and “someone needs to be fired for this logo” and dragging them to the “dislike new logo” category. Through the pattern-based approach, the tool can recognize this pattern, and crawl the remaining untrained posts and place them in their respective categories.
This solution provides the ability to measure the effectiveness of each category, and improve it by modifying the examples and the parameters in each category until it provides the most reliable results.
The ultimate goal is to crawl and provide insights through the pattern the tool has derived from given examples. Providing examples to the tool gives it more flexibility, and allows it to identify these patterns on its own, and automatically classify them into custom categories for the remaining data.

How to Start – Setting Up Custom Categories

Custom categories still require Boolean strings in the beginning to define the overall scope of the conversation. By including keywords and elements such as AND, OR, or NEAR, the researcher can narrow the conversation so that it’s relevant to the research question.
In the example of the company changing the logo, the Boolean string would use keywords such as “X” (company name) AND “logo”. After deciding the scope of the conversation, the tool can be trained to include only results related to the company’s logo.
Ultimately, machine learning allows analysts to group social posts into categories that are relevant to the strategic question at hand. By training machine-learning platforms and teaching them to focus on topics that matter and ignore irrelevant posts, the platform can provide a more profound level of insights.

Basic Frameworks of Custom Categories

In our publication about Gaining Business Value from Unstructured Data, we discussed the value of using custom categories. Depending on the business goal, the use of these categories varies. Although there are infinite frameworks researchers can work with, the following examples show how custom categories can help derive business insights.
In this section, we’ll dive into the three following frameworks:

  • Product or Service
      • Product qualities such as features, design, performance
        • What drives buyers to purchase or not the iPhone 7?
  • Brand
      • Brand values, marketing, customer service
        • How does safety play a role in Volvo’s brand reputation?
  • Unbranded
      • Sentiment around product category as a whole, industry research, trend analysis
        • What are the unique audience interests of greek yogurt consumers and how could this help inform brand strategy and messaging?

Framework: Product or Service
Custom categories are used to analyze social conversation around a particular product or service. In this category framework, basic sentiment around a product or service (such as positive, neutral, and negative) can be divided into several categories that will further drill down into the reason behind the sentiment.
This specific approach is applied when brands want to analyze the success of a new launch, the performance of a current product, or even their competitors’ offerings.
One of the most popular research queries is purchase intent. By creating a category that allows the tool to identify if there’s conversation around purchase intent in social media, companies can understand the potential of their product or service.
Another advantage is that the product development team can search for product feedback in social media to understand how to fix repeated issues and improve their offerings. By knowing what issues or complaints arise in social conversations, they are able to offer the product that customers will be most satisfied with.
A viral example of this is the iPhone 7 and the removal of the headphone jack. Researchers can use social listening to deep dive into this specific, and nuanced conversation to see what people are saying about Apple’s new product feature.
Another example of this product framework is the following analysis of social conversation around Bose’s product in the Beats vs. Bose Case Study.  


One common use case for social media analytics is for brand crisis monitoring. Think of Wells Fargo being sued for creating 2 million fake customer accounts to boost sales. By listening to social conversations, businesses like Wells Fargo can understand to what extent their illegal activity changed public perception of their brand.
In the following example from our guide designed to help agencies create data-driven pitches, we can learn the most prevalent topics of conversation around Warby Parker’s brand. In this case, the main topics were their social engagement, brand decision, and innovation.

Framework: Unbranded

Many organizations use social media analytics to search and understand the external environment surrounding their brand performance. For this approach, brands have to be more forward-thinking, looking for big ideas that stir business uncertainty. This ‘unbranded’ framework has no pre-set definition and it purely depends on what organizations are trying to find.

One of the use cases for this category framework is trend discovery within the industry. Many brands are diving into social media to learn what’s popular, and how they can incorporate that trend into their own offerings.
More specifically, understanding what consumers are looking for when referring to a general product or service can help inspire departments from product development to marketing.
Assume that your business doesn’t exist; what does the landscape look like without you? What external forces may impact your business? If you’re a major brewing company, you would stick to a research topic that is relevant to your business, such as monitoring conversation around emerging beer flavors.

If you were a financial company offering credit cards, wouldn’t you want to know what people say when they talk about credit cards? Wouldn’t it be critical to understand the driving factors for people who choose certain credit cards? In the following visual from the Social Insights Financial Services Trend Report, you can see which credit card rewards led the conversation.
The possibilities and results you can get from using custom categories are not limited by these previously outlined frameworks, though the above represent some common use cases using social data. Applying custom categories to the research or business question can provide enterprises the flexibility to work with data that would otherwise be very complex to handle, or tediously difficult to gain insights from. Custom categories, based in machine-learning, can provide valuable insights to different entities throughout the enterprise – whether on brand perception, purchase intent, competitors, industry-level trends, brand or topic audience, and many, many other areas.
Found this blog informative? Read our “Fundamentals of Social Media Analytics” to learn more.

Request a Demo

Ready to transform your business?

Get a walkthrough of Crimson Hexagon and learn how consumer insights can help you make better business decisions.