Across all industries, business users are learning the value of their raw text. By mining this data, they can save operational costs, uncover relationships previously not available, and gain insights into future trends.
Is it hard to believe that 80 percent of business data is in the form of text?
Examples include call center transcripts, online reviews, customer surveys, and other text documents. This untapped text data is a gold mine waiting to be discovered. Text mining and analytics turn these untapped data sources from words to actions. However, to do so, each company needs to have the skillsets, infrastructure, and analytic mindset to adopt these cutting edge technologies. To better assess your ability to embrace text mining solutions, Zencos has developed a self-evaluation analytics checklist to assess your companies readiness to become analytically driven.
Data scientists analyze text using advanced data science techniques. The data from the text reveals customer sentiments toward subjects or unearths other insights.
There are two ways to use text analytics (also called text mining) or natural language processing (NLP) technology.
You can convert free-form text into structured data for use in predictive models or unearth hidden patterns in your data. With text mining, you can flag potential customers eligible for cross-selling, forecast customers’ sentiments, or understand behaviors that predict fraud.
There are so many ground-breaking ways that companies can integrate previously untapped data sources and NLP into their operations.
Here are a few examples that our data science staff suggested:
The good news is that many companies are already using text to drive business operations successfully. When moving a data strategy from business intelligence reporting into data science – text analytics can be a way for you to optimize your processes.
Let’s look at a few examples that may highlight a chance for your company to implement text analytics.
There are plenty of stats that can tell you consumers are interested in other’s opinions and experiences. These statistics reveal that at least 90% of us are influenced by what we read. Even more so if it’s a negative review – the sentiment resonates. In recent years, multiple sites have collected reviews for local eateries, vacation destinations, and, of course, consumer products.
A positive review is a form of social proof.
If your company is considering entering a new market or needs to research product ideas, why not start with online reviews from real users? This was the very idea that drove a major Amazon case study on pricing by a young analyst team.
They wanted to understand the best speakers available to purchase at the $150 price point. They theorized that if you’re going to develop and market a new product, it is useful to understand what features are most valued. What a great business use case and a reasonable example of analytics!
The team extracted data for five speakers based on popular brands that Amazon customers had reviewed. The data contained consumer ranking, price, and all customer reviews.
This data was a mix of structured data (ratings, price) and unstructured data (review text).
Using the customer rating, these junior data scientists wanted to learn which product characteristics influenced scores. The following figure shows the products with the final text topic extraction analysis.
It makes sense that a speaker’s most outstanding quality should be its sound quality. There are many choices in this market space. Within the target price range, consumers must choose the most valuable features.
By reviewing low-rating topics and considering the reviewers’ sentiment toward that topic, you learn which features are essential. Its battery life, speaker material, and a charge port.
The business analysis paid off! But more importantly, the business is able to under the customer experience.
When starting your speaker re-design or even a marketing campaign, you understand what features are essential to consumers. This market research from this data could have been expanded to include multiple sites or even all products. It is vital to understand what features drive purchasing decisions and leads to the most product dissatisfaction.
People love the Fitbit because they can quickly get performance feedback along with a coach from a single device. When they are not so happy with the invention, they turn to Twitter and talk to the brand’s customer support team. As the company assesses the tweets over time patterns emerge as well as the customer’s attitude or sentiment. No one expects a happy customer when they are having issues, but it can be positive if the support is quick and useful.
In this social media customer study, the analyst pulled tweets consumers sent to @FITBITSUPPORT. In one six-month period, there were over 33,000 posts. Fitbit is popular! That is too much data for one person to digest and attempt to identify all of the trends.
However, with text analytic applications, the analyst was able to break out the tweets by model and then zero-in on specific issues. For the Fitbit Charge HR, the strap was an issue. It would break, bubbles would develop in the band, and the rubber stamp would peel off. The Fitbit Blaze had problems with the operating system where it could not get past the logo screen.
Within a few short mouse clicks, the user comments are available.
Information Extract from Twitter
The product team knows which features are annoying the customer and where to focus their energy.
Understanding the customer experience is essential and these online reviews provide a reliable way to understand it.
It’s basically the words right out of their fingertips. If you are an after-market company, then you might see an opportunity to supply bands.
Just as customers share information when they tweet, your competitors expose information about themselves when they report to public databases. There are several instances where the US government has mandated information. Subsequently, the data is published and made available to all.
Zencos worked with a medical device manufacturer using FDA reports as a source. This data set contains several hundred thousand medical device reports of suspected device-associated deaths, acute injuries, and malfunctions. Our team was able to use the database to match the companies to their products. Then we used the text fields to understand the main issues with some of the company’s devices. Some detective work was required, but the result was fruitful.
A deep dive into the text data revealed that the customer had a much higher placement success rate than one of their leading competitors.
Using topic cluster analysis, we demonstrated the common causes of failed device installations by our client’s competitor that led to many patient deaths. Text clustering and sentiment analysis allowed us to find common problems with very adverse outcomes for many devices. Products such as SAS Visual Text Analytics contains sophisticated text mining algorithms.
Text analysis provides valuable insights into the customer’s own malfunctioning devices and allows comparison of the device’s performance with others in the marketplace. It may not have been apparent before the analysis, but now you have a glimpse into market penetration.
Plus you understand the common issues your competitors have.
Other public data, such as that in the Consumer Financial Protection Bureau, reports on what consumers find annoying about financial institutions. Yes, you can retrieve that text to see your rankings – but what a great way to spy on your competitors.
Government researchers concerned about vaccine safety wanted to understand the adverse reactions. When there are tens of thousands of events that occur, it is difficult for an analyst to understand what might be the most important. It’s more of a challenge when the data is unstructured, free-form text.
Text analytics simplifies the process by allowing the researchers to consider patients who reacted to vaccinations and were taking additional medications. Then the researchers used text analytics to identify the most severe reactions.
When their predictions included text data sources, the accuracy rose to 80 percent of the time versus 40 percent when they excluded the text data.
This approach provided significantly better results than the data alone offered. Researchers can save many lives when they can quickly understand the relationships and the causes by having the text data available.
Human trafficking affects over 40 million people each year. It is a devastating crime that reaches vulnerable populations, such as children. Many have a strong desire to see this crime stopped. One of those is Tom Sabo, who was very disturbed by this issue after attending a human trafficking symposium. He wanted to put his text mining skills into action.
Using text analytics and AI, he was able to create useful models that law enforcement could use. One model pulled together several text-based data sources: police reports, newspaper articles, recent prosecutions, and a shady classified advertising website. The goal was to find relevant patterns in the text that Sabo could then incorporate into a predictive analytics model.
Sabo used the police statements from a specific New York jurisdiction. He then linked this data to other events outside of the state and even outside of the country.
He saw these trends related not only to who was involved, but also, where the events were happening.
This model was subsequently used to identify these situations quicker and allowed law enforcement to act.
With the growth and availability of unstructured text data, companies have tremendous opportunities before them. But having a desire to embrace text mining and predictive analytics is not enough. You need to first understand where you are as a company analytically, and you need to create a plan for how to embrace these new opportunities. Understanding where you stand currently will help identify what your next step should be and prevent you from biting off more than you can chew.
If you have any questions about how your company can use your free-form text, please reach out. We’d love to hear from you!
