Lior Goren Ruck
Senior Data Product Manager @ Yotpo
March 14th, 2018

How Customer Reviews Can Be Used for Opinion Mining

Analyzing customer feedback at scale through opinion mining is the next frontier for customer-driven brands.

Table Of Contents

Human communication involves a multitude of emotion.

Whether you’re writing a comment on a Facebook post or a text message to your boss, your words convey your opinions and attitudes about any number of topics.

Imagine using a computer program to skim the texts your mom sends you. Instead of reading each one to find out what she wants to tell you, it would give you a quick summary:

She is unhappy about the state of your room, your academics, and your love life.

This is essentially what opinion mining does. And it can be used for everything from gauging your mother’s disappointment to assessing public sentiment toward a presidential candidate based on social media mentions.

Opinion mining uses Natural Language Processing to identify a range of opinions about topics in a given pool of text. These opinions are scored along a scale of positive to negative using sentiment analysis technology, making it easy to determine trends in attitude and mood.

Opinion mining and customer reviews

Thanks to its efficiency in extracting sentiment trends from massive amounts of text, opinion mining creates the unique opportunity to analyze customer reviews at scale.

Unlike surveys, which tend to influence responses with targeted questions, reviews provide businesses with a core of organic reactions that authentically reflect customer opinions. Free text means shoppers describe things in their own words, enabling them to raise unexpected issues that brands may not have considered. For example, a beauty retailer that uses opinion mining to analyze its reviews could quickly learn that their best-selling eye shadow has a trend of negative sentiment around the topic of scent. They could also dig further to find out the specific problem, like a scent that is “too strong” or “too sweet.”

With the help of opinion mining, retailers can easily find out what their customers love and hate about their products and service, even if they’re receiving several thousand reviews each month.

How opinion mining & sentiment analysis work

Since opinion mining and customer reviews are such a natural pair, it was only a matter of time before Yotpo’s Data Science team got to work to see what consumer trends they could find. They used NLP to extract topics from the reviews and developed deep learning technology to train their own sentiment analysis model on the opinions expressed. You can take a look at some of the fashion industry findings they uncovered here.

I worked together with our Data Science team leader, Tal Perri, as his team identified 1 million topics and 75 million related opinions in our review database alone.

Just defining an “opinion” required several iterations.

Tal’s team trained the technology on more than 30 million reviews to hone its ability to accurately identify opinions and topics and to group them by similarity of meaning. For example, the words “shipping,” “shipment,” and “delivery” would form a single topic. This allows for more opinions to be tallied and more statistically significant samples per topic.

The team then used sentiment analysis technology to score each topic and opinion on a scale of -100 (the most negative) to +100 (the most positive).

opinion mining and sentiment analysis

Sentiment analysis is designed to distinguish between conflicting sentiments about different topics within the same review. For example: “Great product, but slow shipping.”

Thanks to painstakingly crafted rules, it also has the capacity to sort through complex and contradictory human styles of writing, like sarcasm.

For example, it can tell that this sentence expresses negative sentiment:

“Top-notch delivery service — still waiting on my package four months later.”

And that this one is positive in tone:

“Top-notch delivery service — got my package yesterday!”

Extracting topics and opinions from reviews

Data science and deep learning aside, the really impressive thing about the team’s findings was the sheer speed and accuracy (92%!) with which their algorithms could identify trends in sentiment.

As any busy business owner knows, there are about a million things to do before you can even dream of sifting through customer reviews. Concerns over fulfilment, personnel, product development, suppliers, budgeting, and more make it near impossible to find the time.

“When we were done, we needed to evaluate the accuracy of our model,” Tal said. “So, we asked our professional services (manual moderation) team to take a corpus of reviews and start manually extracting opinions and topics.

“They analyzed a few hundred reviews over 3 or 4 days before they came to us and said there was literally no way they could do it anymore… and that’s even without tracking trends.”

When he gave them the script his team created, it took just a few hours to run the analysis on all the reviews.

“If you have ten reviews per month, you can go over them manually, but if you have thousands, it’s almost impossible,” Tal said.


Customer reviews are directly tied to your product catalog, often include valuable feedback on service, and they come from verified customers who have first-hand experience with your brand. They are, in other words, the perfect place to look for a huge range of customer-initiated reactions to your products and business as a whole.

But, without the tools to comb through them for trends at scale, it’s easy to miss important feedback from your customers. While relying on star ratings may seem like a quick solution to analyzing mounds of reviews, it won’t give you the whole picture.

Reviews aren’t black and white. A five-star review can contain important requests for improved delivery time, while a one-star review mistakenly written off as “negative” may contain plenty of helpful details that can entice customers to buy.

It’s rare that a customer’s experience is wholly positive or entirely negative, so while star ratings give you an idea of customer satisfaction at a glance, brands would be remiss not to dig deeper with the help of opinion mining.

Find Out How Yotpo
Can Help You Grow

Join The World's
Fastest-Growing Brands

Interested in Yotpo?
Schedule a call with one of our marketing consultants to learn more.
Thank you!
We'll be in touch in no time! In the meantime, take a look at what our customers are saying about Yotpo.
Yotpo Success Stories >
Yotpo Customers
Trusted by the worlds
fastest-growing brands
Yotpo Customers