How to Apply Machine Learning to Customer Feedback

Data & Analytics

We are all aware of the benefits that a good Google review has for our business. On one hand, it allows a future client to have initial feedback about the company or product/service they want to acquire, and on the other hand, the more positive reviews we have, the better our SEO positioning will be.

Properly managing these opinions will increase our customers’ confidence and boost the company’s Google searches. As consumers, we demand a lot when it comes to evaluating our experience with brands, and Google understood the importance of quantifying this when it created the well-known “star” rating (between 1 and 5).

In order to enrich this feedback, Google also allows users to qualitatively describe the experience they have had, providing information which is of great value to brands.

Google My Business offers us various opportunities to access this customer feedback and we can work the data differently through Machine Learning to get the most out of the information.

Accessing the data

If we are a company, the first thing we’ll have to do in order to have more visits (both online and offline) is being on Google. Making our business visible to the rest of the world should be one of our very first priorities.

To make our lives easier, we have Google My Business. This is Google’s service focused on businesses, which helps us to gain more visibility, and lets us:

  • Improve our positioning in different search engines.
  • Appear in Google Maps with the exact location of the company.
  • Improve our online presence in the rest of Google’s services.
  • Offer more information about our business: precise address, opening hours, contact phone, photos we want to share, etc.
  • Have a direct interaction with our customers, making it possible for them to generate comments about our business, and being able to respond on the same platform.
  • Have a continuous and global feedback on what our customers think.
  • If we have several stores, we can analyze the differing opinions of each of them and detect possible problems.

All this information can be obtained directly through Google My Business, but in order to access all the reviews, we will need to work with an API.

An API is an application programming interface. We can view it like the instruction book (in this case for My Business) that tells us how to communicate with the site in order to extract the data related to the reviews.

It is important to point out that during this process, the API provides both quantitative information (1 to 5 star rating) and qualitative information (comments made by the user), so we will have two levels of information that we can analyze.

With this data, we will be able to find out how it’s possible to have users who give low scores (1 or 2 stars) and comments that are quite positive, or on the contrary, give high scores (4 or 5) with negative comments (it may seem strange, but I assure you, it happens). And thanks to technology, we will have the ability to analyse this information, ask questions and make decisions. One of them, by way of example, could be: should we reject those reviews that have no comments, only numerical scores? Comments are more reliable, as they include an explanation, so we should at least consider it.

Also, we have to try to answer other questions that go beyond mere data. How can we detect if the comment of a review is positive or negative independently of the numerical evaluation? Can we categorise and detect what is being said in the reviews? The conclusion is that we can do this by applying Machine Learning techniques to the data.

Applying machine learning to Google reviews

Natural Language Processing (NLP) is a branch of artificial intelligence that aims to make a machine understand what is expressed by a person through the use of a language, which can be implemented for both text and audio.

The first thing we will do through NLP, is to analyze the feeling associated with the comments of our reviews. The algorithm will classify them according to whether they are positive, negative or neutral.

When it is not possible to quantify the type of sentiment associated with a review, the NLP does not categorize it. This is fairly common when, for example, there is a single word review in which the machine cannot detect the context of the sentence, the review is written incorrectly or is not very legible, or there are emoticons that make it difficult to understand.

Once the reviews are categorized according to their sentiment we will have a new layer of information, from which we will be able to make further analysis.

What are the most recurrent themes in my business?

Another analysis that we are able to create using the reviews is determining the main themes. In this case, we are able to use different unsupervised learning algorithms that are capable of detecting the topics in which all the comments will be grouped. The type of algorithm we use is the modelling of topics.

For example, if we are a hotel chain and we apply this type of algorithm to our reviews, it is likely that we will have topics related to the rooms, personnel, facilities, etc, which we can use to group our reviews.

Using the algorithm, we will obtain the percentage of probability that the review belongs to one of the categories or themes that it has detected. The higher the percentage, the more confident the algorithm is that the review belongs to one of the categories.

In this way, following the example of a hotel business, we will be able to classify all comments based on concrete themes and filter them to see only those that relate to what we are interested in. This could be the rooms (problems in the room, improvements that need to be made, opinions about their state, noises, etc); the hotel staff (kindness or rudeness of the staff, thanks for the treatment received, highlighting a worker for any reason); or the different facilities available (opinions about the spa, the gym, the bar area, the restaurant). 

Thanks to this type of analysis, it will be easier to have controlled and organised feedback.

As you can see, we should not solely use the analysis that can be obtained from a particular tool; it is important to try to expand our knowledge about the data. The techniques we have seen above can be applied to any type of qualitative information, be it Google Reviews, Tripadvisor, or any other type of comment that our customers make about our business. This information is very valuable and requires us to apply different analysis capabilities to get new results.

At Good Rebels we have extensive experience with this type of project and we always make sure to get the maximum information we can from users’ opinions. In addition to generating the analysis framework by downloading all the reviews, we are able to represent them on a geolocalized map (as we can see in the map below that we made for a retail client). By applying Machine Learning techniques, we can show the information in a unified way and represent it with the visualisation tools chosen by the client or proposed by us according to our experience.

With this type of service (which can be carried out monthly or according to the desired frequency) we can get a total overview of the opinion of our customers, what they need, what changes could improve the business, and what actions we have to take in order to improve overall customer satisfaction.

Menu