Semantic Analysis: how to analyze customer reviews

Sentiment analysis of app reviews allows you to quickly detect user-impacting issues that lead to user dissatisfaction, lower app ratings, and uninstalls. Semantic Analysis toolkit is designed to provide a helicopter view on customer experience and detect the app’s potential lacks before they got critical. ML algorithms in the core deeply analyze the meaning of all reviews no matter how many of them an app gets or what rating they have.

For our analysis we take max. 10 000 reviews, filtered by the language and country (at the moment we work only with RU and EN languages and US, GB and RU countries). If an app is not android based, then we take all the reviews for the last year as well as the last ones publisched every 15 min.

Please note! We do not analyse reviews in the languages, that we do not support i.e. if reviews come from US or GB, we analyse only english reviews

AppFollow Semantic Analysis toolkit consists of 5 tools:

  1. Positive vs Negative — arranges app reviews depending on their emotional tone;
  2. Topics & Bugs — gives you a quick overview on customers current and potential issues;
  3. Sentiment Score — shows the overall satisfaction level;
  4. Feelings Chart — visualizes and groups the topics and questions that customers are talking about;
  5. Wordcloud — shows the most popular words from app reviews together with their emotional tone.

This is where you can find the Semantic Analysis toolkit:

You can use the tool to analyze your favorite apps only. To gather an app’s data, please mark the app as favrite. The data will be gathered within the next 24 hours.

Positive vs Negative

The chart shows user attitude toward the app: positive, negative, neutral, and mixed feelings of users that haven’t chosen the sides.

This tool helps you get the emotional tone of reviews and doesn’t cover the ratings of these reviews. If a user rates an app with a 1-star review and asked a question there, it won’t be counted as negative.

By hovering over the columns, you will see how many reviews each of them has and the average rating of these reviews.

Sentiment Timeline

The chart shows you how the reviews and their mood have been changing during the chosen period.

Topics & Bug

These two pie charts group reviews in accordance with the topics, customer concerns and complaints. By clicking each of the coloured piece you will be redirected to the Ratings&Reviews page with the list of app reviews filtered by the semantic tag. The list of semantic tags is available here.

Sentiment Score

The chart shows the ratio between reviews with positive and negative emotional tone where 100% means the exceptional user satisfaction, and 0% — very poor results. We use the following formula to calculate the ration: Positive reviews/ (Positive reviews + Negative reviews).

Feelings Chart

This chart groups questions and concerns depending on the topic, not the rating. Sometimes the groups will consist of reviews with different emotional tones (red and grey, yellow and green, etc), this means that they’re gathered by a common issue or question. In the example below the lower group is united by the registration issues.

By clicking the review bubble you will be redirected to the page where you can reply to this customer.


This tool displays the most common words that customers use in their reviews.

The color is related to the emotional tone used in a review: red is negative, green is positive, and grey is mixed or neutral reaction. The word can be used several times in different colors: “voice calls” can be an issue, and in this case they will be red. When users are satisfied with them, the word will be green.

The size of words shows how frequently they are used. The bigger a word is, the more often customers mention it.

By hovering over a word you will see how many times it was mentioned and the average rating of the reviews with this word. Click on the bubble to see the whole list of these reviews and be able to reply to them.

You can also export the Wordcloud to use in the reports or share on public. There are 2 varieties available: the regular one (like in the example) and the Export 3 that groups the words into the rounded bubble. If you have thousands of reviews and over 200 popular words, you will have the Export 2 option made of AF letters bubble.

How to set up

If you happen to see a splash screen like in the example below, click the button “Yes, please notify me”, and you will get the access to it within a few hours.

Please, note that the Semantic Analysis is available for Premium and Enterprise customers. Contact our Customer Success team to get the access.

You don’t need any further settings, the data about favorite apps will be gathered and updated automatically.

How to filter

There is a bunch of filters you can use:

  1. Filter by review content: use it to filter out reviews by specific words. Note: If you'd like to include several words, apply '' | '' (e.g. bank | account to see the reviews that contain bank OR account) or '' & " (e.g. bank & account will show those reviews having bank AND account together in one review)
  2. Select reviews: shows critical (one- and two- star reviews); favourable (four- and five- star reviews); and featured ones;
  3. All versions: allows you to analyze different app versions separately;
  4. Date: by default, the data is displayed for the previous 30 days. By clicking the dates, you can choose the time period you need.
  5. All languages: filter to evaluate customer sentiments by language.
By default, the semantic analysis is available for English and Russian languages.
To enroll in Beta for the Portuguese language, click on the filter by language -> select Portuguese and click 'Enroll to Beta' in the popup window.

Report Semantic Tag as Incorrect

If you have noticed that the semantic tag was assigned to the wrong review, now it is possible to report Semantic Tag as incorrect.

By clicking on 'Report as incorrect', the semantic tag will be removed from the review, so that you can see only correct reviews by applying the filter.

The information about the reported tag will be used in the future to update our Semantic model to make it more accurate.

Need help? Hit the chat button — we’re all ears!

Did this answer your question? Thanks for the feedback There was a problem submitting your feedback. Please try again later.

Still need help? Contact Us Contact Us