Detect brand or product perceptions from comments and reviews

By using deep learning for conducting sentiment analysis of free text fields in reviews, surveys, and comments sections, companies can automatically analyse and classify vast amounts of customer generated information about their brand or products in 100 languages.

02/ The problem

Reviews and comments are a great way to let your customers express their opinion on your product or brand. From their comments you can understand what is the overall perception that your customers have about your product and identify the area of improvements. However, reading through and processing manually large volumes of text is costly and time consuming.

03/ The opportunity for deep learning

Deep learning can automate the process of reading and identify the sentiment in the customer reviews. 
A deep learning model can detect the overall sentiment of a text and classify it in positive, neutral and negative. In this way deep learning allows you to extract valuable insights from free text. If you are operating on an international market you can use a multilingual model that can deal with up to 100 languages at the same time, sparing the time and costs for translations.

04/ Platform model to use

For this use case we suggest using a text classification model, possibly multilingual.

05/ How does the model work?

The model takes a piece of text in input and learns to recognize the patterns in the text that are useful to classify the sentiment of the text. The great thing about deep learning models is that they do not focus only on the presence or meaning of single words, but they also take into consideration the structure of the sentence and the context.  This way the models have a deeper understanding of natural languages. Multilingual models can understand text on a language independent level.

06/ Data requirements

For this use case you would need around 200/300 product reviews for each sentiment. Your data needs to be annotated, that means that for each customer you should indicate what is the issue that they are talking about (i.e the class of the sample). For example, you can have a table with the text in one column and the category in another column. Ideally, you should have the same amount of samples for each category.

07/ Model performance and success

The evaluation view will help you understand how good your model is performing. You should check the performance of your model on each sentiment separately. If you see that the model has a low accuracy on a particular sentiment, this category might not be well represented in your dataset.

08/ Where to learn more

If you’re interested in building this use case, our tutorial on movie reviews sentiment analysis is a great place to start. This tutorial walks you through how to build a text classifier that can predict if a movie review is good or bad. You can also check out our cheat sheets in the Knowledge Center for text classification.