In every industry, but especially in the hospitality & services industry, it is key to understand the needs of your customers. Nowadays, there are so many ways to collect customer feedback via social media, email, telephone, etc. that it becomes difficult to track them all manually. Our customer is already providing the basic features to centralize the feedback which is given on the different social media accounts, but receives a lot of requests to give more advanced insights on all the captured data. The goal here is to automatically classify comments with a specific label (e.g. food, service, price, ...) and give a clear indication on the sentiment of the comment (good, bad or neutral). As our customer did not have the required technical skills set to develop the machine learning algorithms for classification and sentiment analysis on social media posts, fine tune them and integrate them in their existing platform, they came to Data Factory.
Machine learning algorithms can only be trained well when there is sufficient data available of good quality. Also here this was the first challenge which we had to overcome and started to label and clean up the available data to train the models. After analyzing several models we used the state of the art Universal Language Model Fine Tuning (ULMFiT) transfer learning model for the NLP classification. For the sentiment analysis the standard XGBoost model gave back the best results. We deployed the models as a microservice via DigitalOcean due to the best price/quality of this platform. The possibility to use the API made sure that the the amount of modification to the existing platform were limited.
Thanks to the approach we took, our customer can now provide more advanced insights via the dashboard on their existing platform showing automatically what the end customers are talking about, and if they are happy about it or not. These are very actionable insights which help the users of their platform to better understand their customers, take corrective actions and see the impact of their actions in real time. They are doing this without having to lose time or to read every single post about their company on social media.