Customer feedback clustering using state of the art NLP
Using recent advancements in Natural Language Processing (NLP), the Modulai team developed a model for clustering customer feedback into topics, making it possible to monitor sentiment across these and detect potential negative responses in real-time so that companies can take immediate action.
The team was handed a large dataset containing customer feedback in the form of raw text. The subjects varied greatly, and with very little other contextual information, grouping them into distinct topics was non-trivial. However, modern NLP models have become very powerful and can be trained to identify that texts are semantically similar even though the actual wording is very different. Such a model was implemented and used to cluster the customer feedback into topics making the monitoring process much more efficient.
The client is a Stockholm-based startup founded in 2015 that helps businesses grow by the voice of their happy customers, providing a cloud-based solution for customer referrals, responses, recommendations, rewards, reviews, retargeting, and retention. Staying alert to customer opinion is key since reviews and testimonials are an important part of many companies marketing strategies.
In recent years there has been a revolution in the subfield of NLP in machine learning, opening the door for solving problems that were not possible previously.
Using these advancements, the team developed a model that uses a state-of-the-art Transformer architecture to detect semantic similarities in customer feedback, making it possible to identify clusters or topics. The solution then allows the client to monitor these, and also potentially new topics over time, alerting if there is a clear shift in sentiment.