Published on 1|7|2021 by Lucie Kolářová & Peter Kesch
At SentiSquare, we specialize in natural language processing (NLP), a discipline that uses artificial intelligence to simulate the human ability to read and understand text.
How is it possible for SentiSquare robots to forward customer messages with the accuracy near to a human agent? How is it possible that SentiSquare robots manage such a high success rate in answering FAQs? How can we understand a transcribed call from a call centre? Today we'll let you take a peek under the hood of our technology :-).
Good recipes have a secret ingredient that makes them special and unique. We also have a good recipe for our algorithms. But what's the secret ingredient?
It’s our unique approach to problem solving. SentiSquare's NLP technology is based on distributional semantics. This approach allows us to understand the meaning of text without human supervision or input, that’s why it's called unsupervised learning. We assume that words occurring in similar contexts have similar meanings. This allows us to derive the meaning of words - every word can be expressed as a vector in a high-dimensional semantic space. The algorithms create millions of contextual relationships based directly on the client data. Thus, our algorithms learn the meaning of any text, making SentiSquare's AI language-independent.
In the interpretation phase, we use supervised learning and combine all the patterns found in the data and interpret their meaning in the desired way. Human supervision is used at this stage to correctly interpret the contextual relations in the texts. The resulting solution thus perfectly fits their needs.
In the end, the trick is how to combine both approaches – supervised and unsupervised. And in this step SentiSquare is shining out.
Thanks to our algorithms, we achieve extraordinary results when deploying our solutions in practice with clients:
Due to our unique algorithms and the combination of unsupervised and supervised machine learning, we are getting closer and closer to the accuracy of humans.
To bring true value to businesses, NLP tools must be tuned to work with the different types of data they process. Whether that involves different channels where customer texts come from or different languages. High-end NLP often doesn't include many languages and is only available for a few of the most widely used.
NLP researchers at SentiSquare have mastered the art of adapting NLP algorithms and pre-processing text data. This includes techniques like automatic language detection, tokenization, word normalization, dealing with typos and errors, email segmentation (header, footer recognition, history, etc.) and more. For example, we can work with complex word morphology (different noun tenses, verb tenses, etc.) without telling the AI - our machine learning takes care of that. This makes our tools versatile and language-independent. Languages rich in morphology such as Czech, Hungarian, Polish and others are not a problem.
Sometimes there are really high-value patterns or unexpected trends hidden in the data. Therefore, whenever we get data, we use unsupervised machine learning to create clusters of text parts with similar meanings and produce textual summary for each cluster. In this way, we discover the most important themes and patterns in the dataset the fastest and flag possible false assumptions about the data. We use the results to create a classification system that truly reflects what customers are saying.
Clustering not only provides our clients with valuable insight to help improve the customer service, but also provides the basis for building models for use cases like feedback categorization, routing automation, or customer churn prediction. In short, our machine finds the best way to process the data and offers solutions. SentiSquare AI can do this in a very short time!
We were happy to share our recipe with you. Although most algorithms are nowadays publicly available and published, knowing how to use these algorithms and how to combine them appropriately plays an important role. The way they are combined and used is very unique and hard to replicate. Our valuable years of experience and experience with various projects also help us. Our development team has 18 years of research experience in natural language processing. We use our own libraries and write every algorithm ourselves. This is the only way we have 100% control over the entire solution. We customize solutions for our clients according to their requirements and together we achieve exceptional results.
SentiSquare is a technology company that deals with customer-generated text analysis. As one of the few companies uses artificial intelligence based on principles of distribution semantics, which provides many competitive advantages, such as language independence. The company was founded in 2014 as a spin-off by team of researchers at the Faculty of Applied Sciences of the University of West Bohemia in Pilsen. SentiSquare currently supplies technology to the contact centers of large companies such as T-Mobile, E.ON or Albert.
+420 603 400 124