Meet our Machine Learning Team
Having both a full-time Lead Data Scientist, Ph.D. student (Erhvervs Ph.D.) and a professor involved at BotXO, all dedicated to the improvement of our machine learning methods and artificial intelligence helps us become an organization with concrete innovative advantages.
Acceleration with both knowledge and resources at the same time
Having a vast amount of knowledge available, coming from different sources and academic skepticism on one side as well as having applied research and business understanding on the other – is simply just a dream scenario for any company. For us here at BotXO, it allows us to both research, test, and deliver business results – but also to accelerate by either applying more resources or more knowledge – all at the same time.
The problem with Machine Learning and NLP in the Danish market
The problem with building AI for small languages, such as Danish, is the fact that most of the efforts have gone to building NLP tools for large markets like English or Spanish, where there is the vast amount of data required to make neural networks work well.
Using classic NLP techniques is at the time not enough to create models that work well in Danish. The catch lies in using a combination of standard methods as well as more creative and scientific approaches. It is enabling us to identify the real intention of the bot users, which delivers greater value for our clients in the Danish Market.
At BotXO we are currently testing our learning algorithm for what will be our Natural Language Understanding (NLU) component –the component that understands a user’s intention-and, secondly, we work on the algorithms that drive our Natural Language Generation (NLG) component, to enable our bots to make better decisions based on what is understood.
The secret sauce in small data
The more complex and best-performing algorithms within NLP are created on huge amounts of available textual data – we can do two things here in Denmark – wait until we get sufficient amount of data – or work smarter by building a learning algorithm based on a small amount of data.
Developing the machine learning techniques to create models that are both smarter and faster in understanding the intent of the different user questions, specifically in danish, is what we have done so far here at BotXO.
What we are working on
The learning algorithm and model for understanding the intents in the Danish language (which we aim to extend to other languages) need to go beyond small talk. In order to benefit our Danish speaking clients, our bots need to have more specialized knowledge about the things that matter to our clients.
Currently, we are using different kinds of specific customer support data (both email and chat data) to make our chatbot software understand questions within further niches that our clients work within. It covers finance, e-commerce, membership organizations, insurance, and subscription-based businesses.
Support related, information or buying intent
With the next couple of weeks, we will be able to test our chatbot software on new client dataset, which will reduce the amount of manual work alongside understanding the intent in a user’s question due to the fact that – In comparison to fully rule-based chatbots – the clients will not need to specify all possible ways of saying the same thing.
The clients will be able to use our models to generalize about ways of saying and asking for the same thing, and whether the intent is to ask for human support or continuing a purchase.
Next step – is learning to answer correctly
One thing is that we can find the intent in the question, within specific types of niche related business conversations – the next thing is to give the correct answer. As our service is now used to both cut cost and in some cases increase conversion– we still need to ensure correct answers and thereby keep the customer experience on the high note.