Original NLP 2.0 module in KODA Bots

NLP 2.0 is an original system created by the KODA Bots team, based on thousands of messages exchanged on our platform with end users every day. It allows to recognize and adapt even more efficiently to the natural language of speakers in conversations between a bot and a customer.

Hang on, hang on… Let’s start from the beginning. Natural Language Processing (NLP) and Natural Language Understanding (NLU) are the next steps in the development of automated communication platforms. Chatbots, and then voice bots, became a permanent part of brand communication with consumers about four years ago. Since then, limited capabilities of written text processing (NLP) affected the way how automated communication solutions are developed and, consequently, the way how bots communicate with humans.

Being a partner to businesses and organizations that want to automate some specific business processes, our main goal is to respond as accurately as possible to the needs of end users. Remember that this does not always require extensive NLU or NLP systems. One of our latest implementations, a shopping assistant for Komputronik’s online store, does not even have a dialogue box for written messages. Does it comply with the latest trends? Probably not. Does it do its job successfully? Definitely yes!

The KODA Bots team has always had a very clear goal – build a system that meets the needs of our clients and the possibilities offered by the latest technological developments, yet only within the scope of objectives set for a given solution. We act on real needs and real effects, rather than impulsive declarations based on theoretical, often not tested in practice, trends.

Easy management of content and automated conversation scenarios have been our top priorities ever since we started to work on the platform. We make sure that an interface is as user-friendly as possible, because it is helpful both for our chatbots designers, and for our clients, as they can perform some tasks on their own, e.g. add new blocks, update content, train chatbots, etc.

For this reason, when designing the NLP 2.0 module we took into account the following:

• Scalability: a chatbot coach has to manage hundreds or thousands of keywords and phrases, which might be troubling; now one update is enough to have them reloaded wherever they’ve been used.
• Safety of change: thanks to regression tests, changes in configuration are monitored for potential problems on an ongoing basis
• Precision: the module detects more than one piece of information from user’s message, so it can choose and send the most suitable response.
• User-error tolerance: when looking for responses, the mechanism automatically detects and corrects spelling mistakes in users’ messages.
• Multilingualism: a comfortable solution for effective management of bot’s various language versions.
• Adaptability: facilitates matching responses when selected conditions are fulfilled – from hours or days to user-specific data.

We work on entities and intents. Entities can be understood as synonyms, words or data in one database, while intents are composed of entities and allow to get the full context of utterances.

Thanks to the adopted solutions, the technical potential of the new NLP 2.0 module has been extended significantly with the following functions:

• the possibility to create entities, i.e. data sets detected in users’ utterances;
• entities can be defined as:

• a glossary or collection of phrases,
• a regular expression (e.g. e-mail, phone number, postcode, n-digit number),
• a dynamic entity – searched for in an external database, e.g. client’s database;

• the possibility to create intents containing the above mentioned entities and use the same entities in multiple intents;
• the possibility to define several entities in one intent; the entities are detected regardless of their order in user’s utterance;
• the possibility to transfer the detected entities, thanks to which they can be used in chatbot’s response;
• convenient interface allowing to define entities and intents in all chatbot’s language versions;
• expanded reporting system that allows to analyze how NLU works and optimize its functioning.

All this enables quick and convenient creation and management of intents. The NLP model thus created can process and recognize users’ questions with high effectiveness.

Due to a variety of goals and tasks our chatbots are intended to perform, the NLP 2.0 module is the most suitable for customer service chatbots. It allows to automate customer service to an extent that has never been available before. If you have any additional questions or require further clarification, please contact me. I’ll be more than happy to provide any information you need and help you decide whether this solution is just what you need. We’ll see together how the module can serve you best.