Have you ever entered a site to obtain a service and found that you were talking to a chatbot and not a real person?. Have you noticed that chatbot has become a frequently used term?. Do you know that “The global conversational AI market size is expected to grow from USD 4.2 billion in 2019 to USD 15.7 billion by 2024” according to Markets and Markets?. Many companies have reconsidered the importance of AI & chatbots and the need to rely on them more broadly as soon as possible.

In a world where user experience and conversational channels have become an integral part of business success, it is imperative to find effective, automated, scalable, and optimization solutions with minimal effort. Therefore, reliance on artificial intelligence in general and chatbots, in particular, was inevitable. Thus, delivering a seamless personalized experience to customers.

Using a Chabot instead of a human agent can be easy. But the real challenge lies in the effectiveness of relying on this solution. The more incomprehensible chatbot responses such as “Sorry, I don’t understand you” or “You will be referred to a customer service representative,” the less effective this solution becomes and even a burden to use.

Intent Classification – Natural Language Understanding – The Techniques Needed to Build an Effective NLU System

The main pillar on which the chatbot is based is the Natural Language Understanding (NLU) system. One of the most important components of the NLU system is this component that enables the chatbot to understand the users’ text inputs and then retrieve the appropriate responses, which is called in the chatbot world the Intent Classification.

There are many approaches that can be followed to build an intent classifier, ranging from traditional rule-based methods through machine learning techniques to advanced deep learning techniques. Not only the approach used is the main influencer in the performance of the intent classifier but also the type and size of the data as well as the chatbot language (and dialect as well).

How WideBot improves its engine to face intent classification challenges and breaks the accuracy record

Chatbot users face various obstacles in accuracy rate, especially in the commercial field, which gives an unfortunate CX experience. this experience backfires onto the trust of digitizing the customer support cycle. This is no longer a barrier when incorporating artificial intelligence solutions.

At Widebot, we’ve captured a range of challenges that clients face while building their own chatbots using our sophisticated platform. We focused on those challenges related to the Intent classification task and they are as follows:

  • Clients need to get accurate results without having to enter large numbers of examples for each class (intent).
  • There is a need to provide chatbots capable of supporting different languages and also different dialects. Building a system for each language or dialect is painstaking and expensive.
  • The trade-off between accuracy and speed of the AI component.
  • Users are constantly adding and deleting examples and intents. This led to the need for a system capable of training quickly and generating a model of appropriate size.

These points were the main drive for Widebot to develop its special evolving intent classifier.  WideBot offers a unique Intent Classifier that has a language-independent architecture that can learn from any language. It can train on a dataset that has ~500 training examples divided on 20 intents in only 1 second.  For datasets with bigger sizes, for example, ~50k examples and 20 intents, It takes ~120 seconds in training to generate a few-megabytes model. The prediction process is very fast as we are using an advanced smart architecture for model loading and text processing.

To truly measure the performance of WideBot’s Intent Classifier (WBIC), it was used to train multiple models based on publicly published datasets, following the same configurations found in this paper published in 2021 by Watson under the title “Benchmarking Commercial Intent Detection Services with Practice-Driven Evaluations”. The results of the component were compared with the results of the commercial solutions mentioned in Watson’s paper: IBM Watson Assistant (classic, enhanced), Google Dialgflow, Microsoft LUIS, Haptik, and the open-source solution RASA.

The following graph shows that WBIC outperformed all of the previously mentioned solutions by a significant margin. It is worth noting that WBIC also outperformed RASA in the experiments that were carried out on Arabic datasets. It was also interesting to try WBIC in a sentiment analysis task, and its results were good compared to other systems, which makes it a general scalable solution.

Based on the paper, Watson had achieved the highest accuracy in these experiments using its enhanced system, reaching an accuracy of 72.2%. By relying on WBIC, WideBot was able to break this record and set a new record (72.9%).

Building Conversational Experiences with Machine Learning & Deep Learning.

Our professional, dedicated data scientists have started several initiatives to deploy an enhanced intent recognition and classification model to achieve a higher accuracy rate. WideBot data science team accomplished this by optimizing technological advances in Machine Learning and Deep Learning. 

WideBot artificial intelligence engine can now process intent calcified at the highest accuracy rate in the market at a peaking rate of 72.9%. By achieving this accuracy rate, WideBot beats the record of the market players such as Dialogue flow, IBM Watson, and Microsoft LUIS.

As shown in the figure above.

An AI-powered chatbot can do more than just respond to programmed queries. It can accurately and professionally define the client’s intent and respond to potential clients, often offering them solutions on it’s own without the involvement of a human agent. Even with large numbers of customers, chatbots can still perform as well, increasing conversion rates and providing campaign analytics at minimal costs. As a result, our clients can achieve higher profit rates.


WideBot has managed to outperform the most popular commercial intent classification solutions with its new cutting-edge system. This makes WideBot Intent Classifier (WBIC) the best solution for building your own chatbot to ensure the success of your business and attract more customers at lower costs.

If you need to learn more about our WBIC (WideBot Intent Classifier), Request a demo, and our team will help you.