Conversations based on artificial intelligence are increasingly popular. And rightly so. The best chatbot is one that saves you time and money while improving the end-user experience.
Whether you want to improve customer interaction or your employee experience, we're here to help you create the best chatbot possible. The possibilities are endless with botpress.
Start building today. It's free.
why we got rid of intent classifiers
botpress does not use intent classifiers. It's on purpose. Here's why.
Jean-bernard perron
One of the most common questions we get asked by potential users and customers is, “where are your intent classifiers?”
We don't have. And yes, it's on purpose.
Botpress uses llms to identify user intent. because? It is much better for both the creators and users of an AI agent.
We strongly believe in this position, so I would like to take a few minutes to explain our lack of intent classifiers.
Tldr: It's easier to build, more accurate, and easier to maintain.
The old days (before dellm)
(If you are familiar with what intent classifiers are and what they do, feel free to skip this section).
An intent classifier is a tool that classifies user input into predefined intents based on training data.
Developers have to curate and tag countless examples for each possible intent, hoping that the system can match user input to these examples.
For example, with an e-commerce chatbot, developers could define an intent like “trackorder.” Your example sentences might include: “where is my package?” "track my order" and "can you check the delivery status for me?".
Essentially, they are training the AI agent to recognize the user's intent by giving it examples. And yes, they have to enter them all by hand.
Fortunately, the need to perform this manual assignment of possible statements to an intention has practically disappeared as llms has advanced.
But many conversational AI platforms continue to use them. because? We'll talk about it later.
It's not just that it's a longer process: intent classifiers are terrible for many reasons. Here are some:
1. Data dependency
Intent classifiers need a lot of data. They need a huge, representative data set of user examples for each intent to work accurately. Without them, they have difficulty classifying inputs correctly.
And creating these data sets is a slovenia phone numbers nightmare. Developers spend endless hours collecting and tagging examples, which is certainly not a good use of their time.
2. Limited scalability
Intent classifiers are also not designed to scale. Adding new intents means collecting more data and retraining the model, which quickly becomes a development bottleneck. Plus, they can be a maintenance headache, because as language use evolves, so do the statements.
3. Poor understanding of the language
intent classifiers lack a true understanding of language. They have difficulty with language variations, such as:
synonyms
paraphrase
ambiguous wording
typos
unknown colloquial expressions
fragmented entries
They also tend to process each utterance in isolation, meaning they lack the ability to maintain context throughout a conversation.
