Intent recognition—sometimes called intent classification—is the task of taking a written or spoken input, and classifying it based on what the user wants to achieve. Intent recognition forms an essential component of chatbots and finds use in sales conversions, customer support, and many other areas. Intent recognition is a form of natural language processing (NLP), a subfield of artificial intelligence. NLP is concerned with computers processing and analyzing natural language, i.e., any language that has developed naturally, rather than artificially, such as with computer coding languages. This post explores the basics of how intent recognition works, along with some high-level use cases, and how you can apply it in your business or project with the help of our intent recognition data program.
Intent recognition works through the process of providing examples of text alongside their intents to a machine learning (ML) model. This is known as using training data to train a model. Only once you’ve done this can the model operate in production. But what is training data?
Training data is a representative sample of raw data that is manually labeled or organized in the way you eventually want your model to do automatically. Once you’ve labeled the data yourself, you feed it into your ML model so it can learn what you expect it to do. This process is called training a model. Once you’ve trained your model, you use another sample of raw data to validate your model. The validation process is a means of testing the model to see if it performs well on the type of data it’s going to process in production. You feed the raw, unlabeled data into the model and see if its output is accurate. If the model performs well, it’s ready for production. If it doesn’t, you need to provide it more—and probably better—training data.
When it comes to intent recognition the process of creating training data looks like this:
The process of manually labeling training data is the most time-consuming and laborious aspect of utilising ML. That’s why we developed our text intent recognition data program. You input your raw data and the data program breaks it apart into discrete tasks to produce a coherent labeled output automatically.
Now, the easy part! With your training data assembled you can ask us to train a model for you, which would be accessible via API, or you can create your own or choose from many open-source models (such as one from Google’s BERT) and begin the training process yourself. Once your model performs well on your validation set, you’re ready to let it loose in production.
Intent recognition finds a comfortable home in any situation where there are a large number of requests or questions that are often quite similar. Let’s take a look at some examples.
The rise of chatbots aligns with user trust in chatbots. Helpshift found that twice as many consumers were willing to interact with chatbots in 2019 than in 2018. This leads to huge market growth (the global conversational AI market is expected to hit USD $13.9 billion by 2025) and enormous operational savings (some sectors are expected to save over USD $7 billion a year by 2023).The largest hurdle to overcome with chatbots is that of accuracy. Categorizing users' inputs effectively in training data makes it much easier for an ML model to form a deeper understanding of the users’ needs and provide more relevant and helpful responses.
Cut down on time spent copy-pasting responses to FAQs, sort customer queries by priority and request type, and automate menial, time consuming tasks, such as customer address changes. Many businesses have a pre-existing wealth of user information already available from years of CRM records; this is training data just waiting to be exploited.
Leads go cold fast, so surfacing those with clear purchasing intent for prioritization during inbound and outbound sales processes is essential. Intent recognition can automatically sort responses to email campaigns into categories like “out of office”, “incorrect contact person”, or “not interested”, so you can focus on the leads that really matter. The Harvard Business Review shows the importance of fast contact times:
“Firms that tried to contact potential customers within an hour of receiving a query were nearly seven times as likely to qualify the lead (which we defined as having a meaningful conversation with a key decision maker) as those that tried to contact the customer even an hour later—and more than 60 times as likely as companies that waited 24 hours or longer.”
If intent recognition sounds useful, it’s pretty easy to start out. There are advanced open-source pre-trained models that can get you going. But before you begin using any available model, you will need to produce a relevant labeled dataset to train the model on. That’s where super.AI can help. Our intent recognition data program takes your raw text and sorts it into your desired intent categories. We’ll label the text you send and return a high quality training dataset that you can take to train and tailor your intent recognition model. We can even complete the process for you and train a model in the background that you can access via API. If you’re interested in learning more or have a specialized use case, reach out to us. You can also stay tuned to our blog, where we’re running a series of posts covering different aspects of NLP.