We have some exciting news to share! Two new features have been rolled out for the Watson Natural Language Classifier Service. In an effort to be succinct, here’s some information:
Classify Multiple Phrases
Reaching out to users and hearing their pain-points is key in product development. One of our customers pain-points focused on having the ability to train and classify data at faster speeds. Today, making API calls to Watson Natural Language Classifier just got easier.
Traditionally, a user had to submit the text to be classified sequentially. The new “classify multiple phrases method” supports sending up to 30 text phrases in one request, saving users time when calling the API.
Specifically, the API endpoint seen below will allow for users to try this new feature for the classification of multiple phrases in the same language.
Currently, Watson Natural Language Classifiers multiple phrase classification supports the following languages: English, Arabic, French, German, Italian, Japanese*, Korean, Portuguese (Brazilian), and Spanish.
*Japanese support for classifying multiple phrases is in beta.
Curious to try it yourself? Visit our API Reference to see how it’s done!
Training with larger data sets
The second major update centers around our user’s requests to train with larger datasets.
Initially, when training a model of Watson Natural Language Classifier, the limit of rows per csv file was 15,000. We are happy to announce we have taken steps towards increasing this limit and it has increased to 20,000. More training data will result in stronger classifier models. Utilizing IBM’s Deep Learning as a Service, the larger datasets will leverage graphics processing units (GPUs) for improved training times.
Note: The maximum size of training data remains at 15,000 records for users in the Frankfurt region or those with IBM Cloud Dedicated
Want to give these new features a try? Why not create a quick sample app? See how you can use Watson’s Natural Language Classifier API to identify spam messages and tweets.
Interested in creating a more advanced spam classifier for phishing emails? Follow the same code pattern, but for your training data use sample emails. Feel free to reach out directly to provide feedback or discuss potential use cases!