Watson Natural Language Classifier Best Practices
Get Started with the IBM Watson Natural Language Classifier Service.
Hello Medium! My name is Zia Mohammad and I am on the product management team for Watson Natural Language Classifier and Watson Language Translator APIs.
Watson Natural Language Classifier
Think of all the classifications that are done, from risk analysis and support tickets to product descriptions, classification runs behind the scenes to save time. The development of AI has allowed for strides in natural language processing, where computers are now able to understand human language as it is written. There is no longer a need to look for specific key words.
A way to sort these text-based classifications can be done through natural language classification. A quick way to get started is with the Watson Natural Language Classifier API. Simply input unstructured/structured data, train your classifiers, and let Watson sort the rest.
Need inspiration?
Here are some examples to help you start thinking of what you’ll try with Watson Natural Language Classifier:
- E-commerce and Retail: Think of all those product descriptions or user reviews while shopping, help your users choose products by narrowing the choices by theme. Tag products or identify fraudulent items.
- Higher Education and Government: Sort text or documents into categories. Academia, law, non-profit organizations, all have some form of classification. Put Watson to the test.
- Social media: Tweets, Emails, Posts, and Shares are the new mail. Analyze these and sort them into categories.
- Services: Support tickets, service queries, and messages can be difficult to parse. Use Watson to categorize problems and offer faster solutions.
- Talent solutions: Analyze resumés and applications to derive deeper meaning and soft skills.
Getting Started Tutorial: Watson Natural Language Classifier
Advanced Users: Best Practices Guide
Already familiar with the service? Consider yourself an advanced user? Check out our best practices guide for training your data.
Only 9 slides, the linked PDF is the key reference for any developer already using IBM’s Watson Natural Language Classifier service.
Here’s a quick tip on working with intents.
Intent?
Correct, each statement or phrase that we read, type, or speak has one or more intents that it can be associated with. For example, a sentence in your training data: “Show me pink Audi convertibles?” could have 3 assigned intents during training— Color, Car Type, and Car Model.
So how would you go about extracting multiple intents with the API? When choosing your training data, start by assigning three classes to a training set:
Class 1 - vehicle_color_pink | vehicle_color_blue | …
Class 2 - vehicle_model_audi | vehicle_model_bmw | …
Class 3 -vehicle_type_convertible| vehicle_type_sedan | …
Once you’ve assigned the multiple classes to the text strings in your training data, you’re ready create and test the classifier!
Once the classifier has been trained, we can enter a similar string for testing—
“Do you have blue BMW sedans?” The classifier will output the following classes showing how confident it is:
Class 1 — vehicle_color_blue: 33%
Class 2 — vehicle_model_bmw: 33%
Class 3— vehicle_type_sedan: 33%
Interested in learning more advanced tips? Check out all 7 tips: Watson Natural Language Classifier Best Practices Guide
