Search code examples
training-datawatson-discovery

Training IBM Watson Discovery Service for FAQ customer supporting

I'm trying to build FAQs customer support agent with Watson Discovery Service and uploading FAQs documents to this service, but when I query it shows full documents content as a result. There is answered that Question and Answer pairs should be uploaded as a separate documents, but there are many Question and Answer pairs in FAQ document. So is it possible to train Watson Discovery Service and improve results for supporting FAQ format?

Solution

  • If you want Discovery to split your FAQ into Question/Answer pairs automatically on ingestion, look into the Segmentation feature: https://console.bluemix.net/docs/services/discovery/building.html#performing-segmentation

    You could also try using Passage Retrieval: https://console.bluemix.net/docs/services/discovery/query-parameters.html#passages

    (I am an IBM Employee)