AWS debuts Amazon Kendra, an ML-powered enterprise search service
Amazon Web Services on Tuesday announced Amazon Kendra, a machine learning-powered service for enterprise search. The service aims to help organizations, without machine learning expertise, build a better search engine for their internal documents.
“We think this is a step-change in the capability for internal search,” Matt Wood, AWS’s VP of AI, said at the AWS re:Invent conference in Las Vegas. “We’ve certainly experimented with different technologies within Amazon, and none of them work very well.”
Data within an organization can be hard to search, given that it’s often unstructured and siloed, he explained. Files may use different jargon and formats, and it may exist in SharePoint, Dropbox, or any number of places. Consequently, internal enterprise search tools typically offer limited accuracy, and they often only link up a limited number of data silos.
By contrast, the world wide web is relatively easy to search — you can rely on the structure of HTML and links between documents to infer relevancy.
Customers can set up Kendra from the AWS console, providing the service with your credentials for the services you want Kendra to query. Kendra then uses machine learning to build a search index based on your organization’s documents and any FAQs. It uses natural language understanding to learn the intent and context of information within a document, as well as the relationships between content and documents.
The service can be queried with natural language. In addition to surfacing relevant documents, the search service will provide a direct answer if it can. For instance, if someone were to ask, “Where is the IT support desk?” Kendra could reply, “On the third floor” in addition to providing relevant links.
Within the AWS console, you can test and refine your queries. The service gets better over time, improving with user feedback and data on whether users are actually clicking on the links provided.