Watson Discovery Spring 2020 Discovery pipeline Using NLU, - - PowerPoint PPT Presentation

Watson Discovery

Spring 2020

Discovery pipeline

Using NLU, document conversion, and UI tools

Watson Discovery use cases

• The need to search thousands of product reviews at once: Create a

Discovery collection and build a UI to query the collection and graph the sentiment over time.

• The need to programmatically find text within a document: Use the

passage retrieval feature of Discovery to create an FAQ chatbot.

• There are thousands of documents in different formats and you need to
• rganize them logically: Use Discovery to pull out keywords, concepts, and

relationships to sort them.

Term Definition

Collection A collection is a logical division of your data in an environment and is queried independently. Configuration A configuration can be assigned to a collection and can be used to convert and normalize data as well as specify which enrichments to apply. Enrichments Discovery contains a powerful analytics engine that provides cognitive enrichments and insights into your data. These enrichments include entities, categories, concepts, keywords, and sentiment. Aggregations This refers to Discovery returning a set of data values, such as the top values for selected

• enrichments. For example, it can return the top 10 concepts that appear in a data

collection. Passages When working with large documents, you can use the passage search feature to return short and relevant excerpts related to the best matches. Discovery Query Language The syntax of the queries that you would use to search for results, using field names,

• perators, and keywords.

Natural Language Query As an alternative to strict query language, you can also query the language with simple phrases, such as “How do I save a file.” Watson Discovery News A collection that is included with every created Discovery service. It is an indexed data set that is updated daily with over 300,000 news articles.

Enrichment Definition

Entity People, companies, organizations, cities, and geographic features Sentiment Identifies the overall positive or negative sentiment Keyword Determines important keywords, ranks them, and, optionally, detects the sentiment Concept Identifies general concepts that aren’t necessarily directly referenced Classification Classifies into a hierarchy of categories that’s five levels deep Relationships Parses sentences into subject, action, and action form, and returns additional semantic information Emotion Analyses the emotions such as anger, disgust, fear, joy, and sadness

Demonstrations

https://github.com/IBM/watson-discovery-news

Watson Knowledge Studio

If your documents are specific to some domain, you may need to use domain experts to help customize Discovery to better understand the entities and relationships that are unique to your domain. Teach Watson in the language of your domain