Osaka BMXUG Study Meeting -Kubernates Experience & Watson Discovery Introduction- Participation Report on March 27, 2019

This time, I participated with the aim of experiencing k8s, but there was an unexpected harvest, so I decided to write an article to share it. The hashtag is here #bmxug

https://bmxug.connpass.com/event/117966/

Watson Discovery

What is Watson Discovery??

Simply put, it is a search API made by IBM. It is equipped with unique features of "IBM" different from the familiar full-text search system ElasticSearch. Also, I was very interested because it can be used for free.

What features does it have?

By giving some text data to WatsonDiscovery, it extracts the context in the text from various aspects. It is particularly active when you want to search a large number of documents.

Search by meta-information

It extracts good meta-information from the registered text data. For example, it seems to have the following attributes.

Entity
Relation
Keyword
Category
Concept
Semantic role
Sentiment
Emotion

For example, if you pass an angry text to WatsonDiscovery, it will be given "anger emotion" meta-information. Wow! Amazing! It's certainly a unique feature. This feature was not supported in Japanese, but it seems to have been possible around August 2018!

Find hidden concepts

It seems to find not only the words in the text but also the concept of words not in the text. How does it work!?

A crawler is also provided (WebCrewl)

It seems that there is a function that regularly crawls Japanese news and updates the discovery dictionary. One of the methods is to specify the URL and it will crawl automatically. It's easy... With this, you don't have to prepare data, and you can easily create things like inquiry Q&A.

Sample introduction

For example, search for "White Day gifts that will be appreciated". Also, specify "positive" as a sentimental meta-information search. (The data has already been collected by WebCrewl)

The results were displayed in various ranking formats. For example, in the ranking of "how much money to use?", the TOP was free. This was interesting.