To understand how IBM Watson Discovery's services can help any enterprise gain business insights, it's necessary to comprehend its capabilities in detail.
One of the core assets of IBM's entire Watson program is the ability to use Artificial Intelligence (AI) to sort through masses of data to come up quickly with insights it could take a human mind much longer to perceive. Previously, data analysis required structured data, such as a database, to provide the input from which insights might be gleaned, but Watson can also handle unstructured data (e.g., from videos, emails, and text documents). This ability means at least one of Watson's core products, called Discovery, has the advantage of being able to use as input data from just about anywhere.
This could be handy, but there's some confusion to wade through at the top. "Discovery" refers to more than one thing. First of all, it's an IBM Watson product name with its own web page. On this web page, IBM tells us there are four subproducts under the Discovery product heading: Discovery, Natural Language Understanding, Discovery News, and Knowledge Studio.
Where it gets a little tricky is that Discovery, the subproduct, isn't quite a product; it's a service that IBM defines as an engine. IBM describes it as "a cognitive search and content analytics engine that you can add to applications to identify patterns, trends and actionable insights to drive better decision-making."
What Discovery does is help you define what the data collection you want to use for gaining insights will actually be. It doesn't help that Watson Discovery's main function, proclaimed right on its web page, is to help users "build a cognitive search and content analytics engine." It's already an engine, isn't it? So what IBM is inelegantly expressing is that the Discovery engine will help you build a "discovery engine" of your own. All you have to do is figure out how to use the service.
If you're just starting out, the Discovery Knowledge Base Search page shows how to search existing data collections for answers to questions. It can be useful to consult this page for an overview understanding of what results you can hope to achieve. Of course, the "garbage in, garbage out" concept that has been with us at least since the ÃƒÂ¢Ã¢â€šÂ¬Ã‹Å“70s also applies here. You'll need to use Discovery to define your data universe before you can expect to ascertain anything useful about your particular business.
In Discovery parlance, this data universe is a "collection." You can set up environments that contain multiple collections. Each environment is both a domain for particular queries and a "service instance" that also contains the metadata about all the data sources you'll be using. Each collection has a composition called a "configuration" that can either use its own default structure or a custom one you build yourself.
Watson as a whole is a platform that uses REpresentational State Transfer (REST) APIs to provide its services within the IBM Bluemix PaaS, (which IBM renamed "IBM Cloud" early in November). In this wider world, each knowledge set that any Watson app uses is limited to what data the user provides, and this data entity is called a "corpus" (plural "corpra"). Each data domain is treated as a different corpus, which enables any one API to do its work across multiple corpra without affecting any data results.
This means Watson Discovery service, in addition to being a product and an engine, can also be described as an API suite that can "ingest, enrich, index, and search structured and unstructured data." (The IBM Watson Discovery Explorer page contains a list of APIs for its common functions. For some tips on using Watson APIs, please see David Shirey's recent "How to Get Started with Watson APIs" article.)
By using the tools provided via the Watson Developer Console Services page, a developer can create a data collection and have it "ingest" data. This consumption can be achieved in several ways. First by specifying JSON, HTML, PDF, MS Word, or other files as part of the collection. Second, by automating the process with Ingestion API calls. The third way is by using the IBM Cloud Data Crawler, a command-line utility within the PaaS for specifying documents within existing repositories. After that, it's just a matter of setting up a filter parameter (i.e., what you're looking for), using Natural Language in the query parameter to specify other filters, and letting Discovery do its thing. Discovery returns data that conforms to your filters and issues a confidence rating as to how well Discovery's algorithms judged each result to fit those filters. Users can then further query this new, smaller data set or change parameters and do a new search.
This process should not be confused with "training" Watson, which is the process by which a particular iteration of Watson is taught its basic operating parameters for a particular environment or industry. That's handled by the IBM Digital Experience Manager. The most recent updates and common FAQs on using Watson in the IBM Cloud environment are available at the IBM Cloud Blog page.
Some additional useful features of Discovery include user authentication, versioning, environment controls, the ability to add documents to (or update/replace docs within) existing collections, and other administrative tools.
It may be, though, that some data you want to analyze needs to come from sources you don't already have sitting in a repository, like social media, news feeds, or who knows what. Or you might want to be able to use Discovery's assets more easily. That's when the other products in the Discovery group might be worth a look.
Natural Language Understanding
Watson Natural Language Understanding (NLU) service was launched earlier this year. Based on IBM's former product, AlchemyLanguage, NLU is a standalone Watson service that enables analysis of text and extraction of metadata from unstructured content. NLU keeps AlchemyLanguage's core features but uses a simplified API and runs natively on the Bluemix/IBM Cloud platform, so users don't have to be experts to get started.
NLU analyzes textual data, from which it can extract information about "entities, relationships, concepts, semantic roles, categories, sentiment, emotion, and other metatdata." Like Discovery, NLU lets users feed text into the API itself or specify a repository or other source (e.g., emails, social media) to examine. In addition, users can build custom models for NLU to use as comparisons for sorting out shades of meaning from language particular to their business or industry by using Watson Knowledge Studio (WKS) (see more below). Such customization can improve extraction accuracy.
NLU also provides a set of enrichment features and can be used in conjunction with other Watson services to build solutions for complex situations.
Watson Discovery News
Another type of unstructured data is news feeds. Over the past two years, it sometimes seems there's more news than ever, and some of it is likely to be pertinent to your current or future business plans. Discovery News provides an API that can scan large amounts of unstructured general news across the web and extract information about overall trends. Designed to be simple, Discovery News' API lets users create multiple queries in minutes and scan either web news in general or just the existing collection of 60 days of the most recent news that comes prepackaged with the Discovery service.
The 60-day summary includes items that have already been categorized, been analyzed for syntax and sentiment, and had the major ideas and relationships extracted. Viewing results from this limited domain gives users practice at using the service and helps them learn how to structure their own queries and list of information sources to improve future inquiries.
Discovery News queries can be customized to include or exclude specific news sources, specify topics or date ranges for search, search for specific concepts, and, like Discovery, be pointed to a user's own data sets, documents, and repositories. The API can handle reviewing literally millions of individual data sources. Via Discovery's query language, users can standardize the queries to work with any kind of data, determine patterns and trends from an entire specified data set, or single out a particular entity (e.g., person, product, company) for constant monitoring and reporting of news items related to that target.
Watson Knowledge Studio
That brings us to the final offering in the Discovery stable, Watson Knowledge Studio (WKS), also a cloud-based service. It's probably easiest to visualize WKS as a workbench of tools that facilitates annotation, which is the addition of new metadata to a document or result set. WKS lets users make annotations of different types, such as machine learning (i.e., establish pointers to what should be the most relevant data), rule-based (i.e., set down specific patterns to look for, principally over a large collection), and //medium.com/@IdTypeThat/pre-annotation-dictionary-issues-in-watson-knowledge-studio-9f5b4977d7ff">pre-annotation (i.e., provide a dictionary of specialized terms, preferably by automated means).
WKS helps developers and subject-matter experts (SMEs) collaborate on customizing annotation components for particular uses. Annotations can facilitate tasks such as specifying in advance such concepts as term instances and relationships between terms and concepts that should be given particular weight in searches. This process is based on use of "knowledge artifacts," which more simply means existing knowledge already learned that applies to a particular search or situation and has, most likely, been entered into a text database by a human.
WKS lets developers and SMEs share an integrated UI, development environment, knowledge artifacts, an annotation-building tool, and training processes to improve coordination between those two groups and enhance annotation accuracy. (Chrome is the recommended team browser.) This can help train Watson in the language of a particular business or industry, improve annotation team productivity, enhance the quality of annotations over time, and facilitate arbitration of conflicts between SMEs. Once created, developers can deploy annotation components to Discovery, NLU, or IBM Watson Explorer.
Analyzing Your Own Content
Although it's not always clear from IBM's own web sources about the Discovery product group, itÃƒÂ¢Ã¢â€šÂ¬Ã¢â€žÂ¢s a collection of services that, whether you use all of them or not, can open a door to the kind of data analysis your enterprise may previously have only dreamed about. Discovery can provide a benefit to your enterprise by helping make your planners and other experts aware of pertinent information either from internal data or from the nearly limitless sources of data that are proliferating around us daily.