INTEGRATING MACHINE LEARNING NLP TOOLS INTO THE BUSINESS GLOSSARY OF AN ENTERPRISE DATA CATALOG
Integration of an existing Business Glossary system with a machine-learning NLP tool (the AI Index) to auto suggest new glossary entries.
After research suggested the existing glossary’s usability was sub-par, the initial project was separated into two stages: 1. Improvements to the Glossary, and 2. Glossary Integration with the AI Index.
At its core, Ablution is two products in one: a comprehensive SQL composer and a data catalog. It allows analysts and business users to locate, understand, use, and share their company’s data with everyone who needs it - across all departments, whether tech-savvy or not.
Ablution connects to a company’s data sources, and through the cataloging feature, creates a layer of metadata on top of the physical data objects (tables, schemas, databases of information). Every data object, from vast database repositories down to individual columns of a table, is represented on a dedicated page in a consistent format.
With this metadata catalog, users can add titles to data objects, cross-reference linked data object pages, group related data into sets, add clarifying comments to pages, and create Business Glossary entries and Document pages to help define or explain company terms and processes.
Ablution helps to better identify information through the use of machine learning. For a company with hundreds of thousands (if not millions) of data objects, naming and sorting all the data can be daunting.
Database architects often use shorthand names for data objects, called “tokens” (not to be confused with the encryption process of tokenization). Examples of token shorthand could be “sls_rev” for a Sales Revenue table, or a column for Time Stamps named “TS.” Data tokens save time and make writing SQL easier, but could confuse analysts and other business users who are unfamiliar with the shorthand when trying to find or identify information.
Ablution’s AI Index feature uses natural language processing (NLP) to scan a company’s every data source, identify shorthand token names, and make suggestions for their intended real-world meaning. The AI Index then suggests expanded English-language titles that can be approved by users (or one of the other sixteen languages that Ablution has been translated to). These appear in Ablution as subtitles to the original data object name. This real-world title is called a token expansion, or just expansion.
The AI Index uses machine learning to continually improve its title suggestions. As users approve, reject, or edit the AI Index suggested expansions, Ablution learns from the human input and can make improved expansion suggestions.
The issue with the AI Index feature was it was designed very early in Ablution’s product history. It wasn’t integrated with Ablution’s newer features; the interface for its home screen was old and difficult to navigate. The entire concept had never been tested with users and was in need of a refresh.
Newer Ablution customers would benefit most from the AI Index, but our data told us that most users, both new and established, either knew about the AI Index but did not use it, or were completely unaware of its existence.
My most far-reaching project at Ablution was to integrate the AI Index with Ablution’s Business Glossary. The idea was to take the AI Index’s suggestions, and give users the ability to automatically create a Business Glossary entry from the token expansion.
For example, the token “rev” would suggest a Business Glossary page called Revenue, and place links back to every data object page that uses the token “rev.” Data Stewards could then use the new aggregated Revenue page to define the standards and usage for analysts to calculate company revenue across all channels.
This would provide incredible value for both new and existing customers. Right out of the box, Ablution could suggest making Business Glossary entries for the most relevant terms and data objects based on frequency of use and popularity of tokens. Customers could also take advantage of the integrated AI Index’s ability to automatically group data objects together on a single page. In addition, we would update the AI Index interface, and make it easier for users to find and use.
I began researching and found that before we could integrate an updated AI Index with the Business Glossary, we first needed to address some UX issues with the current Business Glossary workflow.
Looking at our usage data, we found more than half of our customers had fewer than 20 entries in their Business Glossaries. Almost a third had no entries at all.
I began a series of customer interviews to get a clearer picture on why customer engagement with the Glossary was so low. Right away, I found that customers were consistently reporting that Ablution’s process for creating Business Glossary entries was confusing, not intuitive, and a poor user experience.
Several customers reported that they attempted to use the Business Glossary, but gave up early in the process. Others, even customers with robust Business Glossaries and several hundred glossary terms, complained that the process was overly complicated.
I decided to split the project into two parts: Business Glossary Improvements and AI Index Integration.
Our existing workflow was not intuitive: templates were required to organize Documents into individual glossaries, which could only be set on the Document page, not the glossary page. There was no home page for the Business Glossary, and no way to create a new glossary term from the glossary page.
I reviewed other Business Glossaries; how they are organized, their approach to User Interface, their information architecture. I looked to Google Docs and Dropbox for UI inspiration on creating and organizing documents.
I created low-fidelity mockups for the new design, incorporating new features from the Project Requirements, and streamlining the workflow for creating a new glossary term. I created a new Glossary home page, with an “Add New” glossary button. Within individual glossaries, the “Add Term” button made it much easier to start adding entries to a glossary.
Mockups were shared with the design team, product team, and engineering teams for review. After a few rounds of feedback and iteration, the general architecture was agreed upon, and user testing began.
Mockups were shared with the design team, product team, and engineering teams for feedback.
A click-through prototype was created on InVision’s prototyping platform. We first tested in-house and then recruited subjects through Respondent.io for a paid, remote study. We tested for users’ ability to:
Results from the user test showed us that overall, respondents were able to complete these tasks with relative ease. Two of the subjects had some trouble with adjusting the glossary settings; interface changes were made as a result.
After review of the tests, I created the high-fidelity prototypes. These were shared with the engineering team for review and feedback.
A Design Specifications document was then made, which detailed each new feature to be built, with notes to the engineering teams collaborating on the build on specifics such as hover and click states, warning text, and edge cases like unapproved users attempting to alter glossary settings.
Once hand-off was complete, the engineering team began building out the new features, and bi-weekly stand-ups were held so I could conduct quality assurance, testing, and be available to answer any questions. I shifted the bulk of my attention to building the Integrated AI Index.
The same design approach as the Business Glossary was adopted for making Integrated AI Index. Low-fidelity mockups for a new AI Index interface were created and shared with the teams.
Subjects recruited on Respondent.io were tested on their ability to:
High-fidelity mockups were created and shared with the teams for feedback. As with the Business Glossary hand-off, a Design Specifications document was created for the AI Index, with notes for the engineers.
I decided to take my leave at Ablution soon after hand-off of the Integrated AI Index was complete, so I was only available for weekly stand-ups at the early stages of engineering’s build of the project.
Before I left, I did hear feedback from the customer success team that the engineering team had completed Business Glossary improvements, and the new features were being used and appreciated, especially by newer customers who had not yet built a robust glossary on the Ablution platform.
Following up with my former colleagues recently, I learned that the Integrated AI Index had been completed and implemented, and initial feedback from customers was good; those who had begun using the Integrated AI Index features enjoyed the ease with which new Business Glossary entries could be created, and the utility of grouping data objects together in one place.
A data catalog is very much like the digital version of an enormous file cabinet system, or a card catalog like you would find at a library. Digitization allows for better organization, sorting, sharing, and finding of information.
Data Object pages form the backbone of Ablution. Every data object, from a single column on a table up through huge databases, each has its own data object page, presented in a consistent format.
Much like physical data in the real world, digital data takes up storage space. Metadata is the information that describes the physical data, such as its name, title, size, date it was created, date last modified, location breadcrumb, etc. Ablution is the catalog that tracks and manages metadata that's associated with physical data, to help people locate, understand, share, and use the physical data.
The AI Index is a machine learning algorithm that makes suggestions for real-world titles of shorthand data names.
The old version of the AI Index home page had a poor layout: confusing navigation, no search feature, no way to sort, and only nine tokens per page. In addition, it was hidden inside the admin menu, difficult for users to navigate to.
User flow diagram, an example of the sequence of steps for users to create a new Business Glossary entry from a data token.
Product requirements document.
User flow diagram detailing steps necessary for creating a new glossary term with the old architecture, vs. fewer steps with the new proposed architecture.
Product Requirements for Business Glossary Improvements.
Low-fidelity prototype of the Business Glossary improvements. Main glossary page.
Low-fidelity prototype of the Business Glossary improvements. Drop-down menu.
Low-fidelity prototype of the Business Glossary improvements. Overlay.
Low-fidelity prototype of the Business Glossary improvements. Individual glossary.
Low-fidelity prototype of the Business Glossary improvements Drop-down search menu.
Low-fidelity prototype of the Business Glossary improvements.
High-fidelity prototype of the improved Business Glossary. Glossary home page.
High-fidelity prototype of the improved Business Glossary. Search menu.
High-fidelity prototype of the improved Business Glossary. Document page, featuring the new glossary side-bar component.
High-fidelity prototype of the improved Business Glossary. Individual glossary page.
High-fidelity prototype of the improved Business Glossary. Search menu.
High-fidelity prototype of the improved Business Glossary. Warning overlay.
Low-fidelity prototype of the Integrated AI Index. Home page for AI Index.
Low-fidelity prototype of the Integrated AI Index. Glossary entry created from AI Index expansion, with section for links to related data objects.
Low-fidelity prototype of the Integrated AI Index. Warning overlay.
Low-fidelity prototype of the Integrated AI Index. Expansion overlay, showing related tokens and data objects.
Low-fidelity prototype of the Integrated AI Index. Hover state for token approval.
Low-fidelity prototype of the Integrated AI Index. Tokens tab of the AI Index screen.
High-fidelity prototype of the Integrated AI Index. Home screen for AI Index expansions.
High-fidelity prototype of the Integrated AI Index. Glossary entry created from AI Index expansion, with section for related data objects.
High-fidelity prototype of the Integrated AI Index. Warning overlay.
High-fidelity prototype of the Integrated AI Index. Expansion overlay showing related tokens and data objects.
High-fidelity prototype of the Integrated AI Index. Hover state for token approval.
High-fidelity prototype of the Integrated AI Index. Token tab of AI Index screen, with updated table layout.
LinkedIn
AngelList
Dribbble
Twitter
Pinterest
Medium
Behance