Building an ideal knowledge management system

By Jovina Ang

SMU Office of Research & Tech Transfer – There are many reasons why knowledge management is important for an organisation.

Among the many reasons, the most mentioned are:

• Speed up access to information and knowledge, or to people who hold the information you need;
• Improve decision-making processes;
• Promote innovation due to the sharing of ideas, collaboration and access to the latest information;
• Improve the efficiency and productivity via reducing the tendency to “reinvent the wheel”;
• Increase customer satisfaction as the information will help you troubleshoot problems quicker and enhance the value your organisation’s offer to customers; and
• Enable seamless transition and onboarding for new employees.

Budhitama Subagdja (Pak Budhi), who is a Senior Research Scientist at SMU’s School of Computing and Information Systems (SCIS), has been working on the research project titled “K-EMERGE: Knowledge Extraction, Modelling, and Explainable Reasoning for General Expertise” since joining the university in November 2020.

He is also one of the inaugural SMU Research Staff Excellence Awards 2022 recipients.

The research

K-EMERGE is a cross-institutional research project involving researchers and scientists from the Institute of High Performance Computing of the Agency for Science, Technology and Research (A*STAR), Nanyang Technological University (NTU), Singapore University of Technology and Design (SUTD), and Singapore Management University (SMU).

What is K-EMERGE about? “Before I answer the question, I first want to say that this is the first time that I’ve been given the opportunity to work with research scientists from so many institutions,” Pak Budhi told the Office of Research & Tech Transfer.

He added: “The primary aim of the research project is to capture data that is found in textbooks and combine it with data that is in people’s heads. In other words, we want to combine both the theoretical and practical data so that it can be synthesised, categorised and presented in knowledge that is easily accessed and used.

“In this world of work that is dynamic and constantly changing, we cannot just rely on data that is found in textbooks, user manuals or technical documentation. Many external factors, including common sense and experience, affect domain knowledge. Also, any knowledge management system needs to be constantly updated and continuously refined.”

The K-EMERGE project is designed to leverage deep-learning AI-based Natural Language Processing (NLP) and cognitive architecture modeling.

NLP gives computer systems the ability to understand text in the same way as humans, whereas cognitive architecture modeling provides a mechanism similar to how people think, infer and reason.

SMU’s contribution in the project is to capture the data and knowledge from textbooks in an automated system while one of the other institutions on the project is responsible for capturing the data and knowledge from human experts.

Given that Singapore Changi Airport is one of the busiest airports in the world, with almost 400,000 commercial flights arriving and departing annually pre-COVID-19, the research team decided to manage knowledge pertaining to jet engines, specifically jet engines that have been manufactured by Rolls Royce.

Turning data into knowledge

Turning data into knowledge involves connecting the data so that it is manageable, easily accessible and useful for decision-making.

It also involves checking and cross-checking across different sets of data to maintain consistency, as well as to ensure that the relationships across that data are correctly grouped and related.

Using NLP and cognitive architecture modeling, Pak Budhi’s initial role was to establish a baseline knowledge using the textbook-extracted data, using a method called knowledge graphs.

Knowledge graphs, which are also known as semantic networks, use graph-structured data models to connect groups of data such as objects, events, situations or abstract concepts. It also shows all the relationships across that information. As more information is added, the more valuable the knowledge graphs become.

Search engines such as Google and Bing, virtual assistants (Siri and Alexa), and social networks including LinkedIn and Facebook all rely on knowledge graphs to organise and structure their data.

In building the knowledge graphs, Pak Budhi leveraged machine learning for the system to continuously learn from its errors and mistakes.

While machine learning brings along with it advantages such as automated learning, there are also downsides that come with it.

There is inherent noise in any captured data due to biases, corrupted data or data that cannot be read or used. That is why machine learning alone is not enough to build good knowledge graphs. To enhance the knowledge graphs, a lot of refinements have to be done manually so that the integrity of the data can be evaluated and enhanced.

Building a prototype

Having completed the system to capture the data from textbooks, the next phase of the project is to build a prototype that combines this data with the data from human experts.

To accomplish this, Pak Budhi worked with the researchers from the other institutions.

“What’s exhilarating about this phase of the project is learning about the optimal way to combine data that is highly structured in textbooks with the loosely structured data that is in people’s heads. What’s challenging is getting everyone on the cross-institutional research team to commit to a common deadline as some team members have other priorities on top of this project,” he said.

Lessons learned

When asked what he has learned since working on this project, Pak Budhi replied: “Like how a human brain learns, an ideal knowledge management system needs to learn continuously. Knowledge graphs can be enhanced by including new expert knowledge, as well as making continuous refinements including adding new relationships to the different groups of data.”

“And knowledge management using knowledge graphs, can give us a visual representation of how humans learn,” he added.

Back to Research@SMU February 2023 Issue