When I joined as a working student some months ago, I heard a lot of words that seemed like buzzwords to me. "Knowledge Graphs", "Digital Twins", "Semantic Stacks" and so on. I was given a challenge to pursue my master thesis in the direction of knowledge graphs. I was perplexed and perturbed back then. But today, I can assure anybody working with huge amounts of data, that knowledge graphs are the way to go forward. What is a knowledge graph? Let us take our own example. I am a person, I am an engineer, I like sewing and so on. A single entity, in this case me, has a lot of relationships with other things. I belong to the type "Person". I practice a profession called "Engineering" and "Sewing" is my hobby. Now, let us adapt the same principle to a more business-like context. There is a file that has personal details of an employee, including their personal targets. There is a second file that contains the strengths and weaknesses of an employee. There is a third file that contains the salary details of the employees. I am just giving random examples that could fit to any business. It could be anything. But let us consider the following situation. A HR team is looking to formulate strategy for their talent acquisition. They want to hunt for talent inside the organization and pursue active sourcing within. To do this, they give a person all the available data pertaining to all the employees. The usual way forward is that this person analyzes each file, finds a way to connect the data, analyze and present it to the team. What if files 1,2 and 3 seen above could be connected graphically? This is what is represented by a knowledge graph. It connects two files, called data sources, with a relationship. Each of these data sources have something in common which draws a relationship between them. File 1 contains employee ID, name, contact details and personal targets. File 2 contains employee name, strengths, weaknesses. File 3 contains employee ID and salary. Now, every employee has an ID and a name. Using this, they could be mapped to any other detail, like strengths weaknesses, personal target, and salaries. Therefore, a knowledge graph could form relationships like: “Employees with employee ID and name has the following personal target, draws salary X, and has a performance report Y”. Therefore, this employee could be the one we are searching for. This result is fetched in seconds and can reduce a lot of human efforts and errors.
Following is an image of a knowledge graph on geography. Each of the circles below represent an entity. There could be 100s of data sources behind them.
The power of knowledge graphs is so huge that it is used in fields like medicine and crime investigation. If trying to explain it could be this difficult, implementing is infinite times more challenging. But, its worth it!
Comments