E-commerce major Flipkart has partnered with the Indian Institute of Science (IISc) to build a Google-like knowledge graph. This will help the company efficiently catalogue its 380 million and growing listings, aiding customers to find the right products that will boost its business.
Knowledge graph refers to underlying base technologies, which was made popular by Google to enhance results it shows users from a variety of sources.
Partha Talukdar, assistant professor, department of computational and data sciences (CDS) at IISc, is heading the project. The firm has several other ongoing collaborations with premier Indian institutes as well as foreign universities to develop machine learning, voice recognition and even delivery using drones. By utilising the natural language processing (NLP) system, Flipkart wants to analyse product descriptions given by sellers and turn it into descriptions that customers demand. The solution will be able to categorise products, find relations between them and power smarter recommendations on Flipkart.
Flipkart wants to use the NLP techniques developed by professor Talukdar and his team to analyse text written by a human to extract relevant information about a product. This data then can be used to link two or more products to each other using other machine learning and artificial intelligence techniques.
"We get a lot of catalogue data from sellers on Flipkart and one of the problems they face is that they have to adhere to our standards of describing something. We want to ask our sellers for the least amount of information, but at the same time provide our customers with as much data as possible about a product to make a decision," says Nishant Khurana, senior engineering manager at Flipkart, who is working on the new project.
"Our research is focused on how we can read unstructured data written in natural language. We have done quite a bit of work on reasoning and canonicalisation, which is applicable to this space. The way to understand this is that Flipkart has some data and consumers are trying to access this to discover products, so you need to have one common language to put everyone on the same page," says Talukdar.