Bhushan Kotnis

As a Research Fellow at Heidelberg University, I work on bleeding edge research problems in natural language understanding. More specifically, I work on question answering, knowledge graph completion, link prediction. This involves actively contributing to research design, coding prototypes, experimentation and writing research papers.

I formulated algorithms to address sparsity in Linguistic Knowledge Graphs (KG). We use link prediction algorithms that leverage an external text corpus for inferring relations (edges) across normalized entities (vertices). The hope is that Inferring links with high precision will densify the KG eventually making it useful for industrial applications. Research Areas : machine learning, natural language processing

Helped BloomReach learn linguistic relations between product queries such as "product accessory of product" using large query data. I used word embeddings and deep learning for implementing an algorithm which classifies weather a phrase (product title) is related to another phrase (product title) through a given relation (accessory of). This data enables the BloomReach recommendation engine to recommend product accessories, replacement parts, etc after a purchase.

Engineered a solution to extract data from a legacy Insurance Lifecycle data model and migrate it to a new Lifecycle data model. Challenge involved modifying a code generator to generate the required code.

My thesis is primarily focused on studying cascades in complex networks, and developing algorithms for running cost effective information campaigns on social networks. Research Areas : social networks, percolation theory, mathematical modeling, multi agent simulation, optimization.