Knowledge graphs are a structured way of storing and providing logical representation of knowledge about the world. Most of the existing knowledge graphs contain knowledge that is incomplete, and only about a small subset of the entire knowledge about the world. In this thesis, we discuss the construction of company knowledge graph that contains data which deals with insured companies, their site locations and the losses encountered by them due to various causes like hurricanes and tornadoes. An existing knowledge graph from the company is the starting point. It contains information from various data sources, and has been configured by the company risk administrators. We, in this thesis, state that there exist relationships between the entities of the knowledge graph that are not represented by the graph, but can be predicted, either as unreported or potential links in the future. Here we, (a) implement link predictions between loss events and insured companies, using state-of-the-art graph embedding solutions, and (b) evaluate the effectiveness of link predictions using qualitative (with company underwriters) and quantitative (precision and recall) metrics. We use different types of embedding models to generate embeddings of the knowledge graph through extensive research of the existing models and perform various experiments using them on the existing knowledge graph. We give a comparative study between the models and then conclude by choosing the most optimal model that performs the best on our dataset. We first work with one link relationships and then extend our work for multiple links, i.e. with a more complex graph which has multiple link relationships. The performance of the current knowledge graph is encouraging for its use in the prediction of more links and nodes of the graph, which can help the company underwriters to take the knowledge graph towards completion.     «

Knowledge graphs are a structured way of storing and providing logical representation of knowledge about the world. Most of the existing knowledge graphs contain knowledge that is incomplete, and only about a small subset of the entire knowledge about the world. In this thesis, we discuss the construction of company knowledge graph that contains data which deals with insured companies, their site locations and the losses encountered by them due to various causes like hurricanes and tornadoes. An...     »