Knowledge graphs (KGs) are considered to be promising tools for accomplishing many tasks such as question answering, recommendation, information retrieval, etc. due to their ability to store semantically structured information. While there are a huge number of systems to add a lot of data to such KGs, we observe that the schema of most of the popular KGs are weak. Hence automated techniques to enrich the schema of the knowledge graphs so that they cater better to the needs of the application, have become the need of the hour. In this seminar, we present results about solving two schema enrichment problems, namely, identifying relation gaps in a KG and discovering disjoint object property pairs, using unsupervised mechanisms.

We call a class-pair which could potentially be connected by object properties as a relation-gap. We designed several criteria derived from various sources and fed them into the standard multi-criterion ranking approach TOPSIS (Technique for Order Preference by Similarity to Ideal Solution) in order to find the most suitable class-pairs which could be potentially connected by object properties. Through our experimental evaluation on three popular knowledge graphs, we can see that the proposed approach yields promising recommendations for the relation-gaps.

Given a KG, our proposed system DOPLEX finds the disjoint object property pairs in its schema. DOPLEX makes use of Probabilistic Soft Logic (PSL) to aggregate the ontological and lexical pieces of evidence in a unified probabilistic framework and predicts the disjointness scores for the property pairs. We claim and experimentally show that combining these two pieces of evidence along with checking for instance overlaps works well, especially for automatically extracted KGs like NELL. DOPLEX also achieves a higher precision value when compared to the existing system when run on the PATTY dataset.