Tomasz Piechula
supervisor: Władysław Homenda
Unstructured data constitutes today around 90% of data generated by the businesses. Therefore it is of paramount importance to be able to analyze it properly and draw correct conclusions. Moreover, there is a constant flow of new data which needs to be analyzed almost real time.
One method proposed to map such data are knowledge graphs, which were used at greater scale by Google in 2012. Knowledge graphs, known also as semantic networks, represent objects, concepts, situations and events i.e. real world entities as a network and are stored in graph database. However, the standard knowledge graphs are static, i.e. they are unable to account for time factor in embeddings and relations.
One of the most exciting challenges is the graph completion problem, which has been widely studied. On the other hand, the temporal graph completion problem is a very interesting problem which simply enables to predict the future facts, but is still relatively unexplored.
In my research I will explore different algorithms that extract the facts, entities, and relations between them and time using advanced NLP algorithms, time series classification and analysis algorithms, then map the data flow in the form of network (graph).
Ultimately, I will experiment with different temporal knowledge graph completion algorithms and hope to bring some new insight into the field.
Currently I am in the process of literature review.