TL;DR

First summarization method for knowledge graphs workload- and space-aware.

In this paper:

• We summarize a knowledge graph on a personalized basis inferring preferences from user-search or queries.
• We formulate the problem as maximizing the likelihood of answering user queries given the summary and a space constraint.
• We reduce the problem to that of a maximum coverage.
• We propose an efficient greedy algorithm which is at the same time fast and with quality guarantees.
• We validate our algorithms on large knowledge graphs and show that our method is effective and efficient.
• We design effective methods that simulate real users based on preferences harvested manually.
• We show that our method can outperform strong baseline scoring up to 19% in quality.