Robust keyword search in large attributed graphs

• Published in: Information retrieval (Boston) Vol. 23; no. 5; pp. 502 - 524
• Main Authors: Bryson, Spencer, Davoudi, Heidar, Golab, Lukasz, Kargar, Mehdi, Lytvyn, Yuliya, Mierzejewski, Piotr, Szlichta, Jaroslaw, Zihayat, Morteza
• Format: Journal Article
• Language: English
• Published: Dordrecht Springer Netherlands 01.10.2020; Springer Nature B.V
• Subjects: Computer Science; Data Mining and Knowledge Discovery; Data Structures and Information Theory; Graph theory; Graphs; Information Storage and Retrieval; Keywords; Mining Actionable Insights from Online User Generated Content; Natural Language Processing (NLP); Pattern Recognition; Random walk; Robustness; Searching; Shortest-path problems; Social networks; Attributed graphs; Keyword search; Social networks
• ISSN: 1386-4564; 1573-7659
• DOI: 10.1007/s10791-020-09379-9

There is a growing need to explore attributed graphs such as social networks, expert networks, and biological networks. A well-known mechanism for non-technical users to explore such graphs is keyword search, which receives a set of query keywords and returns a connected subgraph that contains the keywords. However, existing approaches, such as methods based on shortest paths between nodes containing the query keywords, may generate weakly-connected answers as they ignore the structure of the whole graph. To address this issue, we formulate and solve the robust keyword search problem for attributed graphs to find strongly-connected answers. We prove that the problem is NP-hard and we propose a solution based on a random walk with restart (RWR). To improve the efficiency and scalability of RWR, we use Monte Carlo approximation and we also propose a distributed version, which we implement in Apache Spark. Finally, we provide experimental evidence of the efficiency and effectiveness of our approach on real-world graphs.