NAGA: Searching and Ranking Knowledge

Research Demo Publications People

NAGA is a new semantic search engine. It can operate on knowledge bases that are organized as graphs with labeled nodes and edges, so called relationship graphs. As of now, NAGA uses a projection of YAGO as its knowledge base. The underlying query language supports keyword search for the casual user as well as graph-based queries with regular expressions for the expert user. One can think of a graph-based query as a labaled graph in which some of the node or edge labels are missing. In order to retrieve answers to a graph-based query, NAGA searches for subgraphs in the knowledge base that match the query strcture and its labels and bind the missing labels in the query. In general, there may be hundreds or even thousands of answers to a given query. The goal is to rank the retrieved answers in such a way that the most important answers are ranked first. For example, for the query that asks for people who were physicists, a good ranking model should give preference to important physicists, such as Albert Einstein, Max Planck, and the like. To achieve this, NAGA's ranking framework formalizes several notions like confidence (the belief in the accurracy of the answer), informativeness (the importance of the answer for the given query), and compactness (the connectivity strength between the entities in the answer). To formalize and combine these notions, we have applied the principles of statistical language models (which are successfully used in document-centric information retrieval) to the unexplored setting of labeled graphs. For more information please refer to our publication: NAGA: Searching and Ranking Knowledge.

Please find more information in the tabs above. If you have any questions about the project, please send a mail to Gjergji Kasneci.