Ranking objects based on relationships and fixed associations
Proceedings of the 12th International Conference on Extending Database Technology Advances in Database Technology - EDBT '09
Text corpora are often enhanced by additional metadata which relate real-world entities, with each document in which such entities are discussed. Such relationships are typically obtained through widely available Information Extraction tools. At the same time, interesting known associations typically hold among these entities. For instance, a corpus might contain discussions on hotels, cities and airlines; fixed associations among these entities may include: airline A operates a flight to city
... , hotel H is located in city C. A plethora of applications necessitate the identification of associated entities, each best matching a given set of keywords. Consider the sample query: Find a holiday package in a "pet-friendly" hotel, located in a "historical" yet "lively" city, with travel operated by an "economical" and "safe" airline. These keywords are unlikely to occur in the textual description of entities themselves, (e.g., the actual hotel name or the city name or the airline name). Consequently to answer such queries, one needs to exploit both relationships between entities and documents (e.g., keyword "pet-friendly" occurs in a document that contains an entity specifying a hotel name H), and the known associations between entities (e.g., hotel H is located in city C). In this work, we focus on the class of "entity package finder" queries outlined above. We demonstrate that existing techniques cannot be efficiently adapted to solve this problem, as the resulting algorithm relies on estimations with excessive runtime and/or storage overheads. We propose an efficient algorithm to process such queries, over large corpora. We devise early pruning and termination strategies, in the presence of joins and aggregations (executed on entities extracted from text), that do not depend on any estimates. Our analysis and experimental evaluation on real and synthetic data demonstrates the efficiency and scalability of our approach.