| Keyword search over relational databases now falls into two ways:tuple graph based way and schema graph based way, the search unit of these two methods is tuple, so they are essentially tuple-level-based way. The disadvantages of tuple graph are the large size, and that the related messages are divided into different tuples because of normalization, this makes the semantic of tuple graph is dispersed, and the search result is neither very good. The object-level data model groups disperse but related tuples, reduces the size of data graph and keeps the semantic integrity of search units.Some research of database keyword search involved object-level information search, these research grouped related information in database as search units, and applied information retrieval techniques on these search units, but these research regarded search units as individual documents, there is no relation between documents, so the reference information between tuples are lost. The size of search units has the problem of too big or too small. In this thesis, existing object-level keyword search research is analyzed, and an object-level data model is proposed, the search units of data model, namely objects, are divided based on database entity, and kept relationship among them, objects and their relationships compose an object-level data graph.An entire construction method of object-level data model is proposed in this thesis, including the method of construct object-level schema graph based on relation classification, the method of construct object graph based on object-level schema graph, and the method of using database fulltext index to implement object index. In this thesis an object-level data model construction prototype system is implemented based on MySQL and DBLP sample database. The prototype shows how to construct object-level data model, and how to run keyword search algorithm on tuple graph and object graph. Extensive experiments show the advantages of object-level data model. |
PDF Full Text Request