Current applications export and exchange XML data on the web. Usually, XML data are queried using keyword queries or using the standard structured query language XQuery the core of which consists of the navigational query language XPath. In this context, one major challenge is the querying of the data when the structure of the data sources is complex or not fully known to the user. Another challenge is the integration of multiple data sources that export data with structural differences and irregularities. In this dissertation, a query language for XML called Partial Tree-Pattern Query (PTPQ) language is considered. PTPQs generalize and strictly contain Tree-Pattern Queries (TPQs) and can express a broad structural fragment of XPath. Because of their expressive power and flexibility, they are useful for querying XML documents the structure of which is complex or not fully known to the user, and for integrating XML data sources with different structures. The dissertation focuses on three issues. The first one is the design of efficient non-main-memory evaluation methods for PTPQs. The second one is the assignment of semantics to PTPQs so that they return meaningful answers. The third one is the development of techniques for answering TPQs using materialized views.
Non-main-memory XML query evaluation can be done in two modes (which also define two evaluation models). In the first mode, data is preprocessed and indexes, called inverted lists, are built for it. In the second mode, data are unindexed and arrives continuously in the form of a stream. Existing algorithms cannot be used directly or indirectly to efficiently compute PTPQs in either mode. Initially, the problem of efficiently evaluating partial path queries in the inverted lists model has been addressed. Partial path queries form a subclass of PTPQs which is not contained in the class of TPQs. Three novel algorithms for evaluating partial path queries including a holistic one have been designed. The analytical and experimental results show that the holistic algorithm outperforms the other two. These results have been extended into holistic and non-holistic approaches for PTPQs in the inverted lists model. The experiments show again the superiority of the holistic approach. The dissertation has also addressed the problem of evaluating PTPQs in the streaming model, and two original efficient streaming algorithms for PTPQs have been designed. Compared to the only known streaming algorithm that supports an extension of TPQs, the experimental results show that the proposed algorithms perform better by orders of magnitude while consuming a much smaller fraction of memory space.
An original approach for assigning semantics to PTPQs has also been devised. The novel semantics seamlessly applies to keyword queries and to queries with structural restrictions. In contrast to previous approaches that operate locally on data, the proposed approach operates globally on structural summaries of data to extract tree patterns. Compared to previous approaches, an experimental evaluation shows that our approach has a perfect recall both for XML documents with complete and with incomplete data. It also shows better precision compared to approaches with similar recall.
Finally, the dissertation has addressed the problem of answering XML queries using exclusively materialized views. An original approach for materializing views in the context of the inverted lists model has been suggested. Necessary and sufficient conditions have been provided for tree-pattern query answerability in terms of view-to-query homomorphisms. A time and space efficient algorithm was designed for deciding query answerability and a technique for computing queries over view materializations using stack- based holistic algorithms was developed. Further, optimizations were developed which (a) minimize the storage space and avoid redundancy by materializing views as bitmaps, and (b) optimize the evaluation of the queries over the views by applying bitwise operations on view materializations. The experimental results show that the proposed approach obtains largely higher hit rates than previous approaches, speeds up significantly the evaluation of queries without using views, and scales very smoothly in terms of storage space and computational overhead.
|
![[NJIT logo]](http://archives.njit.edu/vhlib/images/njit_logo.gif){kind=logo}
![[Van Houten Library logo]](images/vhl-top-image.jpg)
