XML filtering with XPath expressions containing parent and ancestor axes

Abstract

More and more XML data is generated and used for data exchange. In this paper, we address the problem of filtering XML documents with large number of XPath expressions, which may contain ‘ancestor’ and ‘parent’ axes. XPath expressions with these axes are more powerful and flexible for users to describe their interests in publish/subscribe systems. First, we analyze the characteristics of the ‘parent’ axis and propose a series of rules to eliminate it in XPath expressions. Then we propose a new index structure called NIndex, which is designed to efficiently store and index large number of XPath expressions. NIndex offers several features which make it especially attractive for the large scale selective dissemination of information, including the ability to handle complex XPath expressions with ‘ancestor’ and ‘parent’ axes, and efficient pruning. Based on NIndex, we design a new filtering algorithm with low complexity for our problem. Our experiment results show that our algorithm performs well across a range of XPath expressions and documents.

Introduction

The proliferation of the Internet, and the exploding volume of information available on the Internet has fueled the development of a wide range of new applications based on selective dissemination of information (SDI) [3]. These applications include stock exchange, advertisement systems, electronic personalized newspapers, online shopping, online auctioning and entertainment delivery, and require timely distribution of data to a large set of customers. In an SDI (or publish/subscribe) system, users subscribe to a data server with continuous queries or profiles that are expressed in some well-defined languages for expressing their information needs. The SDI system performs the matching task and ensures timely delivery of published data to all interested subscribers.

With XML becoming the standard of data representation and exchange on the Internet, effective and efficient methods have been studied for searching useful information from ordinary and probabilistic XML documents by both structured queries and keyword queries [2], [7], [25], [26], [28], [30], [33], [42]. XML is also adopted for content-based publish/subscribe systems because published XML messages have flexible document structures and subscription rules can be expressed by a powerful language such as XPath [9] and XQuery [5]. In an XML publish/subscribe system, XML filtering is a core part, in which continuously arriving XML documents are routed to users according to their subscriptions expressed as queries. These queries typically specify patterns of selection on multiple elements.

Tremendous effort has been put into efficiently filtering XML streams with a collection of XPath expressions [3], [8], [12], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [32], [36], [41], including XFilter [3], YFilter [12], XTrie [8], XPush [18] and the lazy filtering algorithm (LF) [16]. Most of these systems only support forward axes such as child and descendant axes.

However in real applications, XML queries submitted by users may contain predicates with ancestor and parent axes. Firstly, documents for dissemination usually have different schemas [1], [29], [39], consequently different relationships between ancestor tags of a certain tag t in a query may exist in different documents. Sometimes, users are unconcerned about exact relationships between those ancestor tags as long as they are ancestors of tag t. Moreover, users may not always be aware of the schemas of XML documents. For the above reasons, users have to use the predicates with ancestor or parent axes to constrain the answers.

For example, Alice wants to buy some books from a book selling website. She is interested in those computer books authored by “John”, and published by “Addison-Wesley”. The documents (a) and (b) shown in Fig. 1 all satisfy the Alice’s request, and therefore both of them should be disseminated to Alice. This can be realized by submitting the following query Q with an ancestor axis easily.

$$\begin{array}{c}Q://\mathit{Computer}//\mathit{book}[\mathit{ancestor}\colon\colon \mathit{Addison}\text{-}\mathit{Wesley}]\hfill \\ //\mathit{author}[\mathit{name}=''“''\mathit{John}''”'']\hfill \end{array}$$

However, if Alice does not use the ancestor axis and submits the following query Q_a instead.

$$Q_a://\mathit{Computer}//\mathit{Addison}\text{-}\mathit{Wesley}//\mathit{book}//\mathit{author}[\mathit{name}=''“''\mathit{John}''”'']$$

then only document (a) will be returned because the forward axis in Q_a can express only one kind of relationship between the ancestor tags “Computer” and “Addison-Wesley” of the book tag, i.e. “Computer” is the ancestor of “Addison-Wesley”. Document (b) will not be returned because the ancestor tags “Computer” and “Addison-Wesley” have a different relationship, i.e. “Addison-Wesley” is the ancestor of “Computer”. From the example, we can clearly see that using the ancestor axis provides more expressive power than not using it. For improving flexibility, we also support the parent axis.

In this paper, we propose a new index structure named NIndex, which supports efficient filtering of XML documents based on XPath expressions (XPEs) which contain ancestor and parent axes. NIndex structure offers several novel features that make it especially attractive for large-scale publish/subscribe systems.

Firstly, we analyze the characteristics of the ‘parent’ axis and propose a series of rules to eliminate it from those XPEs with the ‘parent’ axis. Dan Olteanu et al. proposed rewriting rules [34] to transform absolute XPath location paths with reverse axes into equivalent reverse axis free paths, and to rewrite all axes to a minimum set of forward axes. However, the transformed XPEs contain join operations, which are not convenient to handle in an XML document filtering system. In this paper, we only consider transformation of the ‘parent’ axis and transformed XPEs contain only simple axes including ‘ancestor’,‘descendant’ and ‘child’.

Secondly, to improve filtering performance, our NIndex scheme captures common sub-expressions among multiple XPEs. Xtrie [8] is indexed on a set of substrings by using a substring table in which levels of elements are recorded. However, the level attribute may cause redundancy despite the identical sub-patterns. Different occurrences of the same sub-pattern are treated as different items in the substring table if they appear at different levels. Therefore our index is more concise to represent XPEs. It reduces the number of unnecessary index probes and avoids redundant matchings.

Thirdly, NIndex is designed to support effective filtering of XML documents based on complex XPEs containing ancestor and parent axes. Xaos [4] is a query processing system that also supports ancestor and parent axes, and it uses the X-dag structure to express queries. However the X-dag structure can only handle a single XPE, and does not provide support to handle the case of multiple XPEs.

Finally, the experiments show that our algorithm is time-efficient.

The remainder of this paper is organized as follows. In Section 2, we introduce the background and give the definition of a PXPE-tree. Section 3 discusses how to eliminate the parent axis in a PXPE-tree. In Section 4, we introduce the structure of NIndex which is the index for XPEs, and design updating algorithms for maintaining NIndex. In Section 5, we propose a filtering algorithm to address the XPE retrieval problem, which is based on NIndex. Section 6 shows our experimental results. Section 7 discusses related works and compare them with our work. Section 8 concludes the paper.