Rita Hartel

Search and Modification in Compressed Texts

Text compression techniques like bzip2 lack the possibility to search or to update substrings at given positions of texts that have been compressed without prior de-compression of the compressed text. We have developed Indexed Reversible Transformation (IRT), a modified version of the Burrows-Wheeler-Transformation (BWT) that in combination with run length encoding (RLE) and wavelet trees (WT) allows for posi-tion-based searching and updating substrings of compressed texts without prior decompression of the compressed text. As a result, IRT may be useful for a huge class of applications that due to space limitations prefer to search or to modify compressed texts instead of uncompressed texts.

Detecting Suspicious Relational Database Queries

Whenever secret information has been shared among multiple partners and has been illegally leaked to a third party, it is important for the trust among the partners to identify the information leak. We present a forensic approach to privacy violation control that after information has been leaked identifies those partners that had access to the leaked information. Our approach represents secret information as a boolean formula and compares it with the queries and the relational database state to which the queries were applied. We use this technique to identify suspicious queries, i.e., queries that have got sufficient information to infer secret information that has been leaked. Furthermore, we prove that checking where a select-project query is suspicious with respect to a given secret information is NP-complete, but a polynomial time solution exists for interesting subclasses.

Searchable compression of office documents by XML schema subtraction

Starting with Microsoft Office 2007, the Office Open XML file formats have become the default file format of Microsoft Office. As each day a lot of office documents have to be stored and transferred, reducing the document size will yield a benefit when storing and transferring these files. We present a compressed format for XML-based office documents that omits that data from an office document that is already defined by the Office Open XML format. Our evaluation shows that our compressed format reduces the - already compressed - office documents to a data size down to 41% of the original document size. Furthermore, for search operations tested in our evaluation, searching is faster on our compressed office documents than it is on the original documents.

XML Stream Data Reduction by Shared KST Signatures

Within XML data streams, markup as defined e.g. in a DTD is not only being used for structuring large amounts of data, but also for efficiently searching, accessing, and processing the required parts of the data streams. However when huge amounts of XML data are involved, data reduction or compression techniques that still allow finding the required parts of the data fast may become crucial to handle data processing. We present a data reduction and compression technique for XML data streams that not only significantly reduces the amount of data, but also allows for efficient data processing without requiring a full data decompression. Our data reduction technique combines sub-tree sharing with removing structure that is known by a DTD. We have done extensive performance evaluations to compare our compression technique with other approaches to XML compression, and we show that we not only outperform the other techniques, but also outperform string compression techniques like gzip that do not support query processing on compressed data.

Fast multi-update operations on compressed XML data

Grammar-based XML compression reduces the volume of big XML data collections, but fast updates of compressed data may become a bottleneck. An open question still was, given an XPath Query and an update operation, how to efficiently compute the update positions within a grammar representing a compressed XML file. In this paper, we propose an automaton-based solution, which computes these positions, combines them in a so-called Update DAG, supports parallel updates, and uses dynamic programming to avoid an implicit decompression of the grammar. As a result, our solution updates compressed XML even faster than MXQuery and Qizx update uncompressed XML.

Updates on grammar-compressed XML data

In this paper, we present updates on CluX, a grammar-based XML compression approach based on clustering XML sub-trees. We show that updates on CluX-compressed data can be performed faster than decompressing the data, loading it into main memory and compressing it. Furthermore, we show how to support fast multiple updates, e.g. performing 100 updates in parallel is more than 70 times faster than 100 single updates.

Mixing bottom-up and top-down XPath query evaluation

Available XPath evaluators basically follow one of two strategies to evaluate an XPath query on hierarchical XML data: either they evaluate it top-down or they evaluate it bottom-up. In this paper, we present an approach that allows evaluating an XPath query in arbitrary directions, including a mixture of bottom-up and top-down direction. For each location step, it can be decided whether to evaluate it top-down or bottom-up, such that we can start e.g. with a location step of low selectivity and evaluate all child-axis steps top-down at the same time. As our experiments have shown, this approach allows for a very efficient XPath evaluation which is 15 times faster than the JDK1.6 XPath query evaluation (JAXP) and which is several times faster than MonetDB if the file size is ≤ 30 MB or the query to be evaluated contains at least one location step that has a low selectivity. Furthermore, our approach is applicable to most compressed XML formats too, which may prevent swapping when a large XML document does not fit into main memory but its compressed representation does.

Incremental updates on compressed XML

XML tree structures can be effectively compressed using straight-line grammars. It has been an open problem how to update straight-line grammars, while keeping them compressed. Therefore, the best previous known methods resort to periodic decompression followed by compression from scratch. The decompression step is expensive, potentially with exponential running time. We present a method that avoids this expensive step. Our method recompresses the updated grammar directly, without prior decompression; it thus greatly outperforms the decompress-compress approach, in terms of both space and time. Our experiments show that the obtained grammars are similar or even smaller than those of the decompress-compress method.

Computing compressed XML data from relational databases

SQL/XML allows generating an XML document as the result of a query that is evaluated on relational data. This facilitates companies sharing their relational data in form of XML documents with other companies or with a marketplace, but significantly increases the exchanged data volume. To reduce both the volume of the exchanged data by exchanging compressed XML and the time needed for compression, we propose an approach that allows preparation of a compressed XML document as the answer to an SQL/XML query directly, i.e., without the need to create the XML document first and compress it afterwards. Our evaluation has shown that generating the compressed document directly is in most cases faster than generating the uncompressed XML document and compressing it, and in some cases it is even faster than the generation of the uncompressed XML document alone. As our approach of generating compressed XML requires only SQL support from the underlying database system, a second advantage is that it can be used for the generation of compressed XML even for database systems that do not (yet) support SQL/XML (like MySQL).

Information disclosure by answers to XPath queries

We consider information disclosure caused by answering queries to an XML database. We introduce an audit framework to determine ‘suspicious’ user queries, i.e. those queries that returned results being sufficient to derive disclosed secret information. We describe secret information in form of a secret query and provide a formal definition of ‘suspicious’ user queries based on what we call ‘readset fragments’, i.e. XML fragments that have been read to produce certain query result nodes. Our approach performs two steps to detect ‘suspicious’ user queries. First, we analyze the structure of user queries and of the secret query to exclude ‘non-suspicious’ queries. Second, we derive a formula from user query, query result and secret query, which is satisfiable if and only if the query is ‘non-suspicious’.