We addressed the problem of detecting independence between XML queries and updates. Since the problem is undecidable for XQuery queries and updates, and is intractable even for restricted fragments, we adopted an approximating technique based on a schema-based static analysis. Our analysis turned to be precise and, at the same time, fast to run. Main result about this research line have been published in , while the complete study is reported in Federico Ulliana's PhD Thesis (defended in December 12) .

To address the problem of manipulating large XML documents via main-memory XQuery engines, largely used for their efficiency and easiness of integration in a programming environment, we developed partitioning techniques for both XQuery queries and updates. Our technique is based on a static analysis over queries and updates (no schema is used) able to infer information that is used to partition the input document, in a streaming fashion. Besides allowing existing main-memory system to scale up in terms of query/update input size, our technique also admits a MapReduce implementation. Main results have been published in , while the complete study is reported in Noor Malla's PhD Thesis (defended on September 21) .

We also tackled the problem of safe manipulation of JSON data. Some typed and MapReduce-based programming languages for manipulating JSON data have been recently proposed. However, the problem of inferring a schema for untyped JSON data was still open, and having a schema for manipulated data is fundamental for the afore mentioned programming languages. We started investigating technique able to deal with massive JSON data sets. To ensure efficiency, our technique is based on Map-Reduce, while to ensure precision and conciseness it adopts type rewriting rules able to: i) compact as much as possible intermediate inferred types, and ii) to avoid gross approximation when compacting types. Some preliminary results are quite encouraging, and appeared in .

Considerable energy is spent towards enriching XML data on the web with semantics through annotations. These annotations can range from simple metadata to complex semantic relationships between data items. Although the vision of supporting such annotations is spreading, it still lacks the infrastructure that will enable it. To this end we have proposed a framework enabling the storage and querying of annotated documents. We have introduced (i) the XR data model, in which annotated documents are XML documents described by RDF triples and (ii) the query language XRQ to interrogate annotated documents through their structure and their semantics. A prototype platform XRP for the management of annotated documents has also been developed, to show the relevance of our approach through experiments .

A promising method for efficiently querying RDF data consists of translating SPARQL queries into efficient RDBMS-style operations. However, answering SPARQL queries requires handling RDF reasoning, which must be implemented outside the relational engines that do not support it. We have introduced the database (DB) fragment of RDF, going beyond the expressive power of previously studied RDF fragments. Within this fragment, we have devised novel sound and complete techniques for answering Basic Graph Pattern (BGP) queries, exploring the two established approaches for handling RDF semantics, namely reformulation and saturation. In particular, we have focused on handling database updates within each approach and proposed a method for incrementally maintaining the saturation; updates raise specific difficulties due to the rich RDF semantics. Our techniques have been designed to be deployed on top of any RDBMS(-style) engine, and we have experimentally studied their performance trade-offs , , .

We addressed the problem of detecting multiple heterogeneous representations of a real-world object (often referred to as record linkage, duplicate detection, or entity resolution) in two contexts, i.e., for hierarchical data and for data where relationships between entities form a graph.

Concerning XML entity resolution, we contributed to a novel algorithm that uses a Bayesian network to determine the probability of two XML elements being duplicates. The probability is based both on content and on structure information given by the hierarchical XML model. To efficiently evaluate the Bayesian network to find duplicates, we devised two pruning techniques. Whereas the first is lossless in terms of not loosing any true duplicates, the second pruning heuristic trades off runtime for a somewhat lower accuracy of the duplicate detection result. An experimental evaluation shows that the proposed solutions are capable of outperforming other state-of-the art XML duplicate detection methods .

As for duplicate detection in entity graphs, we defined a general framework for algorithms tackling this problem. The general process consists of three steps, namely retrieval, classification, and update. We further proposed an algorithm complying to the framework that leverages an off-the-shelf relational database to store and to efficiently query information (both data and relationships) relevant for duplicate classification. We further extended our framework and algorithm to allow for parallel and batched processing. Our experimental validation on data of up to two orders of magnitude larger than data considered by other state-of-the-art algorithms showed that the proposed methods allow to scale duplicate detection in entity graphs to large volumes of data .

Data warehousing (DW) research has lead to a set of tools and techniques for efficiently analyzing large amounts of multi-dimensional data. As more data gets produced and shared in RDF, analytic concepts and tools for analyzing such irregular, graph-shaped, semantic-rich data are needed. We have introduced the first all-RDF model for warehousing RDF graphs. Notably, we have defined RDF analytical schemas, themselves full RDF graphs, and RDF analytical queries, corresponding to the relational DW star/snowflake schemas and cubes. We have shown how RDF OLAP operations can be performed on our RDF cubes. We have also performed experiments validating the practical interest of our approach.