Active Big Data Analytics

The availability of huge volumes of data characterizes our times as unprecedented in history. Big Data technologies promise to emphasize the role of such resources in terms of computational benefits such as the high speed, low costs and large scale to which they support data processing.

However, with the growth of data in several dimensions, formats and sources, it is evident that the largely different implicit assumptions about their intended interpretation are also amplified: independent design choices across sources are the norm and a consistently integrated reuse of data is increasingly complex and costly.

Lexical Meaning Theories as well as Semantic Inference Methods

A crucial knowledge integration problem is faced by all Big Data applications, especially considering that Web scenarios are mostly faced with unstructured data. Whenever Big Data moves from information about natural phenomena, whose observations and measures rely on long-standing scientific traditions, to human-generated contents, such as the one widespreadly shared across the Web and Social Media, interpretation and agreement across sources becomes hard and play the role of a bottleneck for large scale distributed applications.

Reveal exploit a linguistic perspective (lexical meaning theories as well as semantic inference methods) to the crucial knowledge integration problem.

Artificial Intelligence methods are inspired by the idea that machine-readable formalisms for expressing knowledge and use it in support of rational decisions are not only viable but are the closest approximation to human rationality. Natural language is the main manifestation of such status. On the one side, it results from the adaptation of human beings across centuries in their needs of designing, communicating and applying knowledge. However, natural language requires interpretation whose methods in NLP have been largely studied. In Reveal, we exploit a linguistic perspective (lexical meaning theories as well as semantic inference methods) to the above-posed knowledge integration problem.

Data Reconciliation Processes

Reveal language processing engines and machine learning methodologies are strongly integrated in an overall Data and Knowledge Management framework largely applied as a back-office infrastructure to heterogeneous data sources within newer applications. The services hosted by the framework are able to support the integration of conceptual models characteristic of different data sources and generalize semantic notions from linguistic and non-linguistic data, thus offering a crucial advantage in every scenario where an interplay between structured (e.g. relational) and unstructured (e.g. textual or visual) data is involved.

Active Data Analytics processes have been successfully applied in the banking and system engineering domains as well as in media industry and tourism applications.

Intelligent methods are employed to automate data reconciliation processes aiming at mapping independent data sources and individual data items to shared schema according to the linguistic interpretation of data values, data attributes and data set labels. In this way, linguistic semantics is used to propose classification and alignment functions across data sources: this has a huge impact on costs both on the design stages of complex software systems and on the maintainability levels. Active data analytics processes are always in place when an existing engine for a domain A is available and it is necessary to make it operational onto a second different domain B: in this case, domain adaptation and transfer learning are activated on the basis of linguistic information and methods.