Scientific Board
*Maristella Agosti, ProfessorScientific goals
* Modeling innovative and specialized information retrieval systems for global patent document collections. * Investigating and developing an adequate technical infrastructure that allows interactive experimentation with formal, mathematical retrieval concepts for very large-scale document collections.< * Studying the usability of multimodal user interfaces to very large-scale information retrieval systems. * Integrating real users with actual information needs into the research process of modeling information retrieval systems to allow accurate performance evaluation. * Ability to create different views of patent data depending on the focus of the information needed. * Defining standardized methods for benchmarking the information retrieval process in patent document collections. * Ability to handle text and non-text parts of a patent coherently. * Designing, experimenting and evaluating search engines able to retrieve structured and semi-structured documents in very large-scale patent collections. * Integrating the temporal dimension of patent documents in retrieval strategies. * Improving effectiveness and precision of patent retrieval, based on ontologies and natural-language understanding techniques. * Refining IR methods that allow unstructured querying by exploiting available structures within the patent documents. * Formal (mathematical) identification and specification of relevant business information needs in the field of intellectual property information. * Investigating efficient scaling mechanisms for information retrieval taking into account the characteristics of patent data. * Investigating and experimenting with computing architectures for very high-capacity information management. * Establishing an open eScience platform that enables a standardized and easy way of creating and performing IR experiments on a common research infrastructure. * Discovering and investigating novel use cases and business applications deriving from intellectual property information. * Enabling formal information retrieval, natural language and semantic processing research to grow into the field of applied sciences in the global, industrial context. * Development and integration of different information access methods. * Research on effective methods for interactive information retrieval.Semantic supercomputing
Current technologies to extract concepts from unstructured documents are extremely computationally intensive. To allow interactive experimentation with rich and huge text corpora, the IRF has built a high-performance computing environment, into which the latest technological advances have been implemented: * multi-node clusters (currently 80 cores, up to 1024) * The highest speed interconnect technology * single system image with large compound memory (currently 320 GB, up to 4 TB) * fully integrated configurable computing (currently 4 FPGA cores, up to 256) The combination of these HPC features to accelerateThe World Patent Corpus
The IRF aims to bring state-of-the-art information retrieval technology to the community of patent information professionals. The IRF expects information retrieval (IR) technology to become the focus of information technology very soon. All industry sectors can profit from applying modern and future text mining processes to the special requirements of patent research. Although all ideas and concepts are universally applicable to all sorts of intellectual property information, patents require the most sophistication, and confront us with challenging technical and organizational problems. The entire body of patent-related documents possibly constitutes the largest corpus of compound documents, making it a rewarding target for text mining scientists and end-users alike. What’s more, patents have become a crucial issue, in particular for large global corporations and universities. The industrial users of patent data are among the most demanding and important information professionals. As a consequence, they could benefit the most from technology that relieves the burden of researching the large body of patent information.Research collections
The IRF provides several test data collections that have either been developed by the IRF, by one of its members or by third parties. These data collections can be used freely for scientific experimentation. The Matrixware Research Collection ( MAREC) is the first standardized patent data corpus for research purposes. It consists of 19 million patent documents in different languages, normalized to a highly specific XML format. The collection has been developed by Matrixware for the IRF. The ClueWeb09 collection is a 25 terabyte dataset of about 1 billion web pages crawled in January and February 2009. It has been created by the Language Technologies Institute at Carnegie Mellon University to support research on information retrieval and related human language technologies.References
{{Citation style, date=October 2022External links