Exploiting semantic annotations in math information retrieval

This paper describes exploitation of semantic annotations in the design and architecture of MIaS (Math Indexer and Searcher) system for mathematics retrieval. Basing on the claim that navigational and research search are `killer' applications for digital library such as the European Digital Mathematics Library, EuDML, we argue for an approach based on Natural Language Processing techniques as used in corpus management systems such as the Sketch Engine, that will reach web scalability and avoid inference problems. The main ideas are 1) to augment surface texts (including math formulae) with additional linked representations bearing semantic information (expanded formulae as text, canonicalized text and subformulae) for indexing, including support for indexing structural information (expressed as Content MathML or other tree structures) and 2) use semantic user preferences to order found documents.

The semantic enhancements of the MIaS system are being implemented as a math-aware search engine based on the state-of-the-art system Apache Lucene, with support for [MathML] tree indexing. Scalability issues have been checked against more than 400,000 arXiv documents.