Context guided retrieval of math formulae from scientific documents

Abstract

With increasing popularity of math formula search, a number of Math Information Retrieval (MIR) systems have come into being. Characterized by either conventional text-based techniques or newly evolved formula–centric indexing and search techniques, majority of such MIR systems exhibit applaudable performance. Nonetheless, it is common to observe certain limitations of existing MIR systems. For example, context guided retrieval of math formula, whereby formula can be retrieved given its context, is less researched. To this end, this paper discusses implementation of an MIR system, which employs pattern– based discovery to extract context–formula pairs, creates a Knowledge Base (KB) of such pairs, creates an inverted index for the contexts inside KB, and eventually coordinates with index and KB to retrieve relevant formula search results for a text query. Obtained search results and values of evaluation parameters reflect promising performance and competence of the approach.

Subject Classification:

• Machine Learning
• Math Information Retrieval
• Natural Language Processing

Keywords:

• Math Information Retrieval
• Formula Context
• Knowledge Base
• Indexing
• Apache Lucene
• NTCIR