MathSBERT: A Language Representation Model for Mathematical Information Retrieval
Pankaj Dadure, Partha Pakray, Sivaji Bandyopadhyay
Source abstract
The continuously increasing research in science, engineering, and mathematics has generated a huge amount of scientific information. This scientific information is enclosed with textual and mathematical information. The existing information retrieval systems are able to search the scientific information using the sequence of keywords, however, they are insufficient to retrieve this scientific information using the formula based query. To cope-up with this challenge, in this paper, we have investigated the proficiency of the Sentence-BERT model to retrieve scientific documents. Herein, the textual & mathematical formula has been treated and preprocessed separately and encoded using the Sentence-BERT model. The encoded embedding of these textual and mathematical formulas has been compared with query embedding and estimated the cosine similarity. The proposed approach has been tested on the Math-TagArticles of Wikipedia of NTCIR-12 and measured their proficiency in terms of precision at 5, 10, 15, and 20. The obtained results have shown that the language representation model can capture the syntax and semantics of the mathematical formulae.
Evidence graph
No public relationships recorded yet.
Integrity note: This page is a factual metadata record created by deterministic ingestion. It is not a claim that the work moves a mathematical frontier or has been independently verified.