Mathematical formula similarity calculation method and system

Abstract

The invention discloses a mathematical formula similarity calculation method and system, and relates to the field of education, and the method comprises the steps: recognizing a formula contained in a text, and carrying out the structure conversion of the recognized formula, and obtaining an SLT tree structure and an OPT tree structure of the formula; constructing a corresponding tuple sequence for the SLT tree structure of the formula, and constructing a corresponding tuple sequence for the OPT tree structure of the formula; performing word segmentation processing on tuple sequences of the SLT tree structure and the OPT tree structure of the formula; training a FastText n-gram model, and obtaining two word embedding models which are trained completely are obtained; encoding the formula to obtain an SLT vector representation and an OPT vector representation after the formula word is embedded; obtaining vectors of the formulas through calculation based on the SLT vector representation and the OPT vector representation of the formulas, and obtaining the similarity between the formulas through calculation based on the vectors of the formulas. According to the method, vectorization expression can be accurately carried out on the formulas, and similarity calculation between the formulas can be accurately completed.

Description

Technical field [0001] The present invention relates to the field of education, in particular to a method and a mathematical formula similarity calculation system. Background technique [0002] Currently NLP (Natural Language Processing, NLP) industry, the text information embedded mathematical formula mainly two approaches: one is directly carried along to a word re-training text corpus formulas and, to give the corresponding words embedded vector; another feature element is to map the position vector equation, and the degree of matching calculated by the relative position of the degree of coincidence between the comparison element formulas obtained similarity between two formulas. There are also calculated by the mathematical expression based on similarity binary manner, for example, the formula is converted into a binary tree structure, then the structure of the variable normalized, according to mathematical expressions between the final common subsequence lengths calculated s...

AI Technical Summary

Problems solved by technology

[0003] However, the above-mentioned processing method has the following problems when calculating the similarity between formulas: 1. Most of the formulas are expressed in MathML format or LaTex format. How to accurately extract the formula and correctly express its internal information is the biggest difficulty faced by the above-mentioned processing method; 2. Treat formulas directly as text processing, and violently dismantling the formula structure may damage the important information hidden in the formula itself. For example, there is a certain order in the calculation of formulas. Semantic information; 3. Formulas are specific and have a huge magnitude. When dealing with new formulas, traditional methods cannot accurately vectorize them.

Images