Abstract:
A math word problem (MWP) is a mathematical problem expressed using
natural language. In this research, elementary level set-related word problems in
which information is given in set notation are considered. As per our knowledge,
this is the first research addressing set theory related word problems.
This research introduces an abstract representation to interpret mathematical
semantics of set expressions and relations between sets. Two methods to extract
given set related expressions were implemented: rule based method and a statistical
method. Results show that statistical method is more robust to typing errors
and unexpected expression formats. A parser based on a context free grammar is
introduced to validate set related expressions and give feedback to the user when
there are incorrect expressions. Along with these functionalities, we present a
complete set problem solver system that understand and solve a given set word
problem.
In addition to the solver, we experiment in extracting mathematical expressions
from unstructured plain text using sequential classifiers. Several sequential
classification models including conditional random-fields (CRF) and Long-Short
Term Memory (LSTM) networks were compared with word and character level
features. The results show that using character level features significantly increase
the performance of mathematical expression extraction.
Citation:
Fernando, K. (2019). Automatic answer generation for math word problems [Master’s theses, University of Moratuwa]. Institutional Repository University of Moratuwa. http://dl.lib.mrt.ac.lk/handle/123/15815