Abstract:
Development of experimental techniques and computational models
accelerated the pace of research and development in the eld of mathematics
and text mining. One of the major research area in the eld of automation
is solving word problems using Arti cial Intelligence (AI) techniques. The
existing research, for solving word problems using AI, is unable to provide
on the spot solutions for a given word problem. The existing approaches,
for solving algebraic and arithmetic word problems, have some limitations
that need to be addressed. They are limited to give high accuracy for
solving arithmetic problems only. They also require too much processing
resources and time for extracting and simplifying features from the text
data. In this research work, we proposed a template-based approach that
has been developed by following a two-step process. The rst step is to
predict an equation template from a training dataset. The next step is to
instantiate the predicted template with nouns and numbers. The features
extraction in the proposed approach only relays on POS tags, NER and
dependency relation; no need to extract all words from the given word
problem. To validate the proposed methodology, a prototype system has
been implemented. The system has been compared with the existing systems
using their respective datasets and the proposed dataset. The experimental
results show improvement in accuracy, with an average precision of 80.6% and
average recall of 83.5%. In future, we intend to incorporate the co-reference
resolution technique into our system to further improve its accuracy.
