Enhancing Text Readability using Deep Learning Techniques
PRESENTER:
Wejdan Alkaldi
University of Ottawa
ABSTRACT:
In the information era, reading becomes more important to keep up with
the growing amount of the knowledge. The ability to read a document vary
from person to person depending on their skills and knowledge. It also
depends on the readability level of the text whether it matches the
reader's level or not. We propose a model that uses
state-of-the-art technology in machine learning and deep learning to
classify and simplify a text taking into consideration the reader's level
of reading. The model classifies the text to its appropriate
readability level. If the text readability level is higher than the reader
level, i.e., too difficult to read, the model would perform a text
simplification to the desired level. The classification model is
trained against the readability levels found in Newsela corpus
(https://newsela.com/data/). Once the classification model is trained, it
will be used to classify more corpora for text simplification. Then the
simplification model will be trained to simplify a given document to match
a specific readability level. The model will be able to generate several
simplified versions of a given document based on the readability level
provided. The simplification will be done on a paragraph level, rather
than sentence level. It would also include sentence splitting when
appropriate. The model would help people with low literacy read and
understand any documents they require. It will also be beneficial to
educators assisting readers with different reading levels.