dc.description.abstract |
The outbreak of COVID-19 has set off a worldwide well-being disaster that
affects how we view the world and direct our everyday lives. The impact has
given both positive and negative results. One of the negative results of the
COVID-19 pandemic is Home Violence (HV). HV encompasses many misuses,
including physical abuse, sexual abuse, emotional abuse, and controlling
behavior in a close relationship. People are becoming more reliant on social
media platforms like Twitter, Instagram, Facebook, YouTube, etc. Twitter has
recently emerged as an excellent resource for studying COVID-19 user-generated
material and behaviors in real-time. Analyzing HV-related posts on social media
is beneficial in gauging public sentiment toward sensitive problems, public
expression of feelings, and resource sharing regarding the otherwise personal
experience of HV. Our research proposed a method to analyze the HV incidents
using social media during the COVID-19 pandemic. More than 20,000 Tweets
were retrieved between 2020 April to 2021 July using Twitter API. Data pre
processing and word embedding were done, respectively. Then, to construct the
model, the data set was split into training and testing datasets to detect HV-related
Tweets; a deep learning model, LSTM, with different word embedding
techniques, was used in this research (TF- IDF+LSTM,
BOW+LSTM,
Word2Vec+LSTM,
GloVe+LSTM,
and BERT+LSTM). After, HV-related
Tweets are classified into three main topics: HV incident, HV awareness, and HV
shelter with the help of LSTM with GloVe embedding. Finally, 5W proposed a
model introduced to describe the HV incident Tweets and It’s including ‘What’,
‘When’, ‘Where’, ‘Who’, and ‘Why’ elements. With an accuracy of 89.56%, the
BERT+LSTM model surpassed the other implemented models. The proposed
GloVe+LSTM achieved an accuracy of 98.35% to classify the HV Tweets into
three main categories. HV incidents Tweets reveal that the proposed 5W model
performs well in describing the HV incidents. |
en_US |