Abstract:
The COVID-19 pandemic may have caused unanticipated behavioral changes
that may have led to the upward trend in reported suicidal attempts. COVID-
19 period was one reason to change pre-existing mental health, domestic
violence, anxiety, and depression. And self-isolation and quarantine, may raise
feelings of isolation, sadness, alcohol and drug use, and suicidal thoughts.
When compared, people's mental health changed before, during and after the
pandemic period. Therefore, gathering information on the prevalence of
suicide and suicide attempts worldwide during the pandemic is essential and
crucial. Twitter is one of the main places to share people's thoughts in different
situations. Here we have focused on tweets of those who had suicidal thoughts
and tweets of others who tell the facts of people who have committed suicide.
The labeling was done by reading the full tweets manually one by one instead
of observing only the keywords to increase the labeling accuracy. Out of the
9750 tweets, 3200 expressed suicide-related to COVID-19 and 3000 expressed
suicide-related to other reasons. This study's main objective is sentiment
analysis of suicide-related tweets whether they are Covid-19-related suicides
or not. After collecting data, pre-processing was completed. Then the feature
vectors are produced to apply machine learning algorithms like Long Short-
Term Memory, Artificial Neural Networks and Support Vector Machine to
create a forecast paradigm for suicides during the pandemic. In terms of
classification, the findings demonstrated that ANN performed better than
SVM and LSTM with 91.33% accuracy as well as it had higher recall,
precision, f-measure, and lowest errors.
