Abstract:
The social media platforms serve as forums where people from the different strata of society, across the world, voice their unhindered opinions on a variety of subjects. Such platforms have been utilized to mine patterns that reflect the viewpoints, thought processes, and attitudes of the people. Sometimes the knowledge attained from social media messages highlights the emergent, actionable needs of the people. The rising anxiety and several problems, showcased by the tweets related to menstruation during the pandemic, serve as a ready reckoner for the same.
Psychological studies have established that people worldwide were exposed to grave and persistent psychosocial stressors due to the COVID-19 pandemic. It has also been established by studies that women are more affected by stress. Fear, anxiety, depression, and emotional instability caused due to stress affect the hypothalamic-pituitary-gonadal (HPG) and hypothalamic-adrenal axes, leading to menstrual cycle irregularities, dysmenorrhea, pre-menstrual symptoms, and menorrhagia. Menstrual health is a significant determinant of a woman's overall health and quality of life and is a contributor to the socio-economic burden on women, their families, and society, in general. Research to study the effect of the COVID-19 pandemic on women's menstrual health is an imperative public health initiative and is the need of the hour. It is disappointing that only a handful of medical, survey-based studies have been conducted so far in this important area that affects almost half of the world's population. The work presented in this paper is intended as a novel, small step in the direction of the stated aspiration and urges the machine learning community to break the taboo, and start talking about menstrual health to complement the efforts of the medical fraternity. Technologies like artificial intelligence, data mining, and machine learning have the prowess of algorithmically harnessing gigantic datasets, sophisticated pattern detection, and presenting the knowledge discovered in an understandable form. This ready-to-be-leveraged knowledge can be used by domain experts for drawing conclusions and inferences. The framework presented in this paper uses a blend of supervised and unsupervised data mining techniques to uncover knowledge. The framework is based on the principle of knowledge differentiation and uncovers knowledge at two different levels of abstraction. This allows for analysis of the discovered knowledge from multiple perspectives, enabling its consolidation, comprehension, and actionability.
At the first level, facts and myths being circulated on the net about the women's mensural health during the pandemic are discovered. Facts must be separated from the myths circulated on Twitter, in order to understand the authentic opinions and problems of people. Myths themselves prove as an important resource to understand the reasons behind people's responses and reactions to the government's policies. Identification of myths is also important so that public awareness plans can be launched and myth busters issued by the concerned government agencies. At the second level of abstraction, association rules are discovered from the selected categories of tweets, classified at abstraction level one. To the best of our knowledge, this is one of the first works leveraging the use of association rule mining for deriving meaningful knowledge about menstrual health from tweets. Upon unearthing the associations between the frequently used words in the tweets, they are subjected to four-stage postprocessing rule filters, that enable the bi-directional analysis of these linkages, in order to aid consolidated inferencing about selected classes of tweets.
Results of the framework presented in the paper show that the facts discovered by the system correctly identified the menstrual problems faced by women during the COVID pandemic, and also pointed to COVID-related stress as a probable cause for the same. Some very popular rumors were discovered by the presented framework that could represent some of the valid reasons behind vaccine hesitancy. The framework was also able to discover important associations from facts and myths about menstrual health and the pandemic being circulated on social media, in the form the tweets. The results are also able to clearly establish the efficacy of discovering association rules from tweets, especially when compared to the works that are devoted to just mining the top frequent words in tweets, say, via techniques such as clustering.