15 minute read
How well are your students complying with COVID-19 safety measures?
7. Rate your fear level of contracting the Omicron-variant at your college 8. Do you know of anyone who tested positive for COVID19? 9. Which of the following is correct?
10. How would you rate your Omicron-variant worry-level when resuming to work? compliant fear_rate 1-2-3-4-5 F-1;F-2;F-3;F4;F-5
Advertisement
casePositive Yes; No; Prefer not to say +Yes; -No
awareness There is a drug to treat COVID-19; there is a vaccine for COVID19;there is both a drug for treatment and a vaccine for COVID-19;I am unsure which is correct Awareness+; Awareness-
worry_level 1; 2; 3;4; 5 W-1; W-2; W-3; W-4; W-5
11. In which zone of the country is your college of education located
12. Which of these best describes the community where your college is located? 13. How often do you wash your hands daily? Zone North-East; North-West; North-Central; South-West; South-South; South-East college_loc City; town; village/rural area
hand_wash Always-oftensometimes-rarelynever NE; NW; NC; SW; SS; SE
City; town; village
W-A;W-O;WS;W-R;W-N
14. Are you working under the fear of contracting COVID-19 from office? GROUND TRUTH Yes; No F-Yes; F-No
15. What do you think about the reopening of colleges for academic activities Opinion Open ended Not applicable
C. REPTree Predictive Modelling Machine learning is a branch of artificial intelligence that trains algorithms with data, in either supervised or unsupervised learning approach, and thereby gives capabilities to learn from data without being explicitly programmed and in turn make informed decisions after the learning process in what is referred to as testing the knowledge gain [22]. REPTree is reputable as a quick decision tree learner-algorithm which constructs a decision or regression tree with information gain as the splitting methodology, and prunes it with reduced error pruning method as it results in a more accurate classification tree; size of training and testing notwithstanding. In this work, a 10-fold cross-validation approach as described in Fig. 3 is adopted on the WEKA machine learning software for both the REPTree training and the testing dataset in 60:40 ratio. In the 10-fold cross-validation, the 14:2376 (attribute: instance) sample is randomly apportioned into 10 equal size sub-samples out of which one subsample is reserved as the validation set for testing the model, and the remaining 9 sub-samples are used for training the REPTree algorithm in section Fig. 2. The crossvalidation process is iterated in 10 clocks (the folds), with each of the 10 sub-samples used precisely once as the testing data while the 10 results from the folds is then averaged to generate a distinct estimation. This approach ensures all attributes: instances are deployed for both training and validation while each subset is used for validation once each.
D. Topic Modelling of Preprocessed Data Topic modelling abstract subjects in data corpus based on word clusters alongside their corresponding frequency contained in each text document [23]. It is often deployed for natural language processing (NLP) to uncover topical subjects and thereby extract semantic meaning from unordered text documents in applications such as opinion and text mining and in information retrieval systems. This study aims to deploy topic modelling as part of mechanisms to facilitate in-depth understanding of emotions expressed in the opinions of respondents concerning COVID-19 in an academic environment. The topic modeling widget of the orange data mining tool including the Latent Dirichlet Allocation (LDA), Latent Semantic Indexing (LSI), and Hierarchical Dirichlet Processing (HDP) algorithms are implemented on the attribute 15 text-responses earlier presented on Table 1. LDA is a three-level ranked Bayesian system such that each item of a document is modeled as a fixed mixture over a primary set of topics. LSI however yields topics with negative and positive keywords with negative and positive weights on the topic. The positive weights represent words highly representative of the topic and which influences its occurrence while for negative weights, the topic seldom occur if they appear less in it. Multidimensional scaling (MDS) is used to visualize the modelling, as a low-dimensional projection of the topics represented as points by fitting distances between the points. The proposed topic modelling and sentiment analysis framework is presented in Fig. 4.
E. VA DE R-b a se d Lex i c o n Sentiment Analysis The sentiment analysis uses polarity to classify the opinion expressed by respondents into three categories of Positive, Neutral & Negative while its VADER-approach scores each word-token, as contained in each text-opinion. The approach computes the sentiment scores assuming sentiment is related to the presence of certain known-words or phrase (for bi-gram or more) in an opinion structurally represented. Therefore, opinions are assigned certain sentiment value referred to as lexicon. In line with the work of [24], the occurrence frequency of each word in the dictionary, as used by the respondent, determines the computations of its positive, negative or its neutral state. This study employ the polarity score calculator of the VADER model. The model yields the fourth attribute, the compound score, for each opinions expressed by correspondents in answering poll question 15. The compound score represents an accumulated description of the first three negativity, positivity, neutrality scores, and is computed for each opinion expressed as:
x=
x √x +α (1)
where is the sum of valence scores of component words, and α is default to 15 as the normalization constant. The compound score is then regularized between -1 and +1 which represents the positivity and negativity of each opinion. The VADER-based sentiment analysis framework of this study is as described in Fig. 4.
Fig. 2. Description of stages for the REPTree fear-stimuli detection model
Fig. 3. Concept of machine learning 10-fold cross-validation adopted in the study
Fig. 4. Framework of the topic modelling and sentiment analysis
IV. RESULT & DISCUSSION
Results obtained are discussed in this section. The machine learning phase of the study, using REPTree learning algorithm, is implemented on the preprocessed data to produce the proposed fear-stimuli detection system in the WEKA data mining software on a 1.7 GHz Intel Core i3 CPU with 4 GB RAM. The resulting decision tree model, predicting the Omicron fear-stimuli from the 14no data attribute, is presented in Fig. 5 returning attribute_id ‘hand_wash’ as the major cause of fear in the academic community. The result shows a tree of size 18 built in 0.02 seconds with the 13th attribute at the root node, haven possessed the highest information gain out of the entire 14-no attribute data table. The predictive accuracy of the experimental model is with a 94.68% accuracy level hence the encouraging Precision and Recall weighted average figures. While the Receiver Operator Characteristic (ROC) weighted average shows how the number of accurately classified fearful/fearless instances varies with the number of incorrectly classified instances for the binary problem, the precision and recall weighted averages underscores the efficient performance of the model.
1st iteration
2nd iteration
3rd iteration
10th iteration 14:2376 Training_set
Training folds Test fold
E1
E2
E3
E10
E=1/10∑10
i=1 Ei
The true positive rates of the class labelling, noted as Recall with 0.947 and same value for instances truly classified as positives, representing the Precision, further validates the reliability of decisions made by this proposed model. Experimental results from the natural language processing of opinions expressed in attribute 15 with the question, ‘What do you think about the reopening of TEIs for academic activities’, returns results across the different stages of the framework. The tokenization of the text-corpus computed in a uni-gram approach, claculates the weights of each word contained in the corpus. A document frequency of 0.00-1.00 was set for the scheme and its visualization is shown in Fig. 6 through a word cloud. The size of each uni-gram shows their frequency in the corpus, returning words like hand, wash, washing, important, students, hardly, etc. as those with profound emphasis in the corpus. Result from the LDA and LSI topic modelling, with 9-topic configuration, is presented in Fig. 7. As could be observed, LSI and LDA contains consistent words in topics 4 and 3 respectively with tokens namely: government, college, hardly, wash, hand, water, and nose, which distinctly models topical issues in the minds of respondents. The sentiment analysis part of the framework returns the positive, negative, neutral and compound polarities for each text-respnse to the same attribute 15, indicating emotions expressed therein. In the result, only 25.8% of the respondents expressed opinions clustered as negative sentiments, indicating the state of mind of the majority as either positive or neutral while expressing their opinion about the Omicron-variant and school reopening. Opinions clustered as negative sentiments has compound scores <= -0.05 while compound score >= 0.05 are clustered as positive sentiments with neutrality sentiment captioned between compound score > -0.05 and < 0.05. Fig. 8 shows the outcome of the sentiment analysis of the most negative opinions expressed by correspondents.
Fig. 5. Prediction result of the REPTree fear-stimuli detection model
Fig. 6. Word cloud of text corpus
Fig. 7. LSI and LDA topic modelling with generated keywords respectively
Fig. 8. Result of sentiment analysis
From the REPTree experimental result, handwashing garnered highest information gain hence at the root of the tree, being the deciding factor of the predictive model. A critical study of the decision tree reveals that respondents who wash their hands ‘always’ (W-A) regularly, from the south-west, south-south and north-east zones of Nigeria, are yet Omicron-variant fearful while counterparts from the south-east, north-central and north-west does not dread the variant. The decision tree shows those who never wash their hands are understandably fearful of the variant just as Junior staff, who ‘often’ wash their hands but opines students’ compliance with safety measures are poor with those who feel students’ compliance are good, does not in any way exhibit COVID19 fears. Similarly, senior staff who ‘often’ wash their hands does not exhibit fears while discharging their daily duties; whereas, respondents who ‘rarely’ and ‘sometimes’ wash their hands still does not exhibit Omicron fear according to the machine learning modelled outcome. Result further implies that issues relating to face mask wearing, individual COVID-19 test status, COVID-19 case count, and staff awareness of COVID-19 updates does not have significant information gain to stimulate the Omicron-variant fears in the academic communities. Regular wearing of nose masks likewise does not necessarily implies fear of the variant while experimental result suggest that issues about college precautionary measures and students’ safety measures prominently matters to staff of TEIs in Nigeria, and could stimulate Omicron fear.
I am so worried about what we…
Delta more contagious than other… herbal treatment 100% guarantee… the rapid increase hospitalized… Do not forget like & share this page all unvaccinated must maskup &…
Most students poor cant afford… Folks going find out hard way due emerging data on reduced… -1 -0.5 0 0.5 1 1.5
pos neg neu compound
The sentiment analysis result shows a higher neutral sentiments as expressed by the entire academic communities such that most of their item 15th opinions are hardly of optimism but more of pessimism as regards the subject of the Omicron-variant and school reopening. Their negative sentiments is almost eroded by the neutrality inherent of their opinions. This is a pointer to their constructive criticism of government’s decision for school opening in the face of Omicron-variant prevalence, which echoes their concern and yearn for non-clinical preventive measures other than criticizing the decision to open schools. As may be observed from the topic modelling and word cloud results, compliance with non-clinical safety measures remains topical in the minds of the TEI stakeholders and in relation to the output of the supervised REPTree decision tree prediction, issues surrounding handwashing, provision of water, and student’s compliance stimulates the Omicron-variant fear.
V. CONCLUSION & RECOMMENDATIONS
COVID-19 related personal opinions, especially over the Omicron-variant, acquired from Tertiary Educational Institutions is captured for a predictive modelling experiment. REPTree machine learning algorithm, VADER-based sentiment analysis, and Topic modelling techniques were executed to uncover stimulus of COVID-19 fears in educational communities across Nigeria in the face of the Omicron invasion and government’s decision to open schools. The resulting decision tree model is validated with Precision, Recall and ROC area performance metrics yielding a state-of-the-earth prediction accuracy of 94.68% and similarly to the topic modelling of textual corpus, the experimental model returns the attribute hand-wash as the most significant possessing the highest information gain towards determining the Omicron-variant fear-stimuli in college communities. The sentiment analysis of the opinions expressed likewise shows more of neutral sentiments, lower negative and positive sentiments concerning government decision to open schools in the face of the new wave. Consequently, the fear rate of the virus amongst staff of TEIs is a function of issues associated with compliance of stakeholders with non-clinical precautionary measure of regular hand washing and other issues associated therewith. The palpable fret being exhibited by staff of TEIs therefore need to be addressed by school operators and regulators with the mind view of ensuring adequate compliance with regulations put in place by government for tertiary institutions which will in no small way tame the fear-trend of staff towards the third wave of the virus. Awareness of COVID-19 case-count across the country, the attribute casePositive, which checks if respondents know of any COVID-19 positive acquaintance, surprisingly does not stimulate fear in staff. A research with wider dataset scope is recommended for future work while ensuring a balanced class representation in the training set.
REFERENCES
[1] P. C. Addo, F. Jiaming, N. B. Kulbo and L. Liangqiang, "COVID-19: fear appeal favoring purchase behavior towards personal protective equipment," The
Service Industries Journal, vol. 40, no. 7, pp. 471-490, 2020. [2] WHO, "Coronavirus disease 2019 (COVID-19) situation report," World Health Organization, 8 April 2020. [Online]. Available: https://www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports. [Accessed 8 April 2020]. [3] H. Plecher, "Age structure in Nigeria," Statista, 28 July 2021. [Online]. Available: https://.statista.com/statistics/382296/age-structure-in-nigeria/. [Accessed 1
December 2020]. [4] NCDC, "COVID-19 daily Statistics," Nigerian Centre for Disease Control, 12 January 2022. [Online]. Available: https://covid19.ncdc.gov.ng/. [Accessed 12
January 2022]. [5] L. Wen, H. Wang, L. Yuxing and L. Li, "Psychological status of medical workforce during the COVID-19 pandemic: A cross-sectional study," Psychiatry research, vol. 288, no. 112936, 2020. [6] S. Taylor, L. Caeleigh, M. Paluszek, T. Fergus, D. McKay and G. Asmundson, "Development and initial validation of the COVID Stress Scales," Journal of
Anxiety Disoders, vol. 72, no. 102232, 2020. [7] T. O. Olaleye and O. R. Vincent, "A Predictive Model for Students’ Performance and Risk Level Indicators Using Machine Learning," in International
Conference in Mathematics, Computer Engineering and Computer Science, Lagos, 2020. [8] O. Arogundade, C. Atasie, S. Misra, A. Sakpere, O. Abayomi-Alli and K. Adesemowo, "Improved Predictive System for Soil Test Fertility Performance Using
Fuzzy Rule Approach," in 2nd International Conference on Soft Computing and its Engineering Application, India, 2021. [9] F. Ornell, J. Schuch, A. Sordi and F. Kessler, "Pandemic fear” and COVID-19: mental health burden and strategies," Brazilian Journal of Psychiatry, vol. 42, no. 3, pp. 232-235, 2020. [10] S. J. Eder, D. Steyrl, M. M. Stefanczyk, M. Pieniak, J. M. Molina, P. Ondra, B. Jakub, S. Patrick, F. Scharnowski and A. A. Nicholson, "Predicting fear and perceived health during the COVID-19 pandemic using machine learning: A cross-national longitudinal study," Plus one, vol. 16, no. 3, 2021. [11] E. C. Green and W. Kim, "Fear arousal, sexual behavior change and AIDS prevention," Journal of Health Communication, vol. 11, no. 3, pp. 245-259, 2006. [12] S. Taylor, "The psychology of pandemics: Preparing for the next global outbreak of infectious disease," Cambridge Scholars Publishing, 2019. [13] G. J. Asmundson and S. Taylor, "Coronaphobia: Fear and the 2019-nCoV outbreak," Journal of anxiety disorders, vol. 70, no. 102196, 2020. [14] V. Giallonardo, S. Gaia, V. D. Vecchio, M. Luciano, U. Albert, C. Carmassi, G. Carra, F. Cirulli, B. Dell'Osso, M. Nanni and M. Pompili, "The impact of quarantine and physical distancing following COVID-19 on mental health: study protocol of a multicentric Italian population tria," Frontiers in Psychiatry, vol. 11, p. 533, 2020. [15] S. A. Lee, "Coronavirus Anxiety Scale: A brief mental health screener for COVID-19 related anxiety," Death studies, vol. 44, no. 7, pp. 393-401, 2020.
[16] N. O. Okoi and J. E. John, "Nosophobia, hypochondriasis, and willingness of people to seek healthcare amidst the COVID-19 pandemic in Calabar Metropolis of Cross River State, Nigeria," Open Journal of Psychiatry & Allied Sciences, vol. 12, no. 1, pp. 36-42, 2020. [17] S. Paul, "Nosophobia of corona: A loss or gain for India?," International Journal of Health & Allied Sciences, vol. 9, no. 4, pp. 367-367, 2020. [18] G. Mertens, L. Gerritsen, S. Duijndam, E. Salemink and M. Iris, "Fear of the coronavirus (COVID-19): Predictors in an online study conducted," Journal of
Anxiety Disorders, vol. 74, no. 102258, 2020 [19] P. Kaila, Dr. and D. A. Prasad, "Informational Flow on Twitter - Corona Virus Outbreak – Topic Modelling Approach," International Journal of Advanced
Research in Engineering and Technology, vol. 11, no. 3, pp. 128-134, 2020. [20] F. Martin and M. Johnson, "More Efficient Topic Modelling Through a Noun Only Approach," in Proceedings of Australasian Language Technology
Association Workshop, 2015. [21] T. Olaleye, T. Arogundade, A. Abayomi-Alli and K. Adesemowo, "An Ensemble Predictive Analytics of COVID-19 Infodemic Tweets using bag of words," in
Data Science for COVID-19, Singapore, Elsevier, 2021. [22] T. Olaleye, O. Arogundade, C. Adenusi and S. Misra, "Evaluation of Image Filtering Parameters for Plant Biometrics Improvement Using Machine Learning," in 2nd International Conference on Soft Computing, India, 2021. [23] T. Sherstinova, O. Mitrofanova, T. Skrebtsova, E. Zamiraylova and M. Kirina, "Topic Modelling with NMF vs. Expert Topic Annotation: The Case Study of
Russian Fiction," in Mexican International Confeence on Artificial Intelligence, 2020. [24] H. Hota, D. Sharma and N. Verma, "Lexicon-based sentiment analysis using Twitter data: a case of COVID-19 outbreak in India and abroad," in Data Science for COVID-19, London, Elsevier, 2021, pp. 275-293. [25] NCCE, "List of colleges of education in Nigeria.," Nigerian Commission for Colleges of Education, 28 July 2021. [Online]. Available: http://www.ncceonline.edu.ng/colleges.php. [Accessed 28 July 2021].
