#Walangpasok on Twitter: Natural language processing as a method for analyzing tweets on class suspensions in the Philippines

In this study, we aim to prove that natural language processing (NLP) can be used as a method for analyzing qualitative data like tweets. To validate this, we examine various topics emerging from tweets about class suspensions in the Philippines using #walangpasok (no classes). By utilizing three NLP techniques, we presented various topics. Using Latent Dirichlet Allocation (LDA), the study identified the following: (1) information dissemination and public announcements from government offices, and (2) sentiments of students. Through Word2Vec, we generated (1) monitoring and dissemination of alerts and warnings, (2) local government accountability, and (3) sentiments of Twitter users. An intrinsic evaluation was conducted for word2vec model using cosine similarity. The word-groups have an average cosine similarity of 0.997811. Lastly, the topics that emerged using K-means are (1) weather advisory, (2) class suspension announcements, (3) role of local government, and (4) users' sentiments and outlook towards a situation. The number of clusters generated by the k-means clustering algorithm was decided based on the silhouette score of 0.0153423865462. It turns out that the study provided the same results using NLP techniques. We also performed open coding to analyze the data manually and to ensure that the obtained results using the techniques are accurate. Thus, the use of NLP as a method of qualitative data analysis can be considered reliable and may be recommended to use in other qualitative research, particularly in the field of social sciences.