تحلیل احساس پست های شبکه های اجتماعی در بحران کرونا با استفاده از خوشه بندی دو مرحله ای

در بحران کرونا با طیف وسیعی از افکار، احساسات و نگرش ها در شبکه های اجتماعی مواجه ایم. دستیابی به درک جامعی از نگرش های جامعه نیازمند پردازش این داده‌هاست. هدف این پژوهش شناسایی ویژگی پیام هایی است که منجر به قطبیت های احساسی مختلف در شبکه های اجتماعی می شوند. در این پژوهش از پست های فارسی توییتر، اینستاگرام، تلگرام و کانال های خبری و تکنیک‌های پردازش زبان طبیعی استفاده شده است. در روش پیشنهادی این پژوهش، خوشه بندی دو مرحله ای مبتنی بر شبکه عصبی خود سازمانده و k-میانگین استفاده شده است. نتایج نشان دادند پست های حوزه سلامت و فرهنگ با قطبیت منفی، به احساساتی مانند ترس، تنفر، غم و خشم منجر شده است. پیام های مربوط به عملکرد هیجانی و نادرست مردم با احساس غم، ترس و استرس همراه است و امید در جامعه را کاهش داده است.

sentiment analysis of social media posts in the corona crisis using two-stage clustering

during the corona crisis, we are confronted with a diverse array of thoughts, feelings, attitudes, and behaviors expressed on social media. this data holds valuable insights for individuals and administrators seeking to effectively respond to the crisis. the goal of this study is to identify specific attributes of messages that generate different emotional reactions. the study aims to examine the content shared by twitter, instagram, and telegram users, as well as news pertaining to the covid-19 pandemic in iran. the dataset extracted from these social networks focuses on the period between january 21 and april 29, 2020, encompassing content shared within iran, and in persian language.it is important to mention that the dataset and its corresponding labels were published by the cognitive sciences and technologies council (cstc) in iran. initially, the content of each post underwent pre-processing, which involved the removal of stop words, word normalization, tokenization, and stemming. the emotional labels utilized plutchik's model and encompassed emotions, such as joy, trust, fear, surprise, sadness, anticipation, anger, disgust, stress, or other emotions. for this study, clustering algorithms were employed to analyze the social media posts.a two-stage clustering method was employed in this study. the proposed clustering algorithm combined the self-organized neural network and k-means algorithms. following our proposed algorithm, the data underwent clustering using the self-organized neural network (som) as the initial step. the resulting cluster centers from som were then utilized as the starting points for the k-means algorithm. the implementations were developed using python version 3.7 and matlab r2015a. hazm tools were utilized for data pre-processing, while the clustering process was performed in matlab. the davies-bouldin clustering evaluation metric was employed to determine the optimal number of clusters.the measure was computed for the number of clusters ranging from 2 to 50 within the two-stage clustering method. the findings indicated that the optimal number of clusters was ten. examination of the outcomes revealed that posts related to health and culture, characterized by a negative sentiment, evoked emotions, such as fear, hatred, sadness, and anger. messages concerning individuals' emotional well-being and improper functioning induced feelings of sadness, fear, and stress, contributing to a decline in societal hope. the results demonstrated a strong association between anger and disgust, while fear, stress, and sadness exhibited a positive correlation. to mitigate negative emotions and foster trust in authorities, we recommend providing clear, comprehensive information about the coronavirus pandemic.