Cancer Communication and User Engagement on Chinese Social Media: Content Analysis and Topic Modeling Study

Abstract

Background Cancer ranks among the most serious public health challenges worldwide. In China—the world’s most populous country—about one-quarter of the population consists of people with cancer. Social media has become an important platform that the Chinese public uses to express opinions. Objective We investigated cancer-related discussions on the Chinese social media platform Weibo (Sina Corporation) to identify cancer topics that generate the highest levels of user engagement. Methods We conducted topic modeling and regression analyses to analyze and visualize cancer-related messages on Weibo and to examine the relationships between different cancer topics and user engagement (ie, the number of retweets, comments, and likes). Results Our results revealed that cancer communication on Weibo has generally focused on the following six topics: social support, cancer treatment, cancer prevention, women’s cancers, smoking and skin cancer, and other topics. Discussions about social support and cancer treatment attracted the highest number of users and received the greatest numbers of retweets, comments, and likes. Conclusions Our investigation of cancer-related communication on Weibo provides valuable insights into public concerns about cancer and can help guide the development of health campaigns in social media.

Public Trust in Artificial Intelligence Applications in Mental Health Care: Topic Modeling Analysis

“…Before the analysis, we followed the standard preprocessing procedures designed in previous studies [ 37 , 38 ] to clean the data using Python 3.0 (Python Software Foundation) and to perform word part-of-speech tagging and text processing using the Python library spaCy [ 39 , 40 ]. Through data cleaning, we converted the words in the reviews into lowercase words; removed stop words, punctuation, numbers, and nonword characters; and stemmed the remaining text [ 41 ]. To generate more interpretable topics of high quality, we restricted the parts of speech of words to “noun” (NOUN), “verb” (VERB), “adjective” (ADJ), or “proper noun” (PROPN).…”

“…The statistical methods of unsupervised TM algorithms (which do not need prior labeling or annotations of the documents) were designed to analyze the words (terms) of the original texts to identify the themes (topics) running through a corpus [ 42 , 43 ]. These algorithms allow users to organize and summarize numerous documents that cannot be annotated manually [ 41 ], thereby revealing the hidden topics in the documents [ 43 ]. We adopted the LDA TM technique, which assumes that texts are generated from a mixture of topics [ 44 ].…”

“…LDA is efficient and can generate topics of better quality [ 45 ]. From the data set created, we generated 2 probability distribution outputs: the probability distribution of topics over documents and the probability distribution of terms over topics [ 41 , 43 ]. The number of topics was determined by repeating the analysis with different numbers of topics and by comparing the perplexity of each analysis [ 41 ].…”

“…From the data set created, we generated 2 probability distribution outputs: the probability distribution of topics over documents and the probability distribution of terms over topics [ 41 , 43 ]. The number of topics was determined by repeating the analysis with different numbers of topics and by comparing the perplexity of each analysis [ 41 ]. A lower perplexity value indicates a better model fit [ 44 ], and the perplexity value decreases with the increase in the number of topics [ 41 ].…”

“…The number of topics was determined by repeating the analysis with different numbers of topics and by comparing the perplexity of each analysis [ 41 ]. A lower perplexity value indicates a better model fit [ 44 ], and the perplexity value decreases with the increase in the number of topics [ 41 ]. Both the simplicity and the interpretability of the textual content need to be considered in choosing the optimal number of topics [ 38 ].…”