The data set for this work was gathered from 2 of the most popular social media platforms: Twitter and Facebook. By using the full-archive search of the Twitter Academic Researcher Application Programming Interface, all the tweets posted since 2006 can be retrieved for a given query [63,64]. Using keywords related to Mpox, a query was built ((monkeypox OR “monkey pox” OR smallpox OR “viruela dei mono” OR “variole du singe” OR “variola do macaco”) -is:retweet) to gather all the tweets except the retweets, from May 1 to December 25, 2022. The tweets were cleaned. URLs, addresses with the “@” sign, and hashtag signs “#” were removed, and punctuations were corrected. Finally, 2,333,496 cleaned tweets were obtained. 124,712 tweets related to the 2SLGBTQIAP+ community were extracted using the following keywords: lgbtq, lgbtq+, gay, homosexual, homosexuality, lesbian, intersex, transsexual, transgender, bisexual, queer, “men having sex with men,” “men who have sex with men,” lgbt, lgbtqi, lgbt+, lgbtqi+, and lgbtq+ [65]. All the posts in 30 Mpox-related public Facebook groups from May 1 to December 25, 2022, were gathered using the Facebook_Scraper library [66]. Of the 16,114 retrieved posts, 712 had the 2SLGBTQIAP+ keywords after cleaning and were selected for analysis. All Facebook posts are public and gathered from public groups. The Twitter and Facebook data sets gathered with Mpox- and Mpox plus 2SLGBTQIAP+-related keywords or groups were combined and visualized by means of a word cloud (Multimedia Appendix 1).

The language of the tweets was retrieved using the Twitter application programming interface. Moreover, the language of Facebook posts was recognized using the pycld3 library [67]. The posts were in 102 different languages, and English, French, and Spanish posts with 1,972,637, 124,008, and 33,547 posts, respectively, had the highest frequency.

We have gathered public posts from public pages and public groups on Facebook, accessible by anyone through Facebook. We share the group IDs, page IDs, and post IDs of our data set [68], in compliance with the Association of Internet Researchers ethics [69] and the International Chamber of Commerce/European Society for Opinion and Marketing Research code of conduct [70]. Moreover, the Twitter data set, which is available at [71], includes only tweet IDs and user IDs and is used and shared under Twitter’s privacy policy agreement [72]. Since social media posts are passively analyzed in this research, informed consent from individuals is waived [73].

Topic modeling, a text mining tool for automatic discovery and extraction of hidden topics and semantic structures occurring within a text body or in a collection of documents, was done using the Latent Dirichlet Allocation model available in the sklearn package of Python (version 3.8.8; Python Software Foundation). Topic analysis was performed only on English posts. The optimal number of topics was calculated by maximization of the coherence and minimization of the Jaccard similarity scores [74]. Posts belonging to each topic with a probability higher than 0.7 were studied, and the main subject of concern for each topic was inferred.

In this paper, topic modeling was coupled with sentiment analysis, which is an NLP procedure that classifies a text based on its sentiment. Most sentiment analysis models classify a text into 3 classes: positive, neutral, and negative. However, some models classify a text into 2 classes: positive and negative. The sentiment score is a number between –1 and 1, which indicates the intensity of the sentiment. Generally, models that classify a text into 3 classes have a score close to 1, 0, and –1 for positive, neutral, and negative sentiments, respectively. Moreover, models that classify a text into 2 classes provide a negative score for negative sentiments and a positive score for positive sentiments. In this work, sentiment analysis was performed on English and Spanish posts using the pysentimiento package available on the Hugging Face website. It is estimated that this model, which classifies text into 3 classes, has a macro F₁ score of 0.705 [75-78]. Sentiment analysis was performed on French posts using CamemBERT, which classifies text into 2 classes and is estimated to have 94.55% accuracy [79]. Geotagged tweets were used to study topics and sentiments regarding the 2SLGBTQIAP+ community across different countries. Sentiments on different topics were compared using Mann-Whitney U, ANOVA, and Dunn tests, and studied across different countries using the Mann-Whitney U test.

The temporal trend of the number of the gathered posts related to Mpox and its epidemiology in terms of Mpox cases is depicted in Figure 1. From May 18, when Mpox began to emerge as an outbreak, the volume of the posts significantly increased until May 24, when they started falling. Since May 28, the volume of posts has stayed more or less steady. In more detail, the number of Mpox posts peaked on May 20 and on May 23, 2022, while the number of Mpox and the 2SLGBTQIAP+ community posts peaked on May 24, 2022, just 2 days after the Joint United Nations Programme on HIV and AIDS (ie, UNAIDS) urged media outlets, as well as institutional actors, including governments and communities, to respond to the outbreak with an evidence-based, data-driven, and, at the same time, inclusive and rights-based approach, avoiding attaching a stigma to the 2SLGBTQIAP+ community.

The number of posts concerning the Mpox and those specifically focusing on the relationship between the Mpox and the 2SLGBTQIAP+ community were highly correlated, as expected (correlation coefficient of 0.85, P<.001). Interestingly, the correlation between the number of Mpox cases and the number of posts related to the 2SLGBTQIAP+ community (correlation coefficient of 0.36, P<.001) was higher than the correlation between the Mpox cases and the total number of posts on Mpox (correlation coefficient of 0.24, P<.001). This shows how closely the discussions regarding Mpox on social media are related to the 2SLGBTQIAP+ community.

A total of 10 different topics were extracted from posts related to Mpox and 2SLGBTQIAP+. The topics indicate that the 2SLGBTQIAP+ population is heavily stigmatized for spreading Mpox. Table S1 in Multimedia Appendix 2 shows the identified keywords and the percentage of tweets on each topic that have them. The first 17 keywords shaded in gray are essentially related to Mpox and 2SLGBTQIAP+ and are common among almost all of the topics. The rest of the keywords dominantly belong to one of the topics. The same pattern could be observed in the word clouds created for each topic (Multimedia Appendix 3). This indicates that the topics are well separated and do not overlap. This is the result of maximizing coherence while minimizing the Jaccard similarity score. In other words, posts inside each topic are very much related, and posts from different topics are far from each other. By studying the posts that belong to each topic with a probability higher than 0.7, the subjects identified for each topic listed in Textbox 1.

Topic modeling indicated that the 2SLGBTQIAP+ population is highly stigmatized for spreading Mpox. Sentiment analysis shows that most sentiments are negative, then neutral, and only a few are positive. Since the posts are related to an outbreak after a pandemic and the stigmatization of minor populations, it is expected that the sentiments be mostly negative. English posts have the greatest number of negative polarities. Spanish posts have the fewest negative polarities, and a higher neutral polarity compared to English posts. Moreover, the negative polarity of French posts is significantly higher than the positive polarity (Figure 2). The P value of the Mann-Whitney U test indicates that the intensity of negative polarity is significantly higher than that of neutral and positive polarities (P<.001), and the intensity of neutral polarity is significantly higher than that of positive polarity (P<.001; Figure 2). The P value of the Mann-Whitney U test also indicates that English posts have a significantly higher negative intensity and lower neutral and positive intensities compared to Spanish and French posts (P<.001; Figure 2).

All the topics have a higher negative, then neutral, and finally positive polarity (Figure 3). Additionally, the Mann-Whitney U test shows that all the topics have significantly higher negative, then neutral, and finally positive intensities (Figure 3). However, the ANOVA test indicates that the sentiment scores of different topics are not very similar (P<.001; Figure 4). The Dunn test shows that 3 topics that are strongly related to the stigmatization of the 2SLGBTQIAP+ community, namely, “gay bathhouse/homosexuality/heterosexual; CDC and CNN news,” “Mpox outbreak linked to gay sauna/gay bars/Grindr/fetish festival; Avoid gay sex to protect yourself,” and “Mpox is a stigma against gays/African gays; Stigmatizing gays/African gays” have a significantly lower sentiment score compared to other topics (Figure 4).

The visualization of the geotagged tweets gathered on Mpox and Mpox plus LGBTQ keywords shows that countries that have the greatest number of tweets include the United States, the United Kingdom, Canada, Ireland, France, the Netherlands, Switzerland, Spain, Portugal, Germany, Mexico, Brazil, South Africa, Nigeria, Kenya, Pakistan, and India (Figure 5). The tweets extracted for the 2SLGBTQIAP+ community were mostly concentrated in the United States, the United Kingdom, Canada, Spain, Portugal, India, Ireland, and Italy. However, topic modeling of tweets related to Mpox and the 2SLGBTQIAP+ community was performed only on English tweets, which were mainly from the United States, the United Kingdom, Canada, and India.

After topic numbers 5 and 6, which are about the news and reporting the number of Mpox cases, topics 3, 4, 8, and 9 have the most popularity among different countries, which are all related to the 2SLGBTQIAP+ population being stigmatized for spreading Mpox (Figure 6). Since the posts were in English, it was possible to find the popularity of each topic only in 4 different countries: the United States, the United Kingdom, Canada, and India.

Sentiment polarities were found for English tweets in the United States, the United Kingdom, Canada, and India. Moreover, the sentiment polarities of the Spanish and French tweets were found only for Spain and France, respectively, since the volume was low for other countries. The 3 countries with the highest negative polarity are Canada, the United States, and the United Kingdom (Figure 7). The P value of the Mann-Whitney U test indicates that the distribution of sentiment scores across different countries is less diverse (Figure 7). However, among the countries that were studied for English tweets, Canada has the lowest sentiment score (P<.04).

On July 23, 2022, the World Health Organization (WHO) declared Mpox a public health emergency of international concern [80]. Ever since then, the number of cases around the globe has been increasing. Previously, minority populations such as gay and bisexual communities were blamed for spreading different diseases like HIV/AIDS and hepatitis B and C. The result of such stigmatization was more new cases, depression, mental health problems, and substance use [81]. The same trend is observable with the novel Mpox outbreak that is spreading around the world. This work aims to understand Mpox stigmatization of the 2SLGBTQIAP+ community using Twitter and Facebook.

Social media is becoming increasingly popular among people to share their opinions, ideas, and experiences. People are sometimes more honest on social media than in their real lives. Therefore, it is a reflection of the real world. As a result, it is used in many different areas of research, such as the economy [82], marketing [83], and health care [84].

A few studies have applied social media mining to Mpox. Ng et al [52] extracted from Twitter a body of 352,182 original tweets containing the terms “monkeypox,” “monkey pox,” or “monkey_pox,” in the English language, from May 6, 2022, to July 23, 2022, using Bidirectional Encoder Representations from Transformers named entity recognition. The authors identified 5 topics clustered into three major themes: (1) safety concerns, (2) sexual and gender minority stigmatization, and (3) a general lack of faith in public institutions. Tweets displayed high levels of partisanship and personal health anxiety.

In line with these findings, in this paper, Tweets and Facebook posts are used to discover the popular discussions regarding Mpox and the stigmatization of 2SLGBTQIAP+ communities, the hot spots, and the sentiments of different topics and countries. The results of this study could be used by health officials to combat stigmatization.

This investigation has a number of strengths, including its methodological rigor, transparency, and novelty, as well as its main focus on the 2SLGBTQIAP+ community. On the other hand, it is not without any limitations, which should be properly acknowledged. Sentiments analysis was performed on English, French, and Spanish tweets and Facebook posts, which had the highest frequency among all the different languages. Moreover, topic modeling was performed on English posts, which made up more than 81% of the posts. As a result, the discussions and sentiments of the majority of the posts have been extracted and analyzed. However, there are countries in which people do not speak English, French, or Spanish. The analysis of this study could not be generalized to the countries whose official languages are different.

The number of posts with Mpox and 2SLGBTQIAP+ keywords had a higher correlation with the number of Mpox cases (correlation coefficient of 0.36, P<.001) compared to the number of posts on Mpox (correlation coefficient of 0.24, P<.001). This indicates that social media discussions on Mpox are tightly related to the 2SLGBTQIAP+ community. Out of the 10 topics related to Mpox and LGBTQ, 8 were directly focused on blaming the gay community for spreading Mpox. The sentiments on all topics were very negative. Three of the topics that were strongly related to the stigmatization of the 2SLGBTQIAP+ community had a significantly lower sentiment score compared to other topics (ANOVA P<.001). The sentiment of posts from all the 3 languages, English, Spanish, and French, had a higher negative intensity, then neutral, and then positive (P<.001). Canada had the lowest sentiment score compared to other countries (P<.04). Stigmatization of a minority community on this scale will cause seclusion of people and increase hesitancy for seeking help upon realization of the symptoms. Stigmatization of the gay community, especially in countries where the sentiment polarity is very negative (ie, Canada, the United States, and the United Kingdom), must be prevented in order to contain Mpox and control the disease.

As a contribution to the future of this work, NLP tools could be used to study the sentiments and topics of posts regarding Mpox and the 2SLGBTQIAP+ community in languages and regions other than the ones studied in this manuscript.