Abstract
The present study explored the topics and sentiment associated with gambling addiction during the COVID-19 pandemic, using topic modeling and sentiment analysis on tweets in English posted between 17-24th April 2020. The study was exploratory in nature, with its main objective consisting of inductively identifying topics embedded in user-generated content. We found that a five-topic model was the best in representing the data corpus, including: (i) the public’s perception of gambling addiction amid the COVID-19 outbreak, (ii) risks and support available for those who stay at home, (iii) the users’ interpretation of gambling addiction, (iv) forms of gambling during the pandemic, and (v) gambling advertising and impact on families. Sentiment analysis showed a prevalence of underlying fear, trust, sadness, and anger, across the corpus. Users viewed the pandemic as a driver of problematic gambling behaviors, possibly exposing unprepared individuals and communities to forms of online gambling, with potential long-term consequences and a significant impact on health systems. Despite the limitations of the study, we hypothesize that enhancing the presence of mental health operators and practitioners treating problem gambling on social media might positively impact public mental health and help prevent health services from being overwhelmed, in times when healthcare resources are limited.
Introduction
General background
The recent, global affirmation of online social media has dramatically changed the way individuals represent their health and share relevant information.Citation1 de Rosa et al.Citation2 defined social media as contemporary “social arenas” (p. 16), online virtual communities where users gather to collectively generate and share social representations, in an attempt to make sense of the world, acquire information, and turn it into actionable knowledge. Over the past decade, research has increasingly acknowledged the potential of social media to inform public and global health policy, focusing on the analysis of online user-generated content.Citation3–5 Findings highlight the role of social media as key in health communication, allowing users to share information on prevention and treatment of diseases and engage in mutual support, ultimately to increase control over their physical and mental health.Citation1,Citation2
Social media and public health
Social media have played a pivotal role during the COVID-19 pandemic, especially in the early phases,Citation6 allowing a large number of distanced individuals from different countries to stay connected, interact, and keep updated on the progression of the pandemic.Citation7 The novel coronavirus-2019 (COVID-19) is the acute respiratory disease caused by the novel Sars-CoV-2 virus. Since early cases were observed in China in November 2019, the infection spread rapidly across several countries, triggering the World Health Organisation (WHO) to declare a pandemic on the 11th of March 2020.Citation8 In an attempt to contain the spread of the disease, several national and local governments introduced preventative measures, including, but not limited to, confinement, quarantine, travel and spatial distancing restrictions, commonly referred to as lockdown.Citation9
Studies show that lockdown measures represent a significant challenge to individual’s lives and their mental health, requiring the adoption of novel strategies to cope with symptoms of stress, anxiety, depression, insomnia,Citation10,Citation11 substance abuse,Citation12 and addictive behaviors, including gambling addiction.Citation13 Dubey et al.Citation14 recently defined COVID-19 and addiction in terms of two colliding pandemics, with the future impact on public health systems yet to be estimated. Similarly, Kar et al.Citation10 argued that spatial distancing measures may represent a threat for a number of individuals at risk, on the one hand pushing them to engage in excessive indoor behaviors, such as consuming more media and games, while on the other, increasing the likelihood of experiencing withdrawal symptoms, especially in the light of the current unavailability of in-person mental health services.
Some specific features of the pandemic and associated restrictions might have an impact on gambling behaviors. For example, recent studies have indicated that lockdown policies affect the mental health of individuals, interfering with and disrupting personal, social, financial, and occupational aspects of their lives.Citation15 Some authors have argued that such unprecedented changes may play a significant role in consolidating or generating new patterns of addictive behaviors to overcome stress, anxiety, and preoccupations.Citation13 As discussed by Mallet et al.Citation15 “[s]ocial isolation during spatial distancing, and these stressors in conjunction with substantial changes in gambling markets (land-based, online) during the pandemic may significantly influence gambling behaviors” (p. 5). Moreover, spatial distancing measures and telehealth may represent a challenge for individuals affected by addictive syndromes, particularly for those unable to access online treatments.Citation16
Gambling addiction
Gambling addiction is viewed as being at the extreme end on a continuous scale of disordered gambling behaviors, resulting in detrimental psychosocial, mental, and financial consequences.Citation17–20 The past decade has seen a proliferation in various forms of online gambling due to the diffusion of computers and mobile devices.Citation21 These provide users with accessible, immersive gambling interfaces, and for these reasons it has been hypothesized that online gambling may be comparatively more addictive than offline gambling.Citation22–25
However, empirical findings from a survey conducted by Wardle et al.Citation26 in a representative sample of British adults, showed that most online gamblers tended to gamble offline too, displaying both high involvement with gambling activities and high rates of problem gambling. This means that the online environment may represent a threat for those with problem gambling, as they may use the internet to pursue a variety of accessible gambling opportunities,Citation27 presumably even more in times when land-based gambling establishments are not available.
Gambling addiction during the COVID-19 pandemic
It has been suggested that the lockdown measures implemented to counter the spread of the COVID-19 pandemic may represent a possible driver of gambling behaviors, enhancing risks for a large number of individuals. For example, some authors have argued that significant risk factors for problematic gambling such as boredom and financial preoccupations may be exacerbated by the restrictions associated with the lockdown.Citation28 However, recent studies on online gambling trends during the pandemic present contrasting evidence. PriceCitation29 showed that high-risk gamblers among a larger group of online gamblers had a higher likelihood of gambling during the 2020 Canadian lockdown. A study by HåkanssonCitation30 carried out in May 2020, using self-reports from Swedish participants, showed that only a minority of the interviewees presented increased gambling activity, and those were already at a higher risk of problem gambling and of alcohol-related problems. However, as pointed out by Mallet et al.Citation15 Sweden is a country where lockdown measures were not implemented. A study by Auer et al.Citation31 from a study on a behavior tracking dataset of online sports gamblers (N = 5,396) from a European online gambling operator in Sweden, Germany, Finland, and Norway, showed (i) a significant reduction in the amount of money spent on sport betting, compared to the pre-pandemic period, (ii) no significant conversion of sport betting into more online casino playing, and (iii) a significant decrease in online casino playing for sport bettors. Another, more recent study by Auer and GriffithsCitation32 analyzed behavioral tracking data among a sample of over 133,000 online casino Swedish gamblers, with results indicating overall decreased gambling intensity. In summary, considering all the available evidence, it might be sensible to hypothesize that gambling trends and patterns in response to the pandemic may vary, and that variations may depend on the characteristics of the observed population as well as on the public health measures put in place in different cultural and national contexts.
Nevertheless, some authors have argued that guidelines to help counter the possible detrimental effects of enhanced internet use during the pandemic, including online gambling activity, are needed (Citation13, see also the review by Mallet et al.Citation15). In particular, Király et al.Citation13 highlighted that although several behaviors, under normal circumstances, could help individuals mitigate stress, anxiety, and preoccupations, without necessarily leading to addictions, the current spatial distancing measures and inability to perform a number of common daily activities may turn such behaviors into new, persistent practices and subsequently, into behavioral addictions, with potentially dramatic mental health implications.
In light of the contrasting evidence on the prevalence and patterns of gambling behaviors during the pandemic across different national contexts, it is of foremost importance to study the public’s perception of behavioral addictions. More specifically, the unprecedented circumstances are favoring the emergence of new work-life routines, balance, and patterns of behavior, where individuals are encouraged to stay at and work from home, whenever possible.Citation33 In this scenario, user-generated content could represent a key source of information, aiming to shed light on topics and sentiment underlying the discourse that is produced and shared globally on online “social arenas” such as Twitter.Citation2 This could provide a baseline to examine how the pandemic restrictions are affecting individuals and communities, in an attempt to tackle gambling addiction as a somewhat hidden mental health condition and contribute to inform public health policy.
The role of Twitter in public health communication
A growing corpus of literature has investigated the role of social media in conveying information on mental health, particularly Twitter.Citation34–36 Twitter is one of the most popular and utilized social media in a large number of countries,Citation4 providing a social microblogging platform where users can post and interact by means of short text messages called tweets, limited to 280 characters. As of the first quarter of 2019, Twitter had an average of 330 million monthly active users.Citation37
Recent studies by Houghton et al.Citation38 and Killick and GriffithsCitation39 have shown a significant presence of gambling operators on Twitter, using a variety of strategies to link tweets with major sporting events, ultimately leading to an arguably non-responsible promotion of gambling services. This poses questions regarding the possible implications for the public, particularly whether poorly controlled gambling advertising contributes to instigating a fundamental negative sentiment toward the gambling industry and the capacity of national and local institutions to regulate itCitation40. This may have an impact on public mental health, especially in the current pandemic, based on the evidence that lack of trust tends to be associated to lower adherence to public health measures,Citation41 possibly carrying a vicious cycle of “distrust, non-compliance, hardships and further distrust” (p.89).
Research aims
The present study explored the public’s perception of gambling addiction on a corpus of tweets in English posted in April 2020. We used topic modeling and sentiment analysis to analyze the data. Topic modeling allows for the exploration of semantic patterns and latent meaning emerging from text corpora, rather than testing a set of predetermined hypotheses on the data.Citation42 This makes it “a tool for theory development through exploratory research, unlike traditional quantitative techniques which are mostly used for testing of theories” (p. 794). Therefore, research using topic modeling is “exploratory in nature,” with its main aim consisting of identifying topics embedded in a corpus (p. 791). We used topic modeling and sentiment analysis to analyze Twitter user-generated content and to explore the main topics and sentiment associated with gambling addiction within a specific time frame, rather than testing hypotheses on the prevalence or the characteristics of gambling trends during the pandemic. The specific, relatively short time-frame was chosen to account for approximately one month following the WHO’sCitation43 pandemic declaration and the beginning of spatial distancing restrictions in several countries, including the suspension of traditional land-based gambling establishments.
Materials and methods
Data mining consisted of three main steps: (i) data search and collection, (ii) data pre-processing, and (iii) data analysis. The following sections include a brief explanation of each step and the inclusion and exclusion criteria applied at each level of parsing through the corpus.
Step 1: Data search and collection
We used Twitter’s Representational State Transfer (REST) Application Programming Interface (API) and the rtweet R packageCitation44 to search and collect tweets. Twitter’s REST API is a freely available user interface that is used in data mining and machine learning applications as an endpoint to search and collect Twitter data, providing information and content from individual accounts and allowing researchers to filter content by means of selected keywords.Citation45 rtweet is a package based on R statistical programming language that provides several functions to extract and filter data from Twitter’s REST and streaming APIs. This is a previously peer-reviewed instrumentCitation46 which has been already used in several studies where data mining and machine learning applications were used to inform public health policy.Citation4,Citation46–49
Specifically, we searched for tweets posted between 17 and 24th April 2020, in English, using the following search terms: gambl* AND addict*, excluding retweets. We chose this specific period as the majority of countries had implemented lockdown measures in a time-period between the first and the second half of March 2020.Citation50 This allowed for approximately one month for observing meaningful, prevalent topics and sentiment.
Following, we ran a preliminary search and investigated results. We observed a significant number of tweets that were not relevant to our analysis, due to the fact that terms such as “gambling” and “addiction” are often used metaphorically in everyday language, representing a variety of meanings.Citation51 More specifically, these tweets mentioned “gambling” and “addiction” as figures of speech to comment on the conduct of public individuals such as politicians and sports player, with little or no relation to the context of mental health. For these reasons we reviewed the search options by carefully excluding a selection of terms that in all cases had led to non-meaningful tweets (i.e., -jordan -michael -mike -mikes -mjs -documentary -lastdance -mikebow -mj -trump -james -ryzin). The final search returned 371 tweets. We performed a further, manual inspection of results, and this time we observed a corpus that was more internally consistent, showing a high prevalence of tweets relevant to gambling addiction, as required. However, in one tweet the word gambling was used figuratively to describe drug abuse, and therefore we decided to remove it. After removing duplicates, we retained a final corpus of 144 unique tweets, posted by total 143 different users ().
All data were collected according to internet and Twitter research ethical guidelines. An institutional research committee approved the use of the data for the purposes of the study.
Step 2: Data pre-processing
Consistent with recommendations from the literature,Citation52,Citation53 the text corpus was prepared for the analyses by performing a sequence of pre-processing operations. In particular, at this stage, the following refinement criteria were applied to the collected tweets, in the following order: Removal of hashtags, due to their relatively small number across the whole corpus; removal of unicode strings (e.g., strings like “\u002c” and “\x06”), accidentally retained in the extraction procedure; replacement of URLs and user mentions; replacement of abbreviations (e.g., “ty” with “thank you”); replacement of contractions (e.g., “won’t” and “don’t” with “will not” and “do not,” respectively); removal of numbers; removal of special characters (e.g., ampersands); removal of punctuation; lowercasing all words; spelling correction, where required. Subsequently, we proceeded with the semi-automated removal of English stopwords and manual lemmatization of the corpus. Stopwords are common words in English, but with little or no relevance to the subject of interest (e.g., words like “and,” “or,” “however”). We used a list combining stopwords from the tidyverse R packageCitation54 and additional custom words that we had previously identified through manual inspection. At this stage, exclusion criteria encompassed filtering for redundant, non-substantive semantic content relative to the domain of analysis (for example, but not limited to, words such as: “basically,” “woulda,” “possibly,” “couldve,” “gonna”). Lemmatization consists of the analysis of words’ morphology and the replacement of inflectional endings with their base forms.Citation53 Finally, in order to predispose the corpus for the topic analysis, we restructured it into a tokenized data frame using the tidytext R package.Citation55
Step 3: Data analysis: Topic modeling and model selection
We used topic modeling and sentiment analysis to analyze the data. Topic modeling is a “method for unsupervised classification of such documents, similar to clustering on numeric data, which finds natural groups of items even when we’re not sure what we’re looking for.”Citation56(Chapter 6) Several methods and algorithms are available, each suiting specific types of corpora. We used Biterm Topic Modeling (BTM).Citation57
Biterm is a topic modeling algorithm that classifies words and documents from a text corpus into a smaller number of latent topics, being particularly suitable for short texts such as tweets.Citation58 Specially, the algorithm generates a set of topics from biterms, namely patterns of unordered word co-occurrences within the text. In doing so, BTM differs from other methods based on word-document co-occurrences (e.g., Latent Dirichlet AllocationCitation59). In fact, short texts are characterized by a significant degree of sparsity, due to their limited amount of information. BTM generates topics from co-occurring, unordered word-pairs (biterms), ultimately tackling the issue of sparsity. This allows for the generation of more coherent topics, and consistently, research shows that BTM outperforms other established methodologies in short texts such as tweets.Citation57
Aiming to identify the best-fitting model, we ran 99 different BTM analyses by progressively varying the number of topics (k) anticipated for the corpus (from 2 to 100), and eventually compared the results across models. We used the collapsed Gibbs sampling Markov chain Monte Carlo algorithmCitation57,Citation60 with five iterations per model via the BTM R package.Citation61 The algorithm assigns biterms to topics based on the conditional distributions of word co-occurrences, using the first extracted topic as a background topic, corresponding to the empirical word distribution. In addition to k, the algorithm requires the specification of two hyperparameters, namely α (i.e., the symmetric Dirichlet prior probability of a topic), representing the granularity of topic distribution over the corpus, and β (i.e., the symmetric Dirichlet prior probability of a word given the topic), representing the specificity of topics.Citation52 The outcome of the process consists of a model identified by a topic-word probability distribution (ϕ) and an overall topic probability distribution (θ), the latter quantifying the proportion of biterms per topic.Citation57 In summary, a given topic probability is determined by the proportion of biterms accounted by that topic over the whole corpus (e.g., a topic probability of 0.60 indicates that a given topic explains more than half of all word co-occurrences in the corpus).
In order to select and retain the best model, we used two complementary approaches: (i) We estimated a measure of topic coherence and exclusivity, and (ii) we manually inspected the model to make sure this was interpretable. In fact, evidence from previous studies suggests that automated detection alone does not allow for optimal decision-making over cutoff values, warranting expert manual inspection and evaluation in any cases.Citation52 We first estimated topic exclusivity by adapting the relevant function from the stm R packageCitation62 to BTM. This is a measure of relative word exclusivity and coherence across topics, originally developed for LDA.Citation62,Citation63 We used the mean to aggregate topic scores per model and normalized results to facilitate their interpretation. Second, we manually inspected the model by examining the topics extracted and their associated words.
Regarding the two hyper-parameters of BTM, commonly used values are α = 50/k and β = 0.1.Citation64 However, smaller values are recommended when dealing with distributions that are sparse over topics,Citation52,Citation65 and for this reason we fixed values to α = 5/k and β = 0.01. Finally, we performed a sentiment analysis of the corpus by means of the syuzhet R package.Citation66 Sentiment analysis clusters the corpus into smaller, comprehensive categories, based on pre-coded dictionaries, with each category representing a distinct, prevalent emotion or sentiment.Citation4 In particular, we used sentiment analysis to detect the prevalent emotional tone associated to the discourse on gambling addiction on Twitter, aiming to improve the interpretation of results from the analysis based on the classification of information in topics.Citation67 We used the NRC Word-Emotion Association Lexicon dictionary,Citation68 including a list of English words and their associations with eight basic clusters of emotions (anger, fear, anticipation, trust, surprise, sadness, joy, and disgust, respectively). We used the probabilistic inference obtained through topic modeling to describe the semantic patterns underlying the text and to interpret the topics, and we supported the interpretation by performing a qualitative analysis of text when required.
Results
shows a summary of the characteristics of the corpus of tweets. displays the most frequent words observed in the corpus, in decreasing order.
The results of topic modeling and the distribution of normalized exclusivity across models are presented in . We observed a major rise in topic exclusivity between models accounting for 10 to 20 topics, respectively, peaking at approximately 40 topics. However, the manual inspection challenged the interpretability of those models, showing high sparsity and inconsistencies within and between topics. Instead, we found that a model accounting for five topics (mean exclusivity across topics = 25.22; normalized mean exclusivity = 0.46) was the best to represent the corpus, maximizing word exclusivity, topic internal consistency, and the overall model’s interpretability. Therefore, we decided to retain the five-topic model for further analyses and interpretation.
presents the network model of relations across words, resulting from the analysis of top 100 bigrams across the corpus of tweets, namely the most common sequences of consecutive words observed within tweets.Citation56
In the following paragraphs, we present each topic individually, focusing on their top 25 words (i.e., words with highest co-occurrence probability) (). Aiming to present a meaningful interpretation of topics, we used portions of text from tweets as elementary context unitsCitation69 (i.e., significant parts of text as defined by words co-occurrences).
Topic modeling
Topic 1: Gambling addiction amid the COVID-19 outbreak
Topic 1 showed the highest mean topic probability (0.40). It incorporated words relevant to the impact of gambling addiction on individual’s “lives” amid the lockdown, particularly in relation to specific forms of gambling such as “bet” and “casino.” For example, users engaged in discussions on betting companies, perceived as responsible for reinforcing problem behaviors in individuals with gambling addiction during the lockdown. In this early phase of the pandemic, users reported witnessing increased “marketing” efforts by betting companies through a variety of different media such as TV, radio, and the internet. Gambling addiction was defined in relation to words such as “money,” “drug,” and “stop,” ultimately comparing the condition to other addictive syndromes like drug abuse or alcoholism.
Topic 2: Risks and support available for those who stay at home
The mean topic probability for Topic 2 was 0.16. This topic included the word “advert,” referring to the detrimental impact of online gambling advertising on the mental health of those who stay at home during the lockdown. Users reported a pervasive presence of gambling adverts and concerns of the possible effects on individuals with gambling addictions. The addictive potential of gambling was compared to that of highly addictive substances such as “heroin,” in an attempt to debunk the view of gambling as a “comfort hobby.” Staying at home was described as a challenge for those who struggled with addictive behaviors, and dedicated support was advertised and promoted, for example, helplines offering advice, information or a listening ear when needed. Ultimately, users outlined a framework of risks and opportunities during the lockdown, in which “home” is perceived as not the “safest place to be.”
Topic 3: The user’s psychology of gambling addiction
The mean topic probability for Topic 3 was 0.14. One of the key themes underlying this topic was the “psychological” progression of gambling at the individual level, to the point of becoming a “devastating addiction.” This suggested that individuals with gambling addiction will not know they have a problem until “it’s too late,” since from a psychological point of view, the condition just seems to “take you over.” The public understanding of gambling addiction in psychological terms centers around the concept of “impulsivity.” Problem gambling was considered as a form of “impulsive behavior” leading to a “psychological addiction.” Interestingly, another word co-occurring with “gambling” and “addiction” in this topic was “hustler,” used in an attempt to correct a common misrepresentation of individuals with gambling addiction, the latter needing help to overcome their condition rather than being someone who engages in illegal activities to increase their wealth.
Topic 4: Forms of gambling at the time of pandemic lockdown
The mean topic probability for Topic 4 was 0.13. Words like “horse” and “race” were present in this topic, functional to questioning which forms of gambling are most common during the pandemic outbreak. Lockdown-related restrictions have affected traditional forms of gambling, triggering a reflection on what consequences they may have on individuals with gambling addiction. Users referred to the recent suspension of professional “sport” leagues and “casinos” in several countries. The most commonly represented forms of gambling during the lockdown were online games, casinos, and esports, replacing traditional sports betting and land-based establishments.
Topic 5: Gambling advertising and impact on families
The mean topic probability for Topic 5 was 0.18. Since the lockdown, users reported seeing a number of gambling and betting webpages posting adverts “on the few sporting fixtures left as well as esports.” Gambling “companies” and operators were criticized for their supposed unethical conduct, particularly for trying to influence individuals who might be at risk of developing gambling problems (“vulnerable”). Some users explicitly attacked sport institutions and commercial operators, viewing them as encouraging adults and children into “gambling” and “addiction” through a number of online initiatives. Addiction was associated with “family,” and gambling to escape boredom was seen as a tangible risk for those who stayed at home, potentially carrying dramatic psychosocial and financial consequences for individuals and their families (e.g., as exemplified in the following excerpt from a tweet: “[…] they’re making the most of people’s addictions and boredom during the lockdown. “Broke lads” addicted to gambling are keeping the parasites going, often at the cost to their own families who go without when these often losers go into to feed their addictions.”)
Sentiment analysis
The results of the sentiment analysis showed that fear was the most common sentiment in the corpus (18.43%), followed by trust (17.20%), sadness (15.48%), and anger (12.29%), highlighting an overall prevalence of negative sentiment ().
The less common sentiments observed were surprise (5.41%), disgust (9.58%), and joy (9.83%).
Discussion
The main aim of this study was to explore the topics and sentiment associated with gambling addiction during the COVID-19 pandemic lockdown, based on Twitter online user-generated content. We collected a corpus of tweets posted for one week, in April 2020 and we analyzed it by means of topic modeling and sentiment analysis. Results showed that a model accounting for five topics was a good fit to the corpus of tweets. Topics identified were (i) gambling addiction amid the COVID-19 outbreak; (ii) risks and support available for individuals who stay at home; (iii) the user’s psychology of gambling addiction; (iv) forms of gambling at the time of the pandemic lockdown; and (v) gambling advertising and impact on families. The results of the sentiment analysis showed that fear was the most common sentiment in the corpus, followed by trust, sadness, and anger, highlighting an overall prevalence of negative sentiment. The less common sentiments observed were surprise, disgust, and joy, respectively.
One of the primary observations emerging from the study was that, at the time when lockdown measures were implemented in several countries, Twitter users viewed gambling mainly as an online activity emerging under spatial distancing measures, with an underlying fear, and sadness toward a number of possible consequences for public mental health. In this regard, although we did not collect data on the presence of online gambling operators, our qualitative analysis of tweets allowed us to isolate individual users’ concerns about their observation of an increase in online presence of gambling adverts (e.g., “[…] mate I’m concerned about the amount of online gambling adverts on at the moment. I have mates that have been or are gambling addicts, it seems like the adverts are on constantly at the moment.”), in a time when most traditional land-based gambling establishments had to close in response to the spread of the COVID-19 infection. The results from topic modeling presented in the study indicate that the pandemic lockdown in March 2020 was perceived as a possible driver of problematic gambling behaviors, exposing a large number of unprepared individuals and communities to new forms of gambling, with potential long-term consequences and a significant impact on public health systems.
Another interesting finding from the present study was the perception of the role of boredom, highlighted by Twitter users as main triggers for gambling during the pandemic lockdown, along with financial preoccupations associated with the upcoming economic crisis. In this regard, the increased amount of time that individuals have due to staying at home, is considered as a key factor. Previous studies have shown that a lack of social interaction among problem gamblers is, in fact, associated on the one hand with a higher likelihood to engage in addictive behaviors, and on the other, to an increased need to reconnect.Citation21,Citation25,Citation70 However, Auer et al.Citation31 suggested that the reduced trends in online gambling that they had observed in the first half of 2020 may be associated with (i) individuals’ financial conservatism in view of a possible upcoming economic crisis, (ii) a general reluctance to gamble in front of members of their household, and (iii) the fact that they may be engaging in more “quality time” with their families (p.7). Therefore, although our results indicate a possible misalignment between the public’s perception of gambling addiction on Twitter and other empirical studies on gambling during the pandemic, we suggest caution in their interpretation, inviting future research to monitor the discourse on gambling addiction in the next stages of the pandemic, possibly including data from different media sources and multiple national contexts.
For all these reasons, despite the notable limitations of the present study, we hypothesize that enhancing the presence of mental health operators and practitioners treating problem gambling and gambling addiction on social media could positively impact patients and their families in such unprecedent times. Moreover, evidence from recent studies shows that online intervention for problem gamblers improves gambling outcomes.Citation71,Citation72 Having online support could help address the demand for dedicated services, helping individuals fight such dramatic conditions, and this could also prevent public health services from being misused or overwhelmed, in times when human and material resources in health care are of foremost necessity.Citation73
Another interesting finding emerging from the present study was the public’s general lack of trust in the gambling industry, and more generally, toward governments and institutions. In fact, trust was one of the main sentiments identified, along with fear, sadness, and anger, suggesting a widespread concern toward the conduct of gambling operators and their ethics in advertising, hitting a high number of exposed and unprepared individuals, globally, and potentially leading to decreased adherence to necessary public health measures.Citation41 We believe that such findings could be of specific interest to policymakers, and they could be used to better address efforts and collaboration with commercial operators in the gambling sector, aiming to define clear and sustainable social responsibility guidelines, necessary to regulate online advertising and increase trust among the general public. Unsurprisingly, around the same time our data were collected, the Gambling CommissionCitation74 provided additional formal guidance for remote gambling operators in the UK during the COVID-19 pandemic, in an attempt to regulate the way gambling operators interact with potential customers, and aiming to minimize the risk of harms associated with gambling. We believe this represents an important step forward, whose outcomes will deserve further evaluation in the near future, in comparison with similar initiatives in place in other national contexts.
This study has limitations. First, all tweets were collected in English, affecting the generalizability of results to other linguistic contexts. Second, the data were drawn from many countries and it is possible that some countries were in different stages of lockdown at the time the data were collected. Third, the tweets did not necessarily reflect the lived experience of gamblers but may have reflected speculation produced by several diverse users and agencies involved in online communication. Moreover, a search strategy using variations of the words “gambling” and “addiction” may have led to biased results, considering that individuals with addictive gambling behaviors may not necessarily be willing to discuss it publicly. Fourth, there was not a comparison of the data corpus with other datasets collected from a non-COVID period, which would help determine the uniqueness of the semantic associations discussed in the present study. Fifth, data were collected within a limited time, namely one week. This was in response to the need to capture topics and sentiments immediately following the outbreak of the pandemic and in the early phases of the lockdown. However, this may represent a limitation because a number of recent developments of the scenario could have determined fluctuations and changes in topics and sentiments. Finally, although biterm topic modeling was specifically designed to overcome the problem of data sparsity that affects other topic modeling techniques applied to short text,Citation57 the number of tweets collected was limited, affecting the generalizability of results.
Conclusions
The results presented here suggest a possible shift in the users’ perception of the forms of gambling during the lockdown, with increased concerns expressed by individual users about the presence on Twitter of online gambling operators, and the potential effect of online gambling advertising on individuals living under lockdown restrictions. The latter is arguably viewed as unethically promoting their services and potentially influencing individuals, ultimately generating fear, sadness, and lack of trust among the general public.
We consider three main implications and possible applications of these results: (i) they could prompt a reflection among policymakers and practitioners, with regards to best practices and strategies to optimize online communication concerning gambling addiction during the pandemic; (ii) as a consequence, they could inform institutional efforts to promote trust and adherence to prevention practices in the community, ideally in conjunction with evidence from cross-sectional and longitudinal research; and (iii) they could ultimately provide local and national governments with an updated view on the public’s perception of gambling addiction, as evident from the analysis of online discourse, possibly suggesting changes to the mental health agenda in the attempt to address the public’s demands.
Despite the limited generalizability of the findings, we believe that the results from the present study provide preliminary evidence on the development of topics and sentiments on gambling addiction during the pandemic lockdown. We recommend that future research should investigate specific risk factors of gambling addiction associated to the altered living conditions of individuals, globally, aiming to provide updated public health guidelines and prevent mental health crises during and after the COVID-19 pandemic.
References
- Zhang L, Hall M, Bastola D. Utilizing Twitter data for analysis of chemotherapy. Int J Med Inform. 2018;120:92–100. doi:https://doi.org/10.1016/j.ijmedinf.2018.10.002.
- de Rosa AS, Fino E, Bocci E. Addressing healthcare on-line demand and supply relating to mental illness: knowledge sharing about psychiatry and psychoanalysis through SN in Italy and France. In: Kapoor A, Kulshrestha C, editors. Dynamics of competitive advantage and consumer perception in social marketing. Hershey, PA: IGI Global; 2013. p. 16–55.
- De Martino I, D'Apolito R, McLawhorn AS, Fehring KA, Sculco PK, Gasparini G. Social media for patients: benefits and drawbacks. Curr Rev Musculoskelet Med. 2017;10(1):141–5. doi:https://doi.org/10.1007/s12178-017-9394-7.
- Sansone A, Cignarelli A, Ciocca G, Pozza C, Giorgino F, Romanelli F, Jannini EA. The sentiment analysis of tweets as a new tool to measure public perception of male erectile and ejaculatory dysfunctions. Sex Med. 2019;7(4):464–71. doi:https://doi.org/10.1016/j.esxm.2019.07.001.
- Stellefson M, Chaney B, Ochipa K, Chaney D, Haider Z, Hanik B, Chavarria E, Bernhardt JM. YouTube as a source of chronic obstructive pulmonary disease patient education: a social media content analysis. Chron Respir Dis. 2014;11(2):61–71. doi:https://doi.org/10.1177/1479972314525058.
- Hinjoy S, Tsukayama R, Chuxnum T, Masunglong W, Sidet C, Kleeblumjeak P, Onsai N, Iamsirithaworn S. Self-assessment of the Thai Department of Disease Control's communication for international response to COVID-19 in the early phase. Int J Infect Dis. 2020;96:205–10. doi:https://doi.org/10.1016/j.ijid.2020.04.042.
- Limaye RJ, Sauer M, Ali J, Bernstein J, Wahl B, Barnhill A, Labrique A. Building trust while influencing online COVID-19 content in the social media world. Lancet Digit Health. 2020;2(6):e277–8. doi:https://doi.org/10.1016/S2589-7500(20)30084-4.
- Hu Y, Sun J, Dai Z, Deng H, Li X, Huang Q, Wu Y, Sun L, Xu Y. Prevalence and severity of corona virus disease 2019 (COVID-19): a systematic review and meta-analysis. J Clin Virol. 2020;127:104371. doi:https://doi.org/10.1016/j.jcv.2020.104371.
- Ibarra-Vega D. Lockdown, one, two, none, or smart. Modeling containing Covid-19 infection. A conceptual model. Sci Total Environ. 2020;730:138917. doi:https://doi.org/10.1016/j.scitotenv.2020.138917.
- Kar SK, Arafat SMY, Sharma P, Dixit A, Marthoenis M, Kabir R. COVID-19 pandemic and addiction: current problems and future concerns. Asian J Psychiatr. 2020;51:102064. doi:https://doi.org/10.1016/j.ajp.2020.102064.
- Rossi R, Socci V, Talevi D, Mensi S, Niolu C, Pacitti F, Marco AD, Rossi A, Siracusano A, Lorenzo GD. COVID-19 pandemic and lockdown measures impact on mental health among the general population in Italy. An N = 18147 web-based survey. MedRxiv. 2020; 2020.04.09.20057802. doi:https://doi.org/10.1101/2020.04.09.20057802.
- Mackolil J, Mackolil J. Addressing psychosocial problems associated with the COVID-19 lockdown. Asian J Psychiatr. 2020;51:102156. doi:https://doi.org/10.1016/j.ajp.2020.102156.
- Király O, Potenza MN, Stein DJ, King DL, Hodgins DC, Saunders JB, Griffiths MD, Gjoneska B, Billieux J, Brand M, et al. Preventing problematic internet use during the COVID-19 pandemic: consensus guidance. Compr Psychiatry. 2020;100:152180. doi:https://doi.org/10.1016/j.comppsych.2020.152180.
- Dubey MJ, Ghosh R, Chatterjee S, Biswas P, Chatterjee S, Souvik D. COVID-19 and addiction. Diabetes Metab Syndr. 2020;14(5):817–23. doi:https://doi.org/10.1016/j.dsx.2020.06.008.
- Mallet J, Dubertret C, Le Strat Y. Addictions in the COVID-19 era: current evidence, future perspectives a comprehensive review. Prog Neuropsychopharmacol Biol Psychiatry. 2021;106:110070. doi:https://doi.org/10.1016/j.pnpbp.2020.110070.
- Satre DD, Hirschtritt ME, Silverberg MJ, Sterling SA. Addressing problems with alcohol and other substances among older adults during the COVID-19 pandemic. Am J Geriatr Psychiatry. 2020;28(7):780–3. doi:https://doi.org/10.1016/j.jagp.2020.04.012.
- American Psychiatric Association. Diagnostic and statistical manual of mental disorders, 5th Edition: DSM-5. 5th ed. Arlington, VA: American Psychiatric Publishing; 2013.
- Griffiths MD. Problem gambling and gambling addiction are not the same. J Addict Depend. 2016;2(1):1–3. doi:https://doi.org/10.15436/2471-061X.16.014.
- Griffiths MD. A “components” model of addiction within a biopsychosocial framework. J Subst Use. 2005;10(4):191–7. doi:https://doi.org/10.1080/14659890500114359.
- Iliceto P, Fino E, Schiavella M, Candilera G. Individual differences in interpersonal security predict suicidal ideation and problem gambling. Pers Individ Diff. 2020;162:110031. doi:https://doi.org/10.1016/j.paid.2020.110031.
- Gainsbury SM, Hing N, Suhonen N. Professional help-seeking for gambling problems: awareness, barriers and motivators for treatment. J Gambl Stud. 2014;30(2):503–19. doi:https://doi.org/10.1007/s10899-013-9373-x.
- Abbott MW. The changing epidemiology of gambling disorder and gambling-related harm: public health implications. Public Health. 2020;184:41–5. doi:https://doi.org/10.1016/j.puhe.2020.04.003.
- Binde P, Romild U, Volberg RA. Forms of gambling, gambling involvement and problem gambling: evidence from a Swedish population survey. Int Gambl Stud. 2017;17(3):490–507. doi:https://doi.org/10.1080/14459795.2017.1360928.
- Gainsbury SM. Online gambling addiction: the relationship between internet gambling and disordered gambling. Curr Addict Rep. 2015;2(2):185–93. doi:https://doi.org/10.1007/s40429-015-0057-8.
- Griffiths M. Internet gambling: preliminary results of the first U.K. prevalence study. JGI. 2001;5(5):5. doi:https://doi.org/10.4309/jgi.2001.5.8.
- Price A. Online gambling in the midst of COVID-19: a nexus of mental health concerns, substance use and financial stress. Int J Mental Health Addict. 2020;1–18. doi:https://doi.org/10.1007/s11469-020-00366-1.
- Wardle H, Moody A, Griffiths MD, Orford J, Volberg R. Defining the online gambler and patterns of behavior integration: evidence from the British Gambling Prevalence Survey 2010. Int Gambl Stud. 2011;11(3):339–56. doi:https://doi.org/10.1080/14459795.2011.628684.
- Sharman S, Roberts A, Bowden-Jones H, Strang J. Gambling in COVID-19 Lockdown in the UK: Depression, Stress, and Anxiety. Front Psychiatry [Internet]. 2021 [cited 2021 Feb 10];12. Available from: https://doi.org/https://www.frontiersin.org/articles/10/3389/fpsyt.2021.621497/full.
- Gainsbury SM, Delfabbro P, King DL, Hing N. An exploratory study of gambling operators' use of social media and the latent messages conveyed. J Gambl Stud. 2016;32(1):125–41. doi:https://doi.org/10.1007/s10899-015-9525-2.
- Håkansson A. Changes in gambling behavior during the COVID-19 pandemic—a web survey study in Sweden. IJERPH. 2020;17(11):4013. doi:https://doi.org/10.3390/ijerph17114013.
- Auer M, Malischnig D, Griffiths MD. Gambling before and during the COVID-19 pandemic among European regular sports bettors: An empirical study using behavioral tracking data. Int J Mental Health Addict. 2020;1–8. Advance online publication. doi:https://doi.org/10.1007/s11469-020-00327-8.
- Auer M, Griffiths MD. Gambling before and during the COVID-19 pandemic among online casino gamblers: an empirical study using behavioral tracking data. Int J Mental Health Addict. 2021;1–11. Advance online publication. doi:https://doi.org/10.1007/s11469-020-00462-2.
- Vasudevan M, Mehrolia S, Alagarsamy S, Balachandran AK. Work from home in the pandemic era: loss of mental equilibrium? Asian J Psychiatr. 2021;55:102490. doi:https://doi.org/10.1016/j.ajp.2020.102490.
- Charles-Smith LE, Reynolds TL, Cameron MA, Conway M, Lau EHY, Olsen JM, Pavlin JA, Shigematsu M, Streichert LC, Suda KJ, et al. Using social media for actionable disease surveillance and outbreak management: a systematic literature review. PLoS One. 2015;10(10):e0139701. doi:https://doi.org/10.1371/journal.pone.0139701.
- Himelboim I, Han JY. Cancer talk on Twitter: community structure and information sources in breast and prostate cancer social networks. J Health Commun. 2014;19(2):210–25. doi:https://doi.org/10.1080/10810730.2013.811321.
- Sugawara Y, Narimatsu H, Hozawa A, Shao L, Otani K, Fukao A. Cancer patients on Twitter: a novel patient community on social media. BMC Res Notes. 2012;5(1):699. doi:https://doi.org/10.1186/1756-0500-5-699.
- Clement J. Twitter: number of monthly active users 2010-2019. Statista; 2019. https://www.statista.com/statistics/282087/number-of-monthly-active-twitter-users.
- Houghton S, McNeil A, Hogg M, Moss M. Comparing the Twitter posting of British gambling operators and gambling affiliates: a summative content analysis. Int Gambl Stud. 2020;36:1–23. doi:https://doi.org/10.1080/14459795.2018.1561923.
- Killick EA, Griffiths MD. A content analysis of gambling operators’ Twitter accounts at the start of the English Premier League football season. J Gambl Stud. 2020;36(1):319–41. doi:https://doi.org/10.1007/s10899-019-09879-4.
- Wood & Griffiths (2008) to the reference list: Wood RTA, Griffiths MD. Why Swedish people play online poker and factors that can increase or decrease trust in poker Web sites: A qualitative investigation. Journal of Gambling Issues. 2008 Jun 1;0(21):80–97. doi:https://doi.org/http://dx.doi.org/10/4309/jgi.2008.21.8
- Blair RA, Morse BS, Tsai LL. Public health and public trust: survey evidence from the Ebola virus disease epidemic in Liberia. Soc Sci Med. 2017;172:89–97. doi:https://doi.org/10.1016/j.socscimed.2016.11.016.
- Vallurupalli V, Bose I. Exploring thematic composition of online reviews: a topic modeling approach. Electron Markets. 2020;30(4):791–804. doi:https://doi.org/10.1007/s12525-020-00397-5.
- World Health Organization. Novel coronavirus. 2020 [accessed 2020 Dec 14] https://www.who.int/emergencies/diseases/novel-coronavirus-2019.
- Kearney M. rtweet: collecting and analyzing Twitter data. JOSS. 2019;4(42):1829. doi:https://doi.org/10.21105/joss.01829.
- Twitter. Twitter developer—API reference index. 2021 [accessed 2021 Feb 15] https://developer.twitter.com/en/docs/api-reference-index.
- Aguilar-Gallegos N, Romero-García LE, Martínez-González EG, García-Sánchez EI, Aguilar-Ávila J. Dataset on dynamics of coronavirus on Twitter. Data Brief. 2020;30:105684. doi:https://doi.org/10.1016/j.dib.2020.105684.
- Mariano ER, Kou A, Stiegler MA, Matava C. The rise and fall of the COVID-19 aerosol box through the lens of Twitter. J Clin Anesth. 2020;69:110145. doi:https://doi.org/10.1016/j.jclinane.2020.110145.
- Ogundepo E, Folorunso S, Adekanmbi O, Akinsande O, Banjo O, Ogbuju E, Oladipo F, Abimbola O, Oseghale E, Babajide O. An exploratory assessment of a multidimensional healthcare and economic data on COVID-19 in Nigeria. Data Brief. 2020;33:106424. doi:https://doi.org/10.1016/j.dib.2020.106424.
- Trovato CM, Montuori M, Oliva S, Cucchiara S, Cignarelli A, Sansone A. Assessment of public perceptions and concerns of celiac disease: a Twitter-based sentiment analysis study. Dig Liver Dis. 2020;52(4):464–6. doi:https://doi.org/10.1016/j.dld.2020.02.004.
- Dunford D, Dale B, Stylianou N, Lowther E, Ahmed M, de la Torre Arenas I. Coronavirus: the world in lockdown in maps and charts. BBC News; 2020 [accessed 2020 Dec 14]. https://www.bbc.co.uk/news/world-52103747.
- Lopez-Gonzalez H, Guerrero-Solé F, Estévez A, Griffiths MD. Betting is loving and bettors are predators: a conceptual metaphor approach to online sports betting advertising. J Gambl Stud. 2018;34(3):709–26. doi:https://doi.org/10.1007/s10899-017-9727-x.
- Bickel MW. Reflecting trends in the academic landscape of sustainable energy using probabilistic topic modeling. Energ Sustain Soc. 2019;9(1):49. doi:https://doi.org/10.1186/s13705-019-0226-z.
- Symeonidis S, Effrosynidis D, Arampatzis A. A comparative evaluation of pre-processing techniques and their interactions for twitter sentiment analysis. Expert Syst Appl. 2018;110:298–310. doi:https://doi.org/10.1016/j.eswa.2018.06.022.
- Wickham H, Averick M, Bryan J, Chang W, McGowan L, François R, Grolemund G, Hayes A, Henry L, Hester J, et al. Welcome to the Tidyverse. JOSS. 2019;4(43):1686. doi:https://doi.org/10.21105/joss.01686.
- Silge J, Robinson D, Hester J. Tidytext: text mining using Dplyr, Ggplot2, and other Tidy tools; 2016 [accessed 2020 Dec 14]. doi:https://doi.org/10.5281/zenodo.56714.
- Silge J, Robinson D. Text mining with R: a tidy approach. 1st ed. Sebastopol, CA: O’Reilly Media; 2017.
- Yan X, Guo J, Lan Y, Cheng X. A biterm topic model for short texts. Proceedings of the 22nd International Conference on World Wide Web (WWW ‘13); 2013. p. 1445–56. doi:https://doi.org/10.1145/2488388.2488514.
- Blei DM. Probabilistic topic models. Commun ACM. 2012;55(4):77–84. doi:https://doi.org/10.1145/2133806.2133826.
- Blei DM, Ng AY, Jordan MI. Latent Dirichlet allocation. J Mach Learn Res. 2003;3:993–1022.
- Qiu Z, Wu B, Wang B, Shi C, Yu L. Collapsed Gibbs sampling for latent Dirichlet allocation on spark. Proceedings of the 3rd International Conference on Big Data, Streams and Heterogeneous Source Mining: Algorithms, Systems, Programming Models and Applications, Vol. 36, 2014. p. 17–28.
- Wijffels J. BTM: Biterm Topic Models for Short Text. R package version 0.3; 2020 [accessed 2020 Dec 14] https://CRAN.R-project.org/package=BTM.
- Roberts ME, Stewart BM, Tingley D. stm: an R package for structural topic models. J Stat Soft. 2019;91(2):1–40. doi:https://doi.org/10.18637/jss.v091.i02.
- Bischof J, Airoldi EM. Capturing topical content with frequency and exclusivity. Proceedings of the 29th International Conference on Machine Learning; 2012. p. 9–16. https://icml.cc/2012/papers/113.pdf
- Griffiths TL, Steyvers M. Finding scientific topics. Proce Natl Acad Sci. 2004;101(Supplement 1):5228–35. doi:https://doi.org/10.1073/pnas.0307752101.
- Sun L, Yin Y. Discovering themes and trends in transportation research using topic modeling. Transp Res Part C Emerg Technol. 2017;77:49–66. doi:https://doi.org/10.1016/j.trc.2017.01.013.
- Jockers ML. Syuzhet: extract sentiment and plot arcs from text; 2015 https://github.com/mjockers/syuzhet. Accessed 10 May 2020.
- Yada A, Jha CK, Sharan A, Vaish V. Sentiment analysis of financial news using unsupervised approach. Procedia Comput Sci. 2020;167:589–98. doi:https://doi.org/10.1016/j.procs.2020.03.325.
- Mohammad S, Turney P. Crowdsourcing a word-emotion association lexicon. Comput Intell. 2013;29(3):436–65. doi:https://doi.org/10.1111/j.1467-8640.2012.00460.x.
- Reinert A. Une méthode de classification descendante hiérarchique: application à l’analyse lexicale par contexte [A hierarchical descendent classification method: application to lexical analysis by context]. Cahiers L’Analyse Données. 1983;8(2):187–98.
- Sirola A, Kaakinen M, Savolainen I, Oksanen A. Loneliness and online gambling-community participation of young social media users. Comput Hum Behav. 2019;95:136–45. doi:https://doi.org/10.1016/j.chb.2019.01.023.
- Bücker L, Bierbrodt J, Hand I, Wittekind C, Moritz S. Effects of a depression-focused internet intervention in slot machine gamblers: a randomized controlled trial. PLoS One. 2018;13(6):e0198859. doi:https://doi.org/10.1371/journal.pone.0198859.
- Casey LM, Oei TPS, Raylu N, Horrigan K, Day J, Ireland J, Bonnie A. Internet-based delivery of cognitive behaviour therapy compared to monitoring, feedback and support for problem gambling: a randomised controlled trial. J Gambl Stud. 2017;33(3):993–1010. doi:https://doi.org/10.1007/s10899-016-9666-y.
- Ransing R, Adiukwu F, Pereira-Sanchez V, Ramalho R, Orsolini L, Teixeira ALS, Gonzalez-Diaz JM, Pinto da Costa M, Soler-Vidal J, Bytyçi DG, et al. Mental health interventions during the COVID-19 pandemic: a conceptual framework by early career psychiatrists. Asian J Psychiatr. 2020;51:102085. doi:https://doi.org/10.1016/j.ajp.2020.102085.
- Gambling Commission. Customer interaction—additional formal guidance for remote operators during COVID-19 outbreak. Gambling Commission; 2020 [accessed 2020 Dec 14] http://www.gamblingcommission.gov.uk/news-action-and-statistics/Statistics-and-research/Covid-19-research/Customer-interaction-%E2%80%93-Additional-formal-guidance-for-remote-operators-during-COVID-19-outbreak.aspx.