Designing Digital Inclusivity: Mapping and Shaping Online Environments for the LGBTQ+ Community (Data Donation Research)

Team Members

Daniel Jurg
Luuk Ex
Wouter Nieuwenhuizen
Joris Veerbeek
Stijn Peeters
Ike Picone
Sarah Vis
Annika Pihl
Bartika Choudhury
Khadija Kone
Joanna Flannagan
Zhuo Wang

Contents

1. Summary of Key Findings

Building on the observations in de Groene Amsterdammer that the LGBTQ+ community in the Netherlands is increasingly subjected to insults and threats on social media (van Dijk et al., 2023), this winter school project explored the online spaces of three LGBTQ+ individuals on Instagram and TikTok through an innovative methodological lens: data donations. We analyzed the accounts each participant interacted with over a period of approximately ten years on Instagram, categorizing them as queer or unknown to diachronically discern the presence and representation of queer content on their feed. On TikTok, we gathered metadata from all viewed videos of one participant such as hashtags, to map the topics present in the content consumed over a period of six months. While the project was initially mostly concerned with hate speech, we ended up charting the queer spaces navigated by our participants more broadly. Although limited to three participants, our exploratory study testifies to the possibilities, strengths and weaknesses of small-scale data donation studies, illuminating their potential to deepen our understanding of digital media engagement (for corresponding figures please have a look at the poster or find them in the findings section).

Main findings:

1. A co-hashtag analysis of the watched videos of participant 3 on TikTok revealed that queer content is central to the donator’s online environment (Figure 1).
2. Despite only representing 8% of videos watched (see Figure 2), videos we categorized as queer reported the highest rate of video watch time, indicating the TikTok donor pays more attention to this content (see Figure 4).
3. No evidence was found that high engagement with queer content on TikTok influenced received advertising efforts.
4. The graphs (see Figures 5 & 6) represent a trend in the participation and engagement of the participants on Instagram over eleven years. Both graphs present a collective shift in behaviors among the participants that shift from mostly individual accounts to later on, larger, public accounts we categorized as queer accounts.
5. On Instagram, these findings express and establish a difference in how the platform is used by the participants. Within a common identity of being identified as LGBTQ+, these participants engaged with content differently. This visual analysis of the graphs represents an overtime overview of the shift in trends that enable engagement within this community of studies.
6. The graphs generate a view of how engagement among participants on Instagram varies over time. It also highlights the kinds of accounts interacted with, drawing attention to the changing practices of these participants in this period of study. (See Figures 5 & 6).
7. Research with data donations on a small scale seems particularly relevant in relation to further in depth interviews with participants. Suggesting a fruitful combination between more explorative Digital Methods research and more classical audience studies.

2. Introduction

The online safety of social media users has become a central concern in contemporary discussions on digital media. On February 14, 2024, a major U.S. Senate hearing featuring chief executive officers Linda Yaccarino of X (formerly Twitter), Shou Zi Chew of TikTok, Mark Zuckerberg of Meta, and Jason Citron of Discord garnered global attention as U.S. lawmakers aim to crack down on the spread of harmful online content. Amidst this current widespread public scrutiny concerning the spread of online hate and disinformation, it's challenging to recall the initial optimism surrounding the advent of digital media and their potential for democratization and fostering new communities, especially for traditionally marginalized groups. Yet, it has extensively been argued that social media platforms offer a conducive environment for the construction of performative identities that continually reshape (sexual) identities and reinforce their legitimacy, which play a pivotal role in the formation of minority group identities, for example, such as those within the LGBTQ+ community (Das & Farber, 2020).

However, despite this potential for performative play and connection, a growing body of research now indicates a global surge in hate directed at women, teenagers, and LGBTQ+ individuals on social platforms (Negi, 2023). The role of digital media for marginalized groups seems to be caught in a complex tension: on one hand, they provide spaces for like-minded individuals to perform identities and connect with a larger community; on the other hand, these inclusive spaces are increasingly becoming more hostile. Navigating this issue requires not only scrutinizing major social media companies or track and map explicit harm, but also, in more general terms, gain a deeper understanding of the experiences of marginalized groups within various online environments. This raises broader questions concerning the role played by digital and social media in the development of, for example, queer identities within their everyday life.

Within the broader ’Inclusive Online Environments’ (IOO) project of the Rathenau Instituut, this winter school project, Designing Digital Inclusivity (DDI), presents a first step in mapping the online environment of individuals who identify themselves as ‘members’ of the LGBTQ+ community. The overarching aim of the IOO is to examine the current state of LGBTQ+ online environments on platforms such as TikTok and Instagram and subsequently develop improved inclusive digital spaces. It thereby seeks to merge the knowledge and experiences of the LGBTQ+ community with the broader research community and societal actors.

While this project already makes some initial empirical observations, one central element of the DDI project is to explore the possibilities with using data donations in research more broadly. While Digital Methods research has convincingly made the case that digital trace data collected via the methods of the medium itself allows for the study of broader societal and political events and trends, it is important to understand that many so-called ‘small acts of engagements’ (e.g., likes, comments, shares) afforded and captured by the specific platform infrastructure are embedded within the everyday lives of individuals who might have very ambivalent relations towards online spaces they occupy (Picone et al., 2019). One of our central observations and recommendations following this week is that data donations research can (and should) productively combine the power of Digital Methods research and visualization, with more traditional audience studies insights and methods, such as in depth interviews with data donors.

3. Research Questions

Initially, it was the aim of the project to specifically look at the harm LGBTQ+ individuals might experience on social media platforms. However, given the specific API access and ability to scrape content, we were restricted in our ability to detect harmful content. Arguably, much of the harmful activity happens in the comment sections or is hidden from the algorithm with various obfuscation techniques such as ‘algospeak’ that seeks to modify language or symbols imagined to be part of algorithmic content moderation (Steen et al., 2023). It is worth exploring this issue more in depth in future research. For now, within the context of the Digital Methods Winter School 2024, our specific focus was on how much and what type of queer content was engaged with and its potential reflection in terms of advertisements. In essence, we aimed to understand how LGBTQ+ individuals engage on social media and how companies perceive these users. This objective is encapsulated in the following two research questions:

1. What specific queer content is being shown to LGBTQ+ individuals?
   1. Is there a dedicated queer side on their social media feed?
   2. Do we see a pull/push into more queer-specific communities over time?
2. How does the platform see LGBTQ+ individuals in terms of advertisement?
   1. How do social media companies target individuals based on their gender or sexual preferences?

4. Initial Data Sets

The DDI project started with contributions from three data donors. Each of these donors requested and shared their Instagram data, while participant number 3 additionally requested and contributed TikTok data. We opted to divide our analysis on the basis of the platform, given that both platforms provide different data that requires different focus points and methods. From the two platforms we had access to the following information:

TikTok

The TikTok data set of participant number 3 contained 17.788 unique videos watched over 6 months (June until December 2023). This initial data provided by TikTok is a list of ‘watched videos’ and the ‘time watched’. Additionally, we used 4CAT: Capture & Analysis Toolkit to extract metadata of all the watched videos that were still online . This allowed for a dataset within 4CAT enriched with the following metadata:

● Body (caption of video), video id, author name & id., links to the video, music used, hashtags, and thumbnails.

● Date the video was posted by the author and date it was viewed by the donor. In addition to the ‘unique’ video dataset, we constructed another dataset that contained all the videos watched, even if it was multiple times. This resulted in a total of 21,000 videos.

● Video files of the videos watched and image files of their thumbnails.

Further data calculated/obtained by the researchers included:

● Transcripts of all videos in the data set.

● Time spent on watching the video (by the donor), was calculated by the researchers based on the time the video was watched and the next video would start (taking into account watch sessions).

● Watch rate, measured as a portion of the watch time compared to the length of the video (provided in the TikTok data).

Instagram

While TikTok enabled the collection of a significantly larger amount of metadata, the data gathered from Instagram offered less detailed information but spanning a longer timeframe, in some cases over 11 years. The data utilized in this study from the provided dataset were based on the following metrics:

● Follower and following metrics: The ‘follower following’ metrics act as a source to generate an algorithm for the platform to generate personalized content for the user.

● Suggested accounts viewed: These metrics help us analyze the first question of what the user sees on Instagram, these include accounts that appear on their explore feed and accounts that are suggested over time. By clicking and viewing them, it is counted as an engagement.

● Liked posts of varied accounts by the user/participant: According to the published records of the platform, the content that a user sees is dependent on their engagement with various accounts. This content is further catalyzed by the interactivity of the posts (related to the accounts and larger genre) that the user manually likes.

Ethics of Data Collection and Analysis

A key concern in data donation research is the ethical procurement and management of participants' data. This project was granted ethics approval by the UvA ethics board, enabling us to proceed with our research (van Driel et al., 2022). All data donors were thoroughly instructed on how to collect their data and provided informed consent, which authorized us to share and utilize this data during the Digital Methods Winter School 2024. To protect the data, we operated within a secured OneDrive folder of the UvA and a private dataset within the 4CAT: Capture and Analysis Toolkit. Following the completion of the Digital Methods Winter School 2024 report, all data will be permanently deleted from UvA systems.

While our goal was to ensure the anonymity of research subjects, working with small-scale data donations presented certain challenges. In-depth analysis revealed that some researchers could potentially infer the identity of the donor through specific followed accounts or engaged posts (the identity of the data donors was never confirmed or shared). In addressing these concerns, we contend that the solution does not lie in further removal of identifying details, such as anonymizing account handles before data sharing with fellow researchers, but in establishing more stringent agreements among researchers themselves, for instance, through signing data processing agreements. In our situation, all participants agreed to the condition that the data would only be utilized within the context of Digital Methods Winter School 2024 and removed after the project.

In small-scale data donation research, balancing privacy concerns with research goals presents a challenge, potentially compromising the latter for the sake of the former. High ethical standards for using user data, driven by legislation primarily targeting social media companies with vast data reserves, may not directly apply to the comparatively modest datasets collected by social scientists and digital humanities researchers. It would be unfortunate and counterproductive to curtail the exploration of valuable data sources like data donations due to an overly cautious approach to user consent. Within the context of joint data analysis sessions, such as during the Digital Methods Winter School 2024, researchers should be able to access the data relatively freely within the group, even to the point of recognising the donor while at the same time be bound to strict non-disclosure agreements when it comes to sharing information outside the group. This point will be further expanded as one of the central applications of small-scale data donation research is the opportunity for data reflection in subsequent interviews.

5. Methodology

This project was based on a (computational) grounded theory approach (Nelson, 2020) that seeks to theorize from the data itself rather than reject or confirm specific hypotheses from the literature. More specifically, we aimed to move from more general large-scale observations to more specific analyses based on those observations. All of the data was uploaded into the 4CAT: Capture and Analysis Toolkit for further analysis (Peeters & Hagen, 2022). Initially, the project primarily focused on hate speech, but it evolved to more broadly mapping the queer spaces our participants navigated, specifically based on the data that was available. Below we outline the two specific approaches to TikTok and Instagram.

TikTok

The first step of analyzing the TikTok data was to create a co-hashtag network analysis to determine the top content clusters within the data set with over 17.000 uniquely watched videos. A co-hashtag network analysis is done by linking all the hashtags of the videos using the network module in 4CAT. We then uploaded the network file into Gephi, an open-source network analysis and visualization tool, for further cluster analysis and visualization (Bastian et al., 2009). We removed the specific hashtags the related to so-called algorithm hacking, e.g., #foryoupage, #viral, #fyp.

Based on the co-hashtag network, we identified five main clusters: (1) Queer, (2) Neurodiversity, (3) Fitness, (4) Pets, and (5) Beauty. These clusters were determined through a quantitative modularity clustering within Gephi, which calculates the proximity of hashtags to detect (content) communities, in combination with a qualitative thematic analysis of the hashtags within the identified content communities. These clusters were then used to make a selection of hashtags for a further categorization of top content clusters for additional analyses of likes, favorites, and watch rates by clusters which were compared throughout the observation. We added a food cluster, which was not labeled within the final network graph, but which in earlier renderings of alternative networks appeared to also form a large content cluster.

After determining content clusters, supporting metrics were calculated such as time spent watching the video by the data doner, how many percent of the content clusters made up from the whole data set, and the percentage of queer videos over time.

Instagram

The approach to mapping Instagram data differed significantly due to the limitations in data collection. Although Instagram offers information about the accounts mostly engaged with by the user, accessing metadata for those engaged posts is challenging. Which has to do with access of various scraping tools to the platform. Thus, rather than focusing on analyzing platform-provided metadata, we pursued a more qualitative approach to map engagement with specific accounts. Since Instagram provides engagement data spanning the entire account history (unlike TikTok, which offers only six months of data when it comes to watched videos in browsing history), we could analyze user engagement over a longer timeframe. Our strategy involved conducting close readings of the accounts and categorizing them as either queer or unknown. This categorization wasn't based on whether the individuals or organizations were actually queer, as that determination is unfeasible and not directly pertinent to the research. Instead, we coded accounts based on their open and specific engagement with queer issues on their profile pages. For instance, if a participant engaged with a particular queer influencer explicitly presenting itself as queer, for example by using a rainbow icon in the account, we coded the account accordingly. This allowed us to map the development of queer content on the participants' feeds.

These engagements were based on two levels:

1. The posts from accounts that the participants liked.
2. The accounts that were suggested and viewed by the participants. Participants here refer to the data donors, hence the data is seen from their perspective.

To stay consistent, the definition of Liked Posts and Suggested Accounts viewed were established beforehand:

• Liked Posts: This includes the posts of certain accounts that the participant has liked. Since the numbers of engagements vary over time, an analysis of the ‘most engaged with’ and the ‘Least engaged with’ can be made. Due to a large amount of data, the top 15 most engaged accounts were taken into consideration for the research.
• Suggested Accounts viewed: here a certain assumption is made as to how the platform itself suggests content and how much of it was viewed by the participant in general.

Limitations

Initially, it was our aim to also perform a cross-platform analysis of participant number 3, as we had both Instagram data and TikTok data. However, working with data donations depends heavily on the information provided by social media platforms. Between TikTok and Instagram, which structure user data differently, the discrepancies presented too many methodological challenges to cross-platform analysis within the short timespan of the project. Examining data donations provides a distinctive perspective on platforms from individual viewpoints. At this moment, we found it rather difficult, based on following the data provided, to really compare link engagement on Instagram and TikTok. It would be very interesting if future research explores this point further, especially in light of research on media repertoires.

A second important limitation to the research was the discoverability of communities with co-hashtag analysis. We ran multiple network analyses that resulted in slightly different clusters, depending on the specific settings and filtering criteria. In the end our network graph does not highlight the ‘food’ category that was discovered in an initial rendering of the network visualization. However, despite this inconsistency, the methodology for further analysis is detailed in the appendix and can simply be recreated. That being said, this labeling was done within the brief period of the winter school and further research should make a more robust and justifiable categorization.

A final limitation that we experience is that for Instagram it was rather difficult to determine what specific data was kept in which folder. We discovered only near the end of the project that when requesting data in HTML-format, as opposed to folders, that there is additional information on the type of data contained in various folders.

6. Findings

This section presents the main findings from our exploration of engagement with queer content of our three participants on TikTok and Instagram. On TikTok we first provide an overview of the dominant content clusters based on hashtags in the watched videos of participant three, which are then subsequently analyzed per content cluster in terms of engagement over time. Thereafter we present the insights from Instagram, highlighting the emergence of engagement with queer content over a period of ten years and engagement with advertisement over a period of three years (please also check the poster here).

TikTok

The network graph of Figure 1 made with Gephi visually represents the primary topic clusters derived from videos watched from June until December 2023. The clusters are categorized based on the hashtags present within the TikTok video descriptions. Within the network, each node illustrates the number of hashtags used within the dataset. The larger the node, the higher the frequency of a hashtag used within the dataset. Upon analyzing the frequency of the identified categories, it became evident that recurring central themes included: queer, neurodiversity, fashion, hair, pets, and gym, often intersecting with each other.

Figure 1: A queer co-tag network

Figure 2: Content clusters by numbers

After completing the queer co-tag network, we sorted the dataset by the date of the video watched and computed the cluster percentages over time using a rolling window of 300 consecutive videos. A rolling window averages data within a fixed number of points to smooth fluctuations and reveal trends. We then computed the average watch time for each video in a specific cluster by dividing the video's length by the time gap between two consecutive videos. In Figure 2, several observational metrics were added such as the percentage of favorited and liked videos based on the content cluster within the overall data set with over 17.000 unique videos.

For Figures 3 & 4, we sorted the dataset by the date of the video watched and computed the cluster percentages over time using a rolling window. A rolling window averages data within a fixed number of points, moving through the dataset to smooth fluctuations and reveal trends. In our case, the window size is 300, meaning that at any point, the average includes 300 consecutive TikTok videos. We observed a notable variance in volatility across different categories. Beauty-related videos tend to appear in short, intense bursts, at times comprising over 40% of the last 300 videos served. Conversely, videos on queer topics are distributed more evenly, maintaining a relatively consistent and low percentage throughout.

When we contrast this with the average duration each video is watched (determined by dividing the video's length by the interval between two successive videos), we find unexpected contrasts.

Queer videos, though less frequently served, consistently show the longest viewing durations. This suggests that, for such videos, TikTok 's algorithm might not be accurately interpreting the engagement signals from this user.

Figures 3 & 4: TikTok time-series analysis

Instagram

The findings for Instagram are divided between the 3 data donors, but are presented together in Figures 5 & 6. While the streamgraphs (see Figure 5) highlight when ‘recommended accounts’ were viewed and labeled for queer to show how queer content seeped into each data donator’s Instagram feed, the RankFlows (see Figure 6) zoom in on how they individually engage with queer accounts (via likes).

The suggestions made by Instagram's algorithm for individual accounts are shown in the stream graph. This data implies that there is an account recommended by Instagram and which was then engaged with by the user. By distinguishing between queer and unknown content, we can understand how and when queer content became part of each participant's social media experience. Whereas for participants 1 and 3 queer content seemed to be dropped into their Instagram feed at a specific moment (both around 2019). Participant 2, for instance, got a slower and more subtle introduction of queer content (already starting in 2017) which finally increased in 2020.

Figure 5: Instagram Streamgraph

By then zooming in on each participant’s actual engagement (Liked Posts) with these accounts, the 3 separated RankFlows represent the top 15 accounts that the participants liked the most posts from. Here we can see how differently the 3 participants were engaging with the queer content as proposed by the algorithm

Figures 6: Instagram Rankflow

Participant 1: Liked Posts & Suggested Accounts Viewed

The findings suggest that the participant mainly interacted with personal content in the initial period. Gradually they moved to engaging with queer content, this trend starts from the year 2015. A noticeable change is visible in the following years. This engaged content ranges from personal, political, and other (accounts which cannot be classified under one of the former labels). This participant was proposed by a mix of some queer content, the accounts that posted content adhering to the LGBTQ+ community.

• First noticeable trend was seen in the year 2016
• A big increase in engagement with queer public Instagram accounts (accounts that are set to ‘public’ and can be viewed by anyone) in 2022
• In 2022 The data donator is almost solely liking queer posts
• Participation in queer politics in times of election
• Recommended accounts viewed were seen to be queer between 2019-2020.

Participant 2: Liked Posts & Suggested Accounts Viewed

The findings of the participant suggest the engagement with more specific drag content. A slightly earlier engagement can be found starting from 2014 and persisting throughout the feed of constant engagement through likes. Except for queer content on Instagram, it is also interesting to see that this participant interacted a lot with travel-related accounts from 2016 to 2017. And starting in 2019, meme content engagement was most prominent. In general, the participant interacted the most with culturally labeled content on Instagram.

The Suggested Accounts Viewed ranges through a multiple level of queer content. It is a mix of drag content along with content that engages with community-related activities (e.g. the content of Pride Month activities).

Participant 3: Liked Posts & Suggested Accounts Viewed

The data available shows the shift in trends with how the participant interacts with content in general. Eventual shift to political content and content that had a large influence, such as cultural pages. It is worth noting that the political content is queer-related, such as content about queer activism. Meanwhile, the cultural content that this participant interacted with included posts related to queer communities and clubs. Based on the data, views of queer suggested accounts first appeared in 2019. This participant engaged with more culturally relevant queer content, such as content related to the queer community.

7. Discussion

While our investigation initially had the intention to map harms and dangers LGBTQ+ individuals face within their platform feeds, its primary focus shifted. Based on available data, we redirected our efforts toward a broader objective: mapping and assessing the visibility of queer content in the online spaces frequented by individuals within the LGBTQ+ community, utilizing data donations and exploring their potentials and challenges. Consequently, we wish to underscore three key discussion points emerging from our research: (1) accounting for platform differences, (2) assessing the impact of small acts of engagement, (3) assessing the visibility of queer content. These points culminate in a broader conclusion about the power of small-scale data donation studies in relation to more traditional audience studies methods.

Firstly, audience studies employs the concept of media repertoires to argue that our understanding of people's media use hinges on recognizing that individual patterns of media consumption comprise various media and that the relationships between these components within their catalog are crucial (Hasebrink & Domeyer, 2012). To this end, participant three offered the potential to compare Instagram engagement with TikTok engagement. However, we soon discovered that such a comparison was hindered by the different data points offered by the platforms. Instagram offers relatively little metadata capturing possibilities, whereas researchers can easily extract all the metadata from the TikTok videos that were watched by the participant. In addition, TikTok offers a detailed snapshot of engagement over a period of 6 months (possibly to a year) and while Instagram has relatively little detailed data compared to TikTok it does offer, for example, engagement data with accounts (liked, watched, viewed) for over a decade. While allowing a more limited perspective, it does provide an excellent overview of shifts within the users’ life, such as going to study, getting a job, moving to a new city. Additionally, we found evidence that suggested that Instagram and TikTok served very different purposes for participant number 3 (for example the Dutch election was very present on Instagram but not at all on TikTok), but this was a difficult point to make given the different data points. It would be very beneficial for future research to explore the comparability of various data donations on a smaller scale, and especially which data points in the data donations are best linked.

Secondly, we found that especially TikTok allows for interesting data for quantitative mapping of engagement given some additional variables are added to the mix. The detailed and rich information on the ‘liked’ and ‘favorited’ videos in comparison to the ‘watched’ videos allows for a better understanding of how certain interaction of the user with the platform impacts the content that is present on their feed. We could, for example, observe certain intense binge sessions of beauty videos that at certain points comprised more than 43 percent of the videos being served to the user. Here the question is, of course, how to explain such a percentage, which could suggest recommendations of the platform, but also the fact that the user often goes to the page of specific influencers. Here more contextual information would be beneficial to understand these points.

Thirdly, whereas TikTok contained much interesting metadata that helped to map LGBTQ+ content, Instagram requires much more qualitative engagement. Here the most interesting finding is that Instagram served more as a personal friend platform in the early 2010’s but increasingly became a more public platform to follow influencers and establishments. TikTok data allows for the more traditional Digital Methods approach to follow the methods of the medium to label and connect content, such as a co-hashtag analysis. This benefits from the labor that users and the platform put into labeling and targeting content to specific users. However, our Instragram research required a qualitative mapping of ‘queer’ accounts. Defining ‘Queer content’ raised challenges and led to a wide variety of internal discussion within the research team. We employed various strategies, such as identifying rainbow icons and specific pronouns on Instagram where users specifically employed symbols associated with the LGBTQ+ community. However, complexities arise with content like celebrity posts or images of two individuals kissing without contextual information and making assumptions here can be problematic. Here, again, it will be necessary to analyze certain accounts together with the participant and possibly involve the data doner into the coding process.

9. Conclusion

Within the broader framework of the "Inclusive Online Environments" (IOO) project by the Rathenau Institute, our Digital Methods Winter School 2024 project, "Designing Digital Inclusivity" (DDI), embarked on an initial endeavor to map the online experiences of individuals identifying as members of the LGBTQ+ community through data donations. This exploratory study illuminated the diversity of data provided across platforms and the potential for insightful analyses. A notable observation is the highly decontextualized nature of this data, echoing a well-documented critique regarding the study of commercially generated data, which often reflects an economic logic (boyd & Crawford, 2012). Despite this, our project underscores the unique insights and specific impact assessments that data donations can facilitate over time rendering visible individual feeds and various small acts of engagements

However, in light of the justified critique of decontextual commercially generated data, we posit that the true strength of small-scale data donation research lies in its ability to generate targeted hypotheses or questions, which should be further enriched by direct conversations with the data donors. This approach can serve as a compelling elicitation technique, where data donations are reflected back to the user. In this context, the focus on data visualization within Digital Methods research proves invaluable. Rendering user feeds visible in a manner that is accessible not only to academics but also to everyday users demands a specific way of thinking about communicating insights. We suggest that co-hashtag graphs on TikTok and RankFlows on Instagram offer effective ways to convey information to users and act as "data mirrors" for deeper reflection. While efforts to scale up data donation research are well underway and necessary, it's crucial to also develop more qualitative approaches that facilitate more personal and individual reflections.

References

boyd, danah, & Crawford, K. (2012). Critical Questions for Big Data. Information, Communication & Society, 15(5), 662–679. https://doi.org/10.1080/1369118X.2012.678878

Das, S., & Farber, R. (2020). User-generated online queer media and the politics of queer visibility. Sociology Compass, 14(9), e12824. https://doi.org/10.1111/soc4.12824

Negi, C. (2023). An Overview of Worldwide Cyberbullying and Cyberviolence Against Women, Teenagers, LGBTQ on Social Media: Facebook, Instagram, Telegram, WhatsApp, Snapchat, YouTube, LinkedIn and Twitter. SSRN Electronic Journal.

Nelson, L. K. (2020). Computational Grounded Theory: A Methodological Framework. Sociological Methods & Research, 49(1), 3–42.

Peeters, S., & Hagen, S. (2022). The 4CAT Capture and Analysis Toolkit: A Modular Tool for

Picone, I., Kleut, J., Pavlíčková, T., Romic, B., Møller Hartley, J., & De Ridder, S. (2019). Small acts of engagement: Reconnecting productive audience practices with everyday agency. New Media & Society, 21(9), 2010–2028.

Transparent and Traceable Social Media Research. Computational Communication Research, 4(2), 571–589

Van Driel, I. I., Giachanou, A., Pouwels, J. L., Boeschoten, L., Beyens, I., & Valkenburg, P. M. (2022). Promises and Pitfalls of Social Media Data Donations. Communication Methods and Measures, 16(4), 266–282. https://doi.org/10.1080/19312458.2022.2109608

Appendix

Content clustering with specific terms on TikTok:

Queer: wlw, lgbtq, lesbiansoftiktok, lesbiancouple, trans, nonbinary, pridemonth, queertiktok, gay, masc, bisexual, qaypride, queer, kinktok, dom, softdom, kinkcommunity, gender, lgbtqia, lesbian, queertok.

Cat/dog/pets: pets, catsoftiktok, animals, funnycat, cutecat, catlovers, petsoftiktok, kitten, funnycats, kitty, catmom, cats, dog, dogsoftiktok.

Neuro-diversity/autism/adhd: adhd, adhdinwomen, mentalhealth, selfcare, mentalhealthawareness, autism, neurodivergent, neurospicy, adhdcheck, adhdawareness, adhdtok, actuallyaustic, adhdbrain, audhd4.

Fitness: gymtok, gymgirl, workout, glutesworkout, girlswholift, gymhumour, gym, fittok, bodybuilding, gymmotivation, legday, fitnesstips, mindeset, inspiration, bodypositivity, fitness.

Food/recipes: vegan, veganrecipes, recipes, plantbased, foodtiktok, mealprep, highprotein, foodie, foodtok, healthyrecipes, garden, easyrecipe, cooking, food, health, weightloss, recipes.

Hair/Beauty/Fashion: Blonde, hair, bob, shorthairstyles, hair transformation, hairtok, y2k, barbieaesthetic, outfitideas, y2kgrwm, fashiontiktok, fitcheck, designer, vintage, outfitinspo, outfit.