Project solicitation: Prof. Dr. Roland Verwiebe and Dr. Lena Seewann 

Former project staff: Marie Theres-Hesse (MA)

The project has multiple topics. On the one hand, it focuses on the self-perception of YouTubers as an occupational group asking the following questions:

• What kind of meaning do YouTubers attribute to their actions on the platform? How are boundaries of recreational and professional activities defined?
• Which chances and obstacles does this new occupational field offer?
• Which strategies do YouTubers implement in order to use the platform professionally?

On the other hand, the project examines the inequality structures YouTubers are facing and how they come to be. This poses the following questions:

• Which (work) biographies are typical for people using the platform?
• Which role does the YouTube algorithm (as a non-human actor) play when it comes to inequalities of both visibility and income?
• How strongly do characteristics such as age, gender, educational level, descent or line of work influence inequalities on YouTube?
• How do sociotechnical skills or forms of organization affect the positioning on YouTube?

Current questions are: 

• Which content-related patterns of sentiment and discrimination can be identified within YT?
• What differences exist according to social structural characteristics (e.g. gender, migration background) and characteristics of the social media platform (sector of the YouTube channel, number of subscriptions, views and likes (reach / success))?
• What individual coping strategies do content creators develop to deal with sentiment and discrimination practices on social media? How do content creators perceive YouTube's algorithm in their work? To what extent do content creators perceive the algorithmic system of social media platforms as an independent entity that has its own ascribed agency?

Classification survey 

A classification survey is used to generate variables that are partially visible on the platform but cannot be scraped. With the help of this survey, student assistants systematically classified the characteristics of 5000 channels (e.g. gender, age, migration background, but also channel-specific characteristics such as the number of people involved, other sources of income such as Patreon or sponsored videos, and the specific topic of the channel). The classification survey gives the coders three stimuli one after the other (the channel description, the last video and finally other sources such as Wikipedia or LinkedIn), which help to understand where individual pieces of information come from and records each person within a channel individually (see image).

• This data serves as an extension of the scraped data set and allows the analysis of sociological questions on YouTube, which includes demographic characteristics of the channels and is therefore unique in the German-speaking world.
• The next step will be to expand this data set using machine learning methods.

In addition, a variety of analytical methods from the fields of qualitative and quantitative social research, mixed methods and computational social science are used, including:

- A thematic analysis (Braun & Clarke, 2012) to explore the question of how content creators perceive the YouTube algorithm in their work

• Initial findings show the perception of the algorithm as an independent identity with agency and sometimes a focus on the four themes of the unpredictability of the algorithm, the dynamics of the algorithm, the power of the algorithm and inequality through the algorithm.

- A type formation (Kelle & Kluge 1999) on coping and strategies that content creators develop to deal with sentiment and discrimination practices on social media

• First findings (→ see conferences)

- Dictionary, machine learning (e.g. Lasso regression) and deep learning (e.g. Bert algorithm) methodologies for generating variables such as gender, but also sentiment or hate speech from YouTube comments

→ Based on this, various regression and cluster analyses

• First findings ( → see conferences & publications)

To make data from the YouTube platform accessible to us, we use the YouTube Data API. APIs (Application Programming Interfaces) offer the option of integrating functions of the website in question into your own websites or apps by accessing them using various programming languages. In our case, we use Python to retrieve and store data in a structured way using the interface (API). We are therefore able to scrape information on channels (e.g. name, description, number of subscribers, number of videos, profile picture), videos (e.g. number of views, number of likes, topics) and comments (e.g. content, number of likes, name of the channel). The corresponding Python scripts are available here.

In addition, the source code of the channels was read, which contains additional information that cannot be retrieved via the YouTube Data API.

Since December 2022, data has been scraped daily from the 120,000 channels, which now amount to almost 200GB. This is cross-sectional data, although a longitudinal data set has also been created for selected channels and videos since November 2023.

Conferences

• Philipp, Aaron; Verwiebe, Roland; Weißmann, Sarah (2023): „Sentiment und Diskriminierung auf Social Media? Ein Mixed-Methods-Ansatz zur Untersuchung von Kommunikations- und Debattenstrukturen innerhalb der YouTube-Community“, DGS Section "social inequality and social structure analysis" - Spring Conference 2023
• Philipp, Aaron; Verwiebe, Roland; Weißmann, Sarah (2023): “Sentiment and hate-speech on social media. An analysis of the relevance of gender and race for the emergence of new communication structures within the YouTube Community”, General Online Research (GOR 23)

Publications

• Seewann L, Verwiebe R, Buder C, Fritsch NS. "Broadcast your gender." A comparison of four text-based classification methods of German YouTube channels. Front Big Data. 2022 Sep 14;5:908636. doi: 10.3389/fdata.2022.908636

Further informations

• You can find the Phython scripts here via GitHub to extract your own data such as channel information, video information and comments using the YouTube API.
• Article about the research project „Portal Wissen. Das Forschungsmagazin der Universität Potsdam“ 01/2023