1 Introduction: HCI Research on SLSSs

Social Networking Services (SNSs) like Instagram and Facebook are an important element in people’s everyday life. Now, Social Live Streaming Services (SLSSs) are a new emerging field of SNSs, which could possibly even gain the same status in peoples’ lives. They combine elements of social networks, for example the interaction with the viewer and vice versa, and are a form of live-TV, with the streamer being, most of the time, boss, producer, and manager of his or her channel, deciding when he goes online and what content he wants to stream. The most popular examples of streaming platforms are PeriscopeFootnote 1, UstreamFootnote 2, YouNowFootnote 3, YouTube LiveFootnote 4, or Facebook LiveFootnote 5. Some services are focused on a specific interest or business, for example the service PicartoFootnote 6 which is mostly used to broadcast drawings and art, or TwitchFootnote 7, the main representative of the e-Sports industry. Some services are mainly used in a particular country, niconicoFootnote 8 in Japan or YYFootnote 9 in China, for example. To stream oneself, or something else like a live concert, a camera of a mobile phone or webcam is used that can be connected worldwide and sometimes even the whole day. An example would be a webcam on the ISS which streams videos of the earth from space every day. The streams can be found by searching for hashtags which were attached to the stream before broadcasting, or, like in the case of Periscope, via a world map on which current live streams are marked. Since many platforms offer multi-channel options, which means the linking of other SNSs like Instagram or Facebook to the streaming channel, potential viewers get a notification when the streamer is broadcasting.

Since SLSSs are a relatively new field of social media, they are not thoroughly studied yet. A few general studies on SLSSs and information behavior were conducted. Information behavior is classified as human behavior in relation to information and knowledge, for example HII: Human Information Interaction, and information and communication technologies, e.g. HCI: Human Computer Interaction. We found hsuman information interaction studies on Twitch [10, 18] and a few studies on the general SLSSs YouNow, Periscope, and Ustream [5, 6, 8, 9, 20]. General SLSSs are appreciated for the interaction between the streamer and the viewer, and the fact that SLSSs are very authentic [24], since everything happens in real life and cannot be staged. Furthermore, an SLSS can be used to broadcast breaking news and is adopted by citizen and professional journalists [7, 19]. A study on Periscope determined its role in the context of the economics classroom, since its grants the students to gain insight into other people’s lives in distant cultural areas [3]. Some studies were focused on legal problems concerning SLSSs [4, 11, 28]. Results showed that music and video copyright violations could be observed, furthermore, personality rights are another important factor in context with SLSSs. Gamification is an important aspect of SLSSs, since it engages the viewer or streamer while one is streaming or watching a broadcast [21]. Sometimes, the user even experiences flow, which is a state of total immersion into whatever one is doing, making him potentially forget about the time spend with the service [9]. A study on Twitch focused on the user-interaction with the live-stream and if the content can be influenced by the viewers. A “Twitch Plays Pokémon”-like setting was developed in which the audience shares the control of the main character of a game, thus leaving the control of the content mainly by the audience [14].

In accordance with the proposed model for information behavior research on SLSSs [29], this study investigates the aspect of content production.

Since some SLSSs found their expert community, it will be interesting to investigate what kind of content is produced on general SLSSs and if it differs among diverse services. Furthermore, distant cultural areas could be interested in contrasting kind of topics, making it another compelling aspect of this content analysis. Another aspect is the gender of the streamer that will be investigated in context with the content.

A study on information behavior in connection with the content of SLSSs and how it is related to the users, countries or the service, is still missing. To this end, a content analysis was conducted after a research model (Fig. 1) and the following research questions (RQs) formulated:

Fig. 1.
figure 1

Research model for this content analysis of SLSSs.

Full size image

  • RQ1: What kind of content is broadcasted on SLSSs?

  • RQ2: Are there differences regarding the streamed content between genders?

  • RQ3: Does the country of the streamer have an impact on the kind of content that is produced, in this case Japan, the U.S., and Germany?

  • RQ4: Is the content divergent for the three SLSSs Periscope, Ustream, and YouNow?

  • RQ5: Does the motivation of the streamer influence the content?

2 Method

To get a better understanding on the possible kinds of topics and content that people may talk about, perform or display while streaming, a content analysis [13, 15, 16] was implemented. To this end, a codebook based on literature regarding the use of social networking services was developed to get standardized data sets. A team of researches assessed, evaluated, and compared SLSSs’ users’ streaming behavior and the produced content. To guarantee a qualitative content analysis and a high reliability two different approaches were applied. First, the directed approach was used by selecting literature to get guidance on what kind of content gets produced on social network services. Second, the conventional approach was implemented via the observation of live streams to get an idea on what people stream about [12]. Several steps were taken in our content analysis. According to McMillan [16], our steps were: the drafting of the research questions and hypotheses. Then, a sample was selected by watching streams to get a general overview on them. The time span of the collection was set to four weeks, which is the third step. As a result, a spread sheet with the different categories and formalities was generated.

The content categories were marked in a tally chat and are the following: to chat; make music (m. music); share information (share infor.); news; fitness; sport event (sports); gaming; animals; entertainment media (ent. media); spirituality; draw/paint a picture; 24/7; science, technology, and medicine (STM); comedy; advertisement; nothing; slice of life; politics; nature; food; business information (busi. infor.).

The motivation categories were modeled after the uses and gratifications theory [1, 17]: entertainment (boredom, fun, hobby); information (to reach a specific group, exchange of views); social interaction (socializing, loneliness, relationship management, need to communicate, need to belong); self-presentation (self-improvement, self-expression, sense of mission, to become a celebrity, to make money, trolling).

Furthermore, the gender (male, female) of the streamer was listed as well.

The data was collected from three countries, namely Germany, Japan, and the United States of America, to inspect if differences between cultural areas are present. Furthermore, the research team had the required language skills for the three countries. To ensure that the streams originated from these countries, the declaration of the country was checked for each broadcast and service. Twelve research teams á two persons (advanced students of information science in Düsseldorf) were formed. Every coder received a spread sheet and marked everything in it that was applicable to the stream. While watching the stream the ‘four eyes principle’ was used. Every stream was watched by the two coders simultaneously, but independently for two to ten minutes. To reach a 100% intercoder reliability the entries were compared and if consensus could not be reached, the item was discussed. The streams were observed in two phases. First, the data was collected by observing the stream. If questions remained, for example about the age or the motivation of the streamer, he or she was asked via the chat system of the service.

The streams were not recorded, since it would require the consent of the streamer, but not every streamer communicates with the viewer or agrees with the recording. In the end, 7,667 streams in a time span of four weeks, from April 26 to May 24, 2016, were observed.

Descriptive statistics were calculated for the distribution of the content categories among the motivations, gender, countries, and services. Furthermore, for the inductive statistical analysis, the phi coefficient (mean square contingency coefficient) or Cramér’s-V and p-value were determined were applicable. The phi coefficient was used for two binary variables, meaning the content and motivations were matched, as well as the gender and content categories. To determine the correlation between the services and the content, as well as the countries and content, the Cramér’s-V measure was used. The value for the phi coefficient as well as the Cramér’s-V ranges from 0 to 1, with 0 representing no relationship and 1 equal values. Each resulting correlation was checked with their respective p-value for statistical significance.

3 Results

3.1 Content Categories

The produced content on SLSSs will be discussed in the following section.

Like Tang et al. [24, p. 4773] concluded, chatting (44.0%) is the category that can be found the most frequent on SLSSs (Fig. 2). Other content categories like sharing information on various topics (17.2%), 24/7 streams (like the ISS webcam) or slice of life (14.3%) are often represented. Entertainment media (11.7%) is another favored kind of content. Interestingly, a big category is “nothing”, no streamer was present, or any other activity could be observed, just an empty room. All these categories do not need any kind of preparation or a high amount of cognitive effort [25]. Presumably, this means that the lower the relative frequencies of the content categories are, the more effort is needed. If one wants to talk about politics (1.4%), business information (1.4%) or science, technology and medicine (STM) information (0.9%), one needs knowledge about these areas.

Fig. 2.
figure 2

Distribution of the content categories on SLSSs (N = 7,667).

Full size image

The data also suggests that an SLSS that specializes in some form of content has the monopole for it. The content categories drawing (0.8%) and gaming (5.5%) are not well represented on the general SLSSs YouNow, Periscope, and Ustream since Picarto (for drawing) and Twitch (for gaming) are the specialized SLSSs for those areas.

3.2 Content and Gender

In the following section, the differences of the content categories distributed among the genders will be determined. Overall, of the 7,667 streams, 4,548 streams (59.32%) were broadcasted by streamers who identified with either male or female. This means the streamer stated his or her gender, it was displayed in their channel description or the streamers assigned themselves the corresponding tag, e.g. #boy or #girl. For the other streams, the streamers either did not state their gender, or no person could be seen, if an animal was shown, for example. 2,782 (61.17%) streamers were male and 1,766 (38.83%) female. This distribution of genders among streamers is confirmed by the research of Tang, Veniola, and Inkpen [24, p. 4774] as well.

For the distribution of the content categories among the gender, only a few differences were observed (Fig. 3). Men seem to prefer topics related to spirituality (1.87%) and gaming (3.07%), whereas women are interested in streaming food related videos (1.99%) or 24/7 streams (2.51%). It can be assumed that overall, men are generally more drawn to gaming related streams than women, which statistics for the e-sports SLSS Twitch also determine [23]. The distribution of the chatting categories among the gender suggests that women are more likely to talk to their viewers (40.90%) than men (35.46%), but the phi coefficient does not support this assumption (.087), which is highly significant (.000).

Looking at the correlations, more specifically, the phi coefficient of the content distribution among the gender, none of the categories seem to be related to either one gender (Table 1). Even for the category gaming (phi coefficient –.093; p-value .000) we only find a very small or non-existent correlation. Since the data is not normal distributed, it is only possible to determine some trends, but because all correlations do not imply any relationship between gender and content at all, there seem to be no trends.

Table 1. The phi coefficient and p-value for the top ten content categories in relation to gender.
Full size table

In conclusion, there are only a few differences in information production behavior between the genders; this result seems to differ in comparison to other SNSs. Seymour [22] states that striking differences exist between men and women when it comes to sharing personal information on sites such as Facebook, and YouTube. Furthermore, only men with greater degrees of emotional instability were more regular users of SNSs [2].

3.3 Content and Services

Following in this section, the distribution of the content categories among the countries for each service, namely Periscope, Ustream, and YouNow, will be analyzed. Overall, we watched 2,960 streams on Periscope, 2,686 streams on Ustream, and 2,020 (2,021) streams on YouNow. Since YouNow is unknown in Japan, we have only one stream originating from there (excluded in this part of our analysis).

Overall, Periscope and YouNow seem to share their most popular content categories (Fig. 4). Here, just chatting with the audience and sharing information are favored. But a few differences can be marked. The content of YouNow seems to be more self-produced, for example performing a comedy sketch (2.04%), making fitness related videos (1.42%), e.g. showing or explaining exercises, or playing a video game (1.92%). Whereas on Periscope, the content is more passive, for example being in nature (3.18%), or broadcasting a sport event (1.88%). Those differences and the more active engagement of YouNow’s streamers with their content can possibly be explained with some users’ motivation to become a celebrity [5], which can be compared to success stories of some YouTube stars like Justin Bieber or Lindsay Stirling.

Fig. 3.
figure 3

Distribution of the content categories among the gender (N = 4,548).

Full size image

In contrast, Ustream offers another spectrum of content, which is more focused on different kinds of media. Entertainment media (10.51%), gaming (6.66%), news (4.95%), or sport events (4.06%) are broadcasted. Chatting (4.27%) or interacting with the viewer is not a big part of the streaming behavior on Ustream. Furthermore, streaming videos of animals (9.88%) and/or nature (5.48%) for 24 h a day can be observed often as well. The findings are in line with Ustreams agenda of being a SLSS which aims to educate its viewers, as NASA is an official customer of Ustream, as well as offering a platform to companies giving them the opportunity to broadcast live events, like concerts or sport events, for example [26, 27].

The calculations of the Cramér’s-V between the three services for 24/7 (.458; p-value .000) and chatting (.542; p-value .000) streams also provide clues that especially two of these categories are strongly dependent on the service (Table 2). On Ustream, only 4.27% of the streamer chat with their audience, whereas on YouNow, it is 41.84% and on Periscope 38.94%. For all other categories, there are at least low correlations; however, all correlations are statistically significant (.000).

Table 2. The Cramér’s-V and p-value for the top ten content categories in relation to the services.
Full size table

The results could be explained with the three services being used for various kinds of topics and are not focused on one expert area yet, like Picarto for drawing related content, for example.

3.4 Content and Countries

All three countries share chatting as the main content category (Fig. 5). But the number of relative frequencies differ among them. Nearly 40% of all German streams are just people chatting with the audience, in Japan, they are nearly 30% of the overall distribution, however, in the U.S. only about 18%.

Fig. 4.
figure 4

Distribution of content categories among Periscope, Ustream, and YouNow; sometimes multiple assignments.

Full size image

The most 24/7 streams (14.78%) originate from Japan; here other big foci on animals (6.60%) and nature (6.50%) exist as well. In contrast, the streams’ content that is popular in Germany is more related to entertainment, like entertainment media (10.18%), making music (5.34%), gaming (4.89%), or sport events (1.36%).

In the U.S., a lot of 24/7 streams (7.67%) can be found as well, but also, the highest number of streams in which nothing (9.04%) is happening. The U.S. is also the only country of the three which has advertisements (2.85%) as one of its top ten categories (Fig. 5), suggesting that for now, only streamer from the U.S. see live streaming as an opportunity to generate financial gains.

Fig. 5.
figure 5

Distribution of content categories among the U.S., Germany, and Japan; sometimes multiple assignments.

Full size image

But taking a closer look at the Cramér’s-V correlations, all the categories are highly significant weakly correlated (Table 3).

Table 3. The Cramér’s-V and p-value for the top ten content categories in relation to the countries.
Full size table

Overall, the streamers of the countries seem to have slightly different tastes in their favored content, but the correlations suggest that even people from distant cultural areas just want to fight their boredom [9] and talk to other people while sharing their thoughts and daily lives.

3.5 Content and Motives

Are there any differences among the distribution of the content categories and the motives, as well as any correlations?

For chatting, the percentage values shrink with the decreasing ranking of the four main motives (Table 4). But overall, the percentages of the content categories do not differ greatly among the four motives.

Table 4. Distribution of motives and content categories; sometimes multiple assignments; N = 7,667.
Full size table

Some highly significant phi correlations can be observed. Streamers seem to broadcast themselves and just want to chat if they are bored (.349, p-value .000) or want to socialize (.365, p-value .000). Furthermore, the motive of the need to communicate leads the streamer to share information (.318, p-value .000). The content of the information that is being shared can range among various topics. But this also implicates legal issues, since the streamer could handle sensitive data carelessly [28]. This suggests that streamers who are bored or need some form of human interaction just want to chat with their audience to idle away time.

The highest correlation (.401, p-value .000) can be found between streams that showcase spiritual aspects, for example a Holy Mass or citing quotes from the Quran, and if the streamer has a sense of mission. Furthermore, the streamer has a need for self-presentation (3.65%) and information (4.08%). In this case, the streamer feels a need to broadcast his believes or moral values to his or her audience, therefore using the service for an explicit goal. Since Periscope has the highest percentage of spiritual related content, it can be assumed that especially this service is used to this end.

4 Discussion

The content production on three general SLSSs (Periscope, Ustream, and YouNow) from three different countries (U.S., Germany, and Japan) was analyzed. The distribution of the content categories was determined among gender, countries, SLSSs, and motivations. Furthermore, the phi correlation/Cramér’s-V as well as p-values were calculated for the content and the genders, countries, SLSSs, and motivations.

The most produced content overall is chatting, followed by sharing information, 24/7 streams, slice of life, and also, nothing, e.g. streams in which only an empty room is being broadcasted. These kinds of topics imply that not much cognitive effort is needed for preparing the stream, in contrast to broadcasts that display the streamer performing fitness routines or talking about politics.

For the distribution of the gender, there are more male than female streamers active on the analyzed SLSSs. All in all, the calculations of the correlations show that the produced kind of content does not differ between the genders.

If the service is concerned, highly significant strong correlations can be observed for the categories chatting, and 24/7 streams, implying that 24/7 streams are more broadcasted on Ustream, and streams in which the streamer talks to his or her audience on Periscope and YouNow.

The country from which the streamer broadcasts is not correlated to the content, which is supported by highly significant p-values. Even though the countries are culturally different, the streamers only want to chat with other people.

Further correlations can be observed for the motivation of the streamer and the content he or she produces. If people are bored or want to share information, they usually just chat with their audience. Another finding was that streamers who have a strong sense of mission broadcast spiritually related content and vice versa. They also seem to like to present themselves and distribute information.

The study shows some limitations. It would be interesting to analyze other SLSSs and compare the produced content, for example Instagram Live, Facebook Live, or YouTube Live. Another aspect that should be studied is the origin of the stream. Even though we did not find contrasts for the categories and the three countries we choose, there could exist differences; for example, from countries in the middle east, especially Saudi Arabia, where YouNow is popular, or from China, where SLSSs are heavily used. Furthermore, it should be analyzed if the content of the streams differs per service in relation to the streams’ origin. This means that the content that was broadcasted on Periscope could be different for Germany or the U.S.

Another aspect that should be analyzed is the age of the streamer, since generational differences can be observed for SNSs, so it is likely that there are divergent results for SLSSs as well. Moreover, a closer look should be taken at the motivation of the streamer. Since we found statistical significant results for the content and the streamers’ motives, the age and gender could play an important role in this context as well.

All in all, the content analysis on SLSSs is still in its early beginning, but some valuable results could be observed and show a promising groundwork for future research.