The universe of concepts included in the title and keywords of the documents registered demonstrates the correct choice of terms for the documentary search and highlights the relevant and significant aspects of this topic (see Figure 3).

Figure 3. Recurring main concepts in the research

Source: prepared by the authors using Word it out.

With regard to the dimensions identified in the scientific literature related to topics concerning minors and their interests (United Nations Children’s Fund [UNICEF], 2019), which have been investigated using SBD, those that stand out are studies focusing on the analysis of language and opinions on social networks (19%; n=11), online bullying (15.3%; n=9), risky practices that involve the invasion of privacy of adolescents, and educational policies (10.3%; n=6).

The remaining entries (see Figure 4) address topics as disparate as the situation and forecast of the labour market at 8.6% (n=5), depression, sexting, and teenage suicide, each at 5% (n=3), as well as the social influence of social networks and youth radicalisation at 3.5% (n=2). These are followed by juvenile delinquency, sport and leisure, teenage pregnancy, environment, mobile phones, fashion, and privacy, each with 1.7% (n=1) (identified as “All others” in the graph)

Figure 4. Dimensions addressed in the research on SBD and minors

Source: prepared by the authors

4.3. Knowledge domain of the studies

Another of the aspects examined in this bibliographic research was the knowledge domain to which the publications belong. To this end, the publications have been segmented into four categories in accordance with the profile of the journals in the sample:

• Technical: deals with issues such as computing, engineering and informatics.
• Social sciences: these are composed of 13 main branches: Anthropology, History, Legal Science, Economics, Sociology, Psychology, Linguistics, Semiology, Archaeology, Demography, Human Ecology, Pedagogy, and International Relations (Centro de Estudios Cervantinos, 2020).
• Mixed: this combines the categories of technical and social science journals.
• Medicine.

This classification does not include the four European Union reports that do not have a topic bias. Of the remaining 93% (n=54), 61% (n=33) are technical journals, 19% (n=10) are medical journals, 13% (n=7) are mixed, and 7% (n=4) are social science journals (see Figure 5).

Figure 5. Knowledge domain of the publications

Source: prepared by the authors

4.4. Applied SBD techniques and samples

The techniques applied vary according to the object of study, and the authors do not claim that there are differences between SBD and BD techniques. Researchers conduct their studies by applying statistical techniques such as frequency analysis to detect the importance of one or several words in the network, as well as diverse correlational statistical tests. Moreover, the application of various algorithms to detect trends have also been found, in addition to the use of bivariate statistical analysis, cluster analysis, temporal analysis and structural equations, among others. Similarly, studies have also been found that employ the vocabulary used in social networks to apply sentiment analysis and machine learning, which allows for the analysis and coding of large-scale texts. In other cases, data mining, text mining, and opinion mining are used to identify and analyse the comments, posts and messages that young people leave on social networks.

Thanks to the use of SBD, the samples that researchers work with are very large. Among them, Kim, et al. (2019) stand out. In order to understand the current and future experiences of adolescent bullying, these researchers analysed 436,508 web documents from 279 online channels between 2013 and 2017, in which they identified 177 terms related to specific forms of bullying. For their part, Song et al. (2016) reviewed more than 2.35 billion posts over a period of two years from 163 websites, and they consolidated 99,693 records of minors concerning youth suicide. Han, et al. (2019) studied 2,400 terms in three categories concerning cyberbullying, and Song, et al. (2014) compiled web documents from 227 online channels from which they extracted 435,563 cases related to cyberbullying and identified 118 terms as causes of cyberbullying. Finally, Matsumoto, Yoshida and Kita (2019) have created their own corpus for researching youthful jargon-based tweets when communicating on social networks. They have succeeded in creating lists and vectors of emotional expression of up to 30,000 items.

4.5. Main proposals and contributions

In this section, we have focused on the studies that have been the most influential among the corpus analysed with regard to the sector of the population relevant to this research. Of all the records found, 16 have examined specific samples of adolescents or have referred to them explicitly when preparing and developing their corpus related to the topics and dimensions listed below.

4.5.1. Detection of cyberbullying

By using automatic language identification systems in the context of social networks, Kim, et al. (2019) have pointed out that the problems of online bullying are not limited to schools, but have spread to other environments such as teenage “fandom” communities, which bring together young fans of a hobby or pastime. In this regard, they recommend government systems to control the emerging modes of diverse types of bullying. Han, et al. (2019) have also analysed bullying as a social phenomenon by examining the terms used by minors on social networks. This coincides with the study by Kim et al. (2019), which advocates the creation of data security and privacy protection policies for individuals who are investigated by using SBD.

Song, at al. (2014) studied the risk factors for each type of online bullying with the aim of automatically generating a predictive model to explain the variables that affect this type of behaviour by young people. They also suggest that parental education programmes should be put in place to help children in these situations. Tarwani, Chorasia & Shukla (2017), who have extracted the opinions of internet users on Facebook, MySpace and Twitter using machine learning techniques, offer the possibility of instant detection of bullying messages on social networks and propose a solution to the problem of cyberbullying based on the identification of bullying terms.

For their part, Lara and Giancinto (2017) have worked with statistical models for the detection of offensive expressions, as well as with a system that collects these individual communications. The key lies in creating an application that recognises comments and statements with negative connotations. Its main limitation is the interpretation of the contextual analysis of conversations, which can lead to false positives, especially when language can be sarcastic or vulgar, yet inoffensive, in a given situation.

4.5.2. Sexting as a virtual practice among adolescents

Sexting is an emerging phenomenon that is attracting significant attention for its potential to cause harm to minors (Song & Song, 2015). By applying SBD to the communications of young people under 18 years of age on Twitter, Song, Song and Lee (2018) have concluded that adolescents engage in sexting through the use of smartphone text messaging in the hope of gaining increased attention from friends. They also suggest that expert advice and the promotion of children’s integrity are factors that help to minimise the risk involved in sexting. With regard to the use of SBD, they say that while this methodology is useful and effective, it has challenges such as real-time monitoring and individual data analysis. They consider it essential to conduct further research using this technology with a focus on the risk factors related to this practice.

4.5.3. Suicide risk among young people

This aspect of research using SBD has become particularly relevant in South Korea, as the suicide rate among youth in that country is one of the highest in the world (Song, et al., 2016), and conducting searches of suicide-related terms on Google have been growing steadily since 2007 in that country.

Song et al., (2014) point out that real-time investigation of big data is useful for detecting suicidal tendencies. Specifically, these authors have created a statistical model based on analysing the relationship between the number of Google searches for the keywords stress, drinking, exercise, and suicide, and the detection of adolescent suicide risk. They have determined a correlation between a higher number of searches for the words stress and drinking alcohol and the actual number of suicides, which is significant. They are aware that the use and application of SBD for this type of research is in the early stages, and their advice is that to improve this situation and be able to establish national suicide prevention plans, it will be necessary to create training programmes for professionals in the management of big data, improved infrastructures, and the implementation of cloud computing services.

Song, et al. (2016) have confirmed that adolescents vent their stress, depressive emotions, and suicidal thoughts in cyberspace, and they communicate these thoughts and feelings with other internet users. Among the factors associated with increased activity in searching for words related to teenage suicide, some of the terms that stand out include the following: academic pressure, poor body image, being a victim of bullying, preoccupation with illness, low employment rates, increased rental price index, and intimidation. To reduce the problem, the authors say it is necessary to develop a system for monitoring and responding to real-time searches for suicide-related words on social media via bots and instant messaging.

4.5.4. Adolescent language in social networks

Among the articles that specifically study adolescent conversations that take place in social networks are those of Matsumoto, et al. (2019), and Matsumoto, Yoshida and Kita (2019), who focus on the vocabulary and short expressions that young people use in social networks that do not appear in dictionaries. Their work involves designing models based on groups of documents that include these youthful expressions on Twitter and propose a classification method based on the coding of their characteristics. With this data-driven, slang-clustering model, they have been able to identify 40 terms and short expressions related to the way young people express themselves on social networks. Specifically, Matsumoto, et al. (2019) work with emotional recognition and have created a method they call the “bag of concepts”, the purpose of which is to build an emotional corpus through the collection of weblog texts that include these new words. They point out that most of the words in youthful slang that express the feelings of young people are too ambiguous to be recorded in dictionaries.

4.5.5. Other research dimensions

Other dimensions of research have also been addressed using SBD. According to Jung, Park and Song (2016, 2017), social networks contain an enormous amount of information about children’s feelings, thoughts, interests, and behaviour patterns that can be used to understand their tendency toward depression. In their 2017 research, these authors point out that the ontology they have developed is especially innovative compared to similar studies for three main reasons: Firstly, they have included previously unidentified factors of adolescent depression related not only to individual characteristics, but also to environmental factors such as family, school and community, as well as low academic performance, delinquency and truancy; secondly, they have taken into account risky situations, symptoms, diagnoses, preventive measures, and treatments of adolescent depression; and thirdly, they have also considered slang, buzzwords and neologisms used by adolescents, which makes SBD analysis especially suitable for analysing information generated by adolescents on social networks.

With regard to practices that violate the privacy of teenagers when online, Montgomery (2015) highlights that one of the keys to Facebook’s business model is the fact that it has managed to datify social relationships through the data analysis system known as “The Social Graph”. This has allowed it to study and control all the movements that take place in its network, which is a crucial element in its diverse commercial operations. This omnipresent commercial surveillance is unprecedented in its reach and penetration into the lives of young people. At the same time, this author believes it is necessary to increase public policies in order to protect minors from Facebook’s intrusion into their lives and data, and to implement privacy laws to safeguard against the excesses of BD.

Finally, DeJonckheere, et al., (2019) have integrated several research techniques with regard to young pregnant women: the use of a survey, corresponding text message analysis, and social data mining with natural language processing. Their aim is to identify the beliefs and social norms of these young women in relation to weight gain during pregnancy. The SBD part of the study targets accounts on Facebook, Twitter and Instagram of the sample of selected young women at two precise moments: during their visit to the medical centre in the first trimester, and again in the third. By analysing messages and images posted on social media sites, the researchers studied the beliefs and public statements made by these women with regard to pregnancy, weight gain, body image, and nutrition during the months that the women were pregnant.

5. Conclusions and discussion

This paper has approached publications regarding the online behaviour of children and young people in which techniques linked to SBD have been used. Analysis related to the object of study is constantly growing in terms of the number of academic articles found, publication dates, and volume of citations. By carrying out analyses using big social data, researchers are working to discover the activities of minors online, and to create algorithms to identify risky trends affecting adolescents.

The dimensions that have received the most attention are the recognition of emotions and expressions on social networks, cyberbullying, the monitoring and implementation of educational policies, and the study of unethical practices that interfere with the virtual lives of adolescents, such as virtual manipulation, suicide, and depression among young people. Other issues identified include juvenile delinquency, the perception of teenage pregnancy, sexting, dangerous contacts on social networks, sexual abuse of minors on the Internet, the commercialisation of private data, and the arbitrary way in which social networks invade the privacy of minors.

Specialised authors in the field who are working to exploit data from social networks on issues that directly affect minors or young people have also been documented. By using SBD, scientists can investigate large amounts of real-time data with samples of up to 2.35 billion posts (Song, et al., 2016). They can also create algorithms and automated models to detect children’s behaviour and feelings and identify trends.

What stands out is that only 6.8% (n=4) of the publications have appeared in social science journals, and 13.5% (n=8) were found in mixed journals. The rest were published in technical research journals related to Information Technology, Computer Science, and Engineering, and in medical journals as well. In addition to highlighting the value of perspectives that are more technophile and systematic, which have been provided by the sample under study of a subject as complex as the online behaviour of human beings when faced with multiple screens, an issue still pending is the merger and comprehension of other disciplines such as Communication, Sociology and Education.

In this paper, we have chosen to undertake a search focused on SBD. However, there is some discrepancy in the naming of studies and techniques related to this field. Most authors title their research and include the term SBD or BD in their keywords when they investigate social networks, and they use research techniques such as data mining, machine learning, semantic analysis, and sentiment analysis. However, they do so without actually stating that they are applying a specific SBD or BD methodology.

We expect to see an increase in the number of studies that address the virtual lives of minors through the use of these research techniques, as there is still a gap between their potential and their actual use. Of course, SBD is only one approach (Malvicino & Yoguel, 2014), and data alone is not a magic solution for analysing the social reality. One should not fall into quantitative reductionism (Sánchez-Bayón, 2017) and abandon cultural aspects, among others, when investigating social movements. Researchers must approach big data with caution to ensure social progress (Kosinski & Behrend, 2017), and there is no doubt that they will continue to be necessary in order to develop theories and provide the legitimacy that science needs as a social entity that sets standards (Sætra, 2018).

This research has two limitations. On the one hand, it does not incorporate all possible databases that can be accessed to locate scholarly literature in this field. On the other hand, a tag-based search strategy may at times fail to include some aspects that could be of interest.

6. Funding

This study has been funded by the programme entitled, “New scenarios of digital vulnerability: media literacy for an inclusive society” (PROVULDIG-2-CM) (ref. H2019/HUM5775), CAM and the European Social Fund, and by the National Project of the R&D&I programme, “Social networks, adolescents and young people: media convergence and digital culture” (CSO2016-74980-C2-2-R).