An ontology for describing the determinants of fertility, called the determinants-of-fertility ontology, was developed based on the Ontology Development 101 methodology [22] in 5 steps, as described below.

The available methods for evaluating the quality of an ontology include those proposed by Brank et al [34], Obrst et al [35], and Vrandečić [36]. Brank et al [34] classified ontology evaluations into 4 categories: (1) approaches that compare the target ontology to a gold standard (gold standard), (2) approaches that use the target ontology in an application (application-based), (3) approaches that compare the target ontology with a source of data (data-driven), and (4) approaches that assess the target ontology by human (user-based). Since there is no gold standard available, the determinants-of-fertility ontology was evaluated using the remaining 3 approaches proposed by Brank et al [34] (ie, application-based, data-driven, and user-based evaluations). The ontology was revised based on the evaluation results.

The ontology with terminology was used to detect future signals of fertility issues from social media data. Future signals were analyzed based on the text-mining–based weak-signal detection method of Yoon [12], as follows, using R software (version 3.5.2).

A list of 10 CQs was compiled (Textbox 2) that reflect different levels of factors affecting fertility using an ecological model. The domain and scope of the ontology were determined based on CQs for the determinants of fertility related to the individual, family, workplace, childcare and educational environment, socioeconomic and cultural environment, and public policy. Childbirth before marriage is very uncommon in South Korea, and so delayed marriage is an important factor affecting childbirth decision-making [3,4,27,42,43]. Therefore, the scope of fertility determinants included factors related to marriage delay and childbirth decision-making.

In total, 1659 terms covering the domain and scope of the ontology were collected, and 236 unique class concepts were extracted from these terms. We defined hierarchical and attribute relationships of the classes based on the ecological model. The determinants of fertility were organized into the following levels: individual, family, workplace, childcare and educational environment, socioeconomic and cultural environment, and public policy. These 6 levels of the ontology were defined by adding not only the workplace, but also childcare and educational environment to institutional factors, which constitute the third level of the ecological model. Due to the increasing participation of women in the labor market, the workplace and childcare and educational environment are important factors influencing decisions about childbirth among women who are working [3,4,27,32,42,43]. Employed women in South Korea experience difficulties combining working with childbirth and childcare and are often faced with choosing between giving up their jobs to give birth and look after their child or continuing to work, thereby forsaking their desire to give birth and look after a child [32,42,43]. Therefore, we viewed the workplace and the childcare and educational environment as important social institutional factors and determinants of fertility.

Figure 1 shows the determinants-of-fertility ontology with classes up to the second level. The ontology consists of 6 superclasses: individual, family, workplace, childcare and educational environment, socioeconomic and cultural environment, and public policy. The individual superclass has the subclasses of women’s sociodemographic data, which include age, education, employment, and religion; reproductive health, which includes sexual behavior, contraceptive use, and childbirth; and the individual’s attitude, which includes their attitude toward children and marriage. The family superclass has the subclasses of the family’s sociodemographic data, which include family size, family income, family expenditure, sex and age of children, number of children, spouse’s age, and spouse’s income; the family member’s relationship, which includes gender equality, couple’s intimacy, and family’s life satisfaction; and family-formation factors, which include marriage cost. The workplace superclass has the subclasses of workplace structure, which includes workplace type, working hours, and employment insurance, and workplace culture, which includes parental leave availability. The childcare and educational environment superclass has the subclasses of the childcare and educational structure, which includes type, cost, service, and human resources, and childcare and educational environment satisfaction, which includes the belief of the quality. The socioeconomic and cultural environment superclass has subclasses of the sociocultural environment, which includes social value, mass media, and social change, and the economic environment, which includes economic growth. Finally, the public policy superclass has the subclass of policy on low fertility, which includes the area, legal basis, and tasks. These classes had 3 or 4 levels of hierarchy with 230 classes and 41 relationships.

We developed EAV models for the 139 lowest level class concepts. For example, contraceptive use had attributes of hasType and hasLengthOfUse, where the values of the hasType attribute were condoms, contraceptive injection, contraceptive patch, pill, intrauterine device, diaphragm, and sterilization and hasLengthOfUse had the value of number of months. We also developed a terminology linking synonyms for classes, attributes, and values: 90 synonyms for 236 classes, 9 synonyms for 54 attributes, and 501 synonyms for 772 values.

Table 3 lists the results for the computed DoD for each keyword for the socioeconomic and cultural factors and fertility policies. Violence and abuse (socioeconomic and cultural factor) and flexible working arrangement (fertility policy) had a low DF and high DoD growth rates. Economic problems, population aging, and nuclearization of the family (socioeconomic and cultural factors) and child-safety protection system (fertility policy) had a high DF and high DoD growth rates.

Figure 2 shows the KIM. The weak signals are marked with red rectangle (area A) in the second quadrant of the KIM. The strong signals are marked with blue rectangle (area B) in the first quadrant of the KIM. Table 4 presents the signal classification of keywords for socioeconomic and cultural factors and fertility policies in the KIM. The keywords classified as weak signals were violence and abuse (socioeconomic and cultural factor) and flexible working arrangement and financial support for employment security (fertility policies). The keywords classified as strong signals were economic problems, nuclearization of the family, and population aging (socioeconomic and cultural factors), and child-safety protection system and maternity-leave system (fertility policies). The keywords classified as latent signals were gender inequality, employment problems, and changing perspectives about marriage (socioeconomic and cultural factors), and policy public relations, family-friendly work environment, and smart work center (fertility policies). The keywords classified as not-strong-but-well-known signals were conservative values (socioeconomic and cultural factor) and financial support for childbirth and infrastructure for childcare support (fertility policies).

We have developed a determinants-of-fertility ontology as a framework for collecting and analyzing social media data. The ontology was evaluated in terms of the DL, content coverage, and structural and representational layers. We applied the ontology with terminology to detect future signals of fertility issues from social media data.

The developed determinants-of-fertility ontology has 6 main characteristics. First, it is the first ontology to describe the multilevel factors that affect fertility. Various factors and the complex interactions between them determine fertility [24,26,27], and expressing the fertility determinants requires an integrated view of these factors. We therefore developed the ontology by classifying the various environmental factors related to the individuals and in terms of family, childcare, workplace, and community levels from an ecological perspective.

Second, this ontology contains factors related to fertility issues that are unique to South Korea. In most cases, childbirth does not occur until after marriage in South Korea, and delaying marriage is an important factor affecting the decision to have a child [27,42,43]. Therefore, the scope of the ontology spanned individual and social factors that influence not only delayed marriage but also the decision to have children after marriage. For example, the ontology includes concepts of attitude toward marriage in the individual superclass, matters related to family formation in the family superclass, and support for starting a family in the public policy superclass.

Third, the developed ontology includes terminology with synonyms for classes such as consumer terms and abbreviations, which makes it suitable for analyzing social media data. For example, regarding financial support for childbirth, various terms such as childbirth celebration money, childbirth grant, subsidy, baby bonus, and childbirth incentive are used on social media postings. Since the developed ontology includes these terms, it can be used to collect and analyze consumer terms in social media data.

Fourth, each class of this ontology was modeled using the EAV model and included the attributes of each class and the values of those attributes. Like previous research [20,21], value sets of attributes representing the level and status of an entity included terms describing the keyword in detail in social media data. This novel characteristic of our ontology renders it capable of advanced keyword extraction and suitable for analyzing social media data.

Fifth, we ensured quality of the ontology by using a variety of evaluation methods, including the application-based, data-driven, and user-based approaches proposed by Brank et al [34]. Application-based evaluation was performed by testing the ability of the ontology to answer the CQs that cover its domain and scope. The ontology provided correct answers to all of the CQs without any errors using DL. Data-driven evaluation was performed by testing the content coverage of the ontology whereby terms extracted from social media posts on fertility issues were compared with terms included in the ontology. Medical terms related to infertility and pregnancy that appeared on social media were not included in the ontology. As a result, medical terms on social media were added to the ontology. User-based evaluation was performed by asking the experts to rate the structural consistency, errors in class relationships, and representational quality of the ontology using the evaluation criteria of the structural and representational layers [39]. The experts rated all the evaluation criteria at between 3 and 5 on a 5-point scale.

Finally, the ontology with terminology developed in this study was used as a framework to detect future signals of fertility issues from social media data. The ontology allowed us to use social media data to identify the current trends and future changes in fertility issues related to the effective implementation of policies to increase the fertility rate. These trends were economic problems, child-safety protection system, violence and abuse, and flexible working arrangement.

Economic problems and child-safety protection system were strong signals of fertility issues. Economic problems were noted to be an important topic in a survey about public perceptions of the low fertility phenomenon [44]. In a study by the Korean Ministry of Health and Welfare involving 2000 adults that examined the perceptions of low fertility and population aging, the participants reported that the main causes of low fertility were economic problems such as the economic burden of child support and education costs (60.2%) and employment instability (23.9%) [44]. Therefore, there is a need for governments to provide continuous support measures to reduce economic difficulties in family formation, childbirth, and parenting.

A child-safety protection system needs to be implemented as another investment toward avoiding low fertility in future generations [45]. In order to protect the health and safety of children, the government needs to actively respond to risk factors that threaten child safety such as child abuse and school violence [3,4,45-47]. Policies on child safety have been one of the most important national issues since 2013. The Korean government that came into power in 2013 has focused on protecting vulnerable groups such as children, adolescents, and women by introducing policies to deter sexual abuse, school violence, and domestic violence [46,47]. Since the government promoted these policies, both the mainstream media and social media treated them as big issues [46,47]. However, only 13.1% of married couples reportedly felt that their children were growing up in a safe and healthy environment [48], and researchers have continued to point out that the weak child-protection system remains a problem in South Korea [45,48,49]. The government therefore needs to focus more on solving the limitations and problems associated with the child-safety protection system.

Violence and abuse and flexible working arrangement were weak signals that could develop into strong signals in the future. Violence and abuse were previously reported as important factors for low fertility, with experts identifying changes in the social environment as being important [45,49,50]. The proportion of children (younger than 18 years) in the total population has decreased from 19.6% in 2013 to 15% in 2019 due to the low fertility rate, while the reported number of child-abuse cases has increased from 10,943 in 2012 to 36,417 in 2018 [50]. South Korea is still a highly patriarchal and authoritarian society, and a child is viewed as a subject to be parented and disciplined rather than as one with its own rights [49,50]. Thus, it is necessary to change the perception of violence and abuse toward children as social problems rather than family problems. The government needs to continually encourage policies that not only improve the fertility rate but also improve the environment that currently threatens the safety and health of babies and children.

A flexible working arrangement is a policy that allows workers to balance work and life [4,51]. Both the OECD and the Korean Ministry of Health and Welfare promote this as an important policy for increasing the future fertility rate in South Korea [52]. Long working hours, long commuting hours, and the culture of socializing after work in South Korea make it difficult to maintain an optimal balance between work and family [52]. Therefore, solving the low fertility problem in this country will require the government to promote policies enabling workers to have flexible working hours.

The determinants-of-fertility ontology developed in this study comprehensively covers fertility issues relevant to the low fertility phenomenon in South Korea and will be useful for analyzing social media data. However, it is also subject to several limitations.

First, the direct and indirect effects of employment stability, job creation, housing supply, and public education on fertility [53] were not included in the public policy category of the ontology because the Korean government applies these policies separately from the policies for low fertility. A comprehensive approach covering relevant policies is needed to effectively address low fertility. Future fertility research should include policy strategies that address issues related to employment, job creation, housing supply, and public education.

Second, the synonyms of the ontology developed in this study may not include all the terms used by the general public on social media. Many of the terms used on social media are highly transient—rapidly appearing, spreading, and then disappearing [54,55]. The terms included in the ontology therefore need to be updated continuously based on those currently used by the general public.

Third, future signals of fertility issues were detected during the second phases of the policy on low fertility. Low fertility is a demographic issue that requires a long-term approach, and the policy responses of the government should be periodically reviewed and evaluated to ensure that the policies in place at a particular time point are consistent with any changes in the population and the socioeconomic and cultural environment [56]. This situation indicates the need to analyze social media data periodically using the ontology developed in this study to establish the policy direction for addressing the low fertility phenomenon by reflecting current and future trends on fertility issues as accurately and timely as possible.

Finally, only the KIM that uses the DF of keywords was used to detect future signals. Since future signals are generally subjective [10], even a careful analysis of future signals will not always yield accurate results [57]. Yoon [12] proposed a quantitative method using 2 types of keyword portfolio maps—a KIM using the DF and a keyword emergence map using the keyword frequency—to accurately detect future signals. In contrast, Lee and Park [57] proposed using both quantitative and qualitative methods to ensure the accurate detection of future signals. Therefore, we suggest performing a study of the ontology-based detection of future signals of fertility issues in South Korea that employs a quantitative method involving 2 types of keyword portfolio maps and a qualitative method involving experts.

A determinants-of-fertility ontology was developed in this study that comprised 6 superclasses, 230 subclasses, and 41 relationships with terminology that comprised 1464 synonyms for the 236 classes. Class concepts of the ontology were included as an EAV model and contained synonyms of the ontology classes such as consumer terms and abbreviations. The ontology can be used to analyze social media data on fertility issues. The DL, content coverage, and structural and representational layers of the ontology were evaluated. The ontology and its terminology were used to detect future signals of fertility issues in South Korea. Our novel determinants-of-fertility ontology provides a framework for collecting and analyzing social media data toward understanding which socioeconomic and cultural factors and fertility policies should be focused on in the future. The analysis of future signals revealed that violence and abuse (socioeconomic and cultural factor) and flexible working arrangement (fertility policy) were weak signals that might increase rapidly in the future. The findings of this study will help policy makers to develop effective policies for responding to the low fertility rate in South Korea based on examinations of the present and future trends.