The analysis data to be used in this study is divided into two types. The first type consists of academic papers dealing with the topic of life satisfaction. In this study, the abstracts of papers from academic journals managed by the publisher Springer that contain the term ‘life satisfaction’ in their titles are extracted and analyzed. The reason for choosing journals managed by Springer as the subject of analysis is twofold: first, these journals include a significant number of publications related to the topic of 'life satisfaction,' which is the main focus of this study; and second, there is an established license agreement between the big data analysis tool Netminer 4.5, which is being used in this research, and Springer's journals, making data utilization more convenient.

A total of 1,000 papers were selected, based on relevance. Here, relevance is first determined by the frequency with which a keyword or phrase appears within a document. That is, papers in which the keyword 'life satisfaction' appears more frequently may be considered more relevant. Secondly, suppose the keyword appears in both the title and abstract. In that case, greater importance is assigned, under the assumption that these sections succinctly represent the core content and focus of the document.

The second type consists of national survey data collected in 2022 by the Korea Institute of Public Administration (KIPA) (KOSIS; https://kosis.kr/index/index.do; KIPA; https://www.kipa.re.kr/site/kipa/main.do). These data stem from the results of a survey of social integration status that aims to comprehensively investigate the living conditions of Korean citizens. The total number of respondents included in the survey was 8,077.

First, we derive the multiple factors that affect life satisfaction through analysis of the academic papers. After organizing and condensing the factors derived in this way, we draw on the data from the survey questionnaire produced by KIPA in 2022 to conduct statistical analysis.

The first research aim, analysis to extract the factors affecting life satisfaction, utilizes a topic modelling method. Topic modelling is one of the machine learning analysis methods used for discovering and classifying hidden topic patterns in big data, and it is also a method used to identify and classify topics within text data (Rego Rodríguez; Germán Flores; Vitón-Castillo, 2022; Choi; Mohamed, 2023 a , 2023b; Larrosa; Galgano; Gutiérrez, 2023; Panduro, 2023). Here, it is performed on the abstracts of academic papers with ‘life satisfaction’ in the title. The analysis of the second research objective, the relative importance of the factors affecting Koreans' life satisfaction, will be conducted using the neural network analysis method. Through this analysis, the importance weights of the variables affecting life satisfaction will be calculated, and which variables are particularly important will be identified. There are various methods for such analysis, but in this study, the multi-layer perceptron (MLP) method, one of the neural network analysis methods, will be applied. Originally, neural network analysis, especially with large sample sizes, demonstrates excellent predictive power through the application of parallel processing methods, and does not require statistical assumptions necessary for model derivation, making it suitable for studies like this one. Additionally, the MLP method is chosen among neural network analyses because it is the most fundamental algorithm that secures high predictive power in classification and estimation without the need for assumptions such as the independence of independent variables (Lee, 2003; Cho, 2018; Villa-Soto, 2022). Therefore, we analyze KIPA’s survey data using variables related to the derived influencing factors.

Table 1 below presents information on the frequency of words derived through basic data analysis. As Table 1 indicates, ‘life satisfaction’ had the highest frequency, at 4,038 occurrences. However, because life satisfaction is connected to almost all the other words, it will be excluded from the subsequent analysis process.

Table 1 -

	Part of Speech(POS)	Frequency	Word length	Name Type	Author Keyword
life satisfaction	Proper Noun	4038	17	-	False
relationship	Common Noun	1010	12	-	False
effect	Common Noun	927	6	-	False
life	Common Noun	828	4	-	False
health	Common Noun	755	6	-	False
satisfaction	Common Noun	741	12	-	False
well being	Common Noun	650	10	-	False
level	Common Noun	591	5	-	False
adult	Common Noun	482	5	-	False
model	Common Noun	481	5	-	False
association	Common Noun	477	11	-	False
factor	Common Noun	468	6	-	False
student	Common Noun	461	7	-	False
research	Common Noun	450	8	-	False
LS	Common Noun	446	2	Organization Name	False
datum	Common Noun	437	5	-	False
adolescent	Common Noun	431	10	-	False
role	Common Noun	424	4	-	False
age	Common Noun	424	3	-	False
group	Common Noun	418	5	-	False
finding	Common Noun	414	7	-	False
country	Common Noun	387	7	-	False
support	Common Noun	375	7	-	False
individual	Common Noun	369	10	-	False

Note: The table above exemplarily presents words included up to the 25th rank by word frequency.

Before conducting topic modelling on the factors influencing life satisfaction, we sought first to create an ego network of ‘influencing factors’. Ego networks are a form of social network analysis that represents the relationships between a specific individual (ego) and other individuals (alters) directly connected to that individual (Lee, 2003; Cho, 2018; Amaral; Araújo; Moraes, 2022; Tiwari et al., 2023). Ego networks are used to understand the structure and characteristics of an individual’s social relationships. Since the network for the data to be analyzed here was too large, we first constructed an ego network for ‘influencing factors’ and performed topic modelling around this. As a result, this network included a total of 762 words.

The next step was to perform topic modelling. The Figure 1 below shows nine topics derived through topic modeling analysis.

Figure 1 -

The nine topics derived from Figure 1 include words that overlap with each other. To prevent this redundancy and condense further, Pathfinder net (PFnet) analysis and clustering analysis was performed.

As a result of clustering the factors affecting life satisfaction, they are divided into groups, as follows:

G1: loneliness, interaction, health, community, disability, quality, patient.

As the above list shows, in comprehensively analyzing the characteristics of words composed by group, G1 is the degree of loneliness, G2 that of self-esteem, G3 that of awareness of income and inequality, G4 that of satisfaction with work, G5 that of self-esteem, G6 for degree of family relationships, and G7 for the degree of anxiety and stress.

In the topic modelling, we derived the influencing factors that theoretically affect life satisfaction. The next step is to analyze whether these actually function as important influencing factors in people’s lives. The Government of the Republic of Korea conducts an annual survey of the lives of Korean citizens. Surveys of people’s social capital, levels of trust and satisfaction with life are being conducted by the Korea Institute of Public Administration (KIPA), under the overall name Social Integration Survey (respondents 8,077).

In this study, we use survey data of the 2022 Social Integration Survey. Meanwhile, in this study, in addition to the seven influencing factors derived through topic modelling, we will include influencing variables related to personal background (gender, age, education level) and compare and analyze their importance.

Looking at the basic statistics of the variables used in this study, as regards gender, there are 4,004 men (49.6%) and 4,073 women (50.4%), roughly similar numbers. When residents are divided into urban and rural (urban_rural_2), the former comprises 6,711 people (83.1%) and the latter 1,366 people (16.9%). The life satisfaction variable is a continuous variable made up of 0-10 points, but to divide it into a dichotomous form 0-5 points are categorized as 1 (low) and 6-10 points as 2 (high). Life satisfaction was used as a dependent variable. According to this classification, Type 1, with ‘low’ life satisfaction, comprises 2,230 people (27.6%) and Type 2, with ‘high’ life satisfaction, 5,847 people (72.4%).

4.3.2 Analysis of the relative importance of the factors

At the neural network analysis stage, the input layer comprised of gender (sex_2), family relationship (family_2) and urban and rural (urban_rural_2) variables, and the covariates were anxiety, work, esteem, inequality, loneliness, support, age, and health. Income and education variables were also included. The number of nodes in the input layer was 20, and the standardization method was used as a covariate adjustment method. The number of hidden layers was set to 1, the number of nodes in hidden layer 1 was 7, and the activation function was set to the hyperbolic tangent method. Meanwhile, in the case of the output layer, the dependent variable used a binary value dividing ‘life satisfaction’ into ‘low’ and ‘high’ (life_2_1); the number of nodes is 2 and the activation function is the soft max method.

Figure 2 depicts the neural network derived from the neural network analysis. In the diagram, thick lines indicate strong connections. This means that the weight of the connection is large, indicating that it has a significant influence on the neural network’s decisions. A larger weight means that the signal transmitted through that connection has a stronger effect on the next neuron. Additionally, a thick line indicates that the corresponding path is frequently activated. This means that the connection was used frequently during the neural network learning process, suggesting that the feature plays a key role in the output.

Figure 2 -

In light of the parameter estimates presented in Table 1, the neural network analysis results from this study can be judged good. However, when interpreting the parameter estimates of a neural network, it is important to understand how these values interact in the overall network. Also, evaluating the performance of the entire model with only a single weight or bias value can be misleading, so it is important to interpret it in the context of the entire network.

Table 2 shows the relative importance of the independent variables used to analyze influencing factors on life satisfaction. In neural network analysis, the importance of each independent variable (feature) indicates how much it affects the model’s prediction. Understanding the importance of each feature in complex machine learning models such as neural networks helps to interpret the model’s decisions and create more effective feature selection and engineering. As Table 2 shows, satisfaction with one’s work (work_2) appeared to have the greatest impact on life satisfaction. The next most influential variables were self-esteem (esteem_2), the absence of worry or anxiety (anxiety_2) and satisfaction with health (health_2). What is unusual here is that there is a very large gap in influence between satisfaction with one’s work, which is the variable with the greatest influence (work_2), and other variables. In other words, it can be seen that satisfaction with one’s work has a decisive influence on life satisfaction.

Table 2 -

sex_2	.012	3.1%
family_2	.052	13.6%
urban rural_2	.015	4.1%
anxiety2	.085	22.4%
work_2	.380	100.0%
esteem_2	.188	49.5%
inequality_2	.055	14.6%
loneliness_2	.044	11.6%
support_2	.029	7.7%
age_2	.018	4.9%
health_2	.078	20.4%
income_2	.024	6.4%
education_2	.018	4.9%

Figure 3 represents graphically the importance of the independent variables that affect life satisfaction. It can be seen that satisfaction with work (work_2), the variable with the greatest influence, has an overwhelmingly large impact.

Figure 3 -