This study employed a descriptive research design to identify the main concepts and research topics in KJWHN, using quantitative content analysis of text networks and topic modeling.

The study targeted manuscripts with English abstracts among the 373 papers published in KJWHN from January 2011 to December 2021. The research process involved: (1) data collection, (2) word extraction and refinement, (3) keyword extraction and network creation, (4) network centrality analysis and key topic selection, and (5) topic modeling analysis.

Data for this study were collected in February 2022 to identify research trends in KJWHN over the past decade, from January 1, 2011 to December 31, 2021. Full-text articles were accessed from the journal website (https://kjwhn.org/), and published papers with English abstracts were identified. Out of the 379 papers published during this period, 373 were analyzed, excluding six papers that did not have English abstracts available. The analyzed papers were organized by unique number, title, author, year of publication, abstract, and keywords using the MS Office Excel program (Microsoft, Redmond, WA, USA). Typographical errors and English spelling were checked using the spell-check function of the Excel program. Additionally, words such as “background,” “objectives,” “purpose,” “methods,” “results,” and “conclusion,” which are frequently used in standard abstracts, were removed.

Our research team utilized NetMiner (version 4.3; Cyram, Seongnam, Korea) to conduct a semantic network analysis, extracting keywords from the titles, English abstracts, and keywords of articles. The keywords were derived from both the title and abstract, not solely from the author’s keywords. In order to extract and refine these keywords, the researchers repeatedly read and discussed words and abbreviations with the same meaning, refining them into a single word and unifying them with similar words. As a result, 56 similar words were identified. For instance, “breastfeeding” and “breast feeding,” which can differ in spacing, were unified as “breastfeeding.” Likewise, “self-efficacy” and “self efficacy” were unified as “self-efficacy.” When a combination of uppercase letters, lowercase letters, noun phrases, and abbreviations was used, a representative word was designated. For example, “quality of life,” “qol,” and “QOL” were designated as “QOL” to avoid redundancy.

Next, stop words such as pronouns and numbers were excluded using the automatic filtering function within the NetMiner program. General verbs and auxiliary verbs, such as “do,” “make,” “use,” “would,” “could,” and “should,” were excluded through discussion among researchers. Unnecessary nouns and other parts of speech, including “one,” “two,” “participants,” “subjects,” “level,” “group,” “data,” “research,” “test,” “year,” “design,” “day,” “example,” “effect,” “as,” “without,” “though,” and “all,” which were irrelevant for analyzing trends in women’s health nursing research, were also excluded. Additionally, words indicating logical relationships such as “only,” “before,” “toward,” “to,” and “with,” as well as adverbs and prepositions, were excluded. Consequently, a list of synonyms and negative words, which the researchers had discussed and agreed upon, was entered into the NetMiner dictionary. Among the words extracted from the 373 English abstracts, those with a frequency of 10 or more occurrences were selected as keywords. From this group, the top 25 words with the highest frequency of simple occurrences were chosen as keywords.

Keyword network analysis identifies various characteristics by extracting significant words from the text, determining the connections between them, and reorganizing them into a visual network [12]. A network was created based on the co-occurrence relationships between words, using the total frequency (weight) of word pair co-occurrences. The window size was set to three, and the link frequency threshold was set to two in order to extract all relationships that appeared more than once. The direction was set to nondirectional, allowing for the formation of a network between keywords regardless of the order in which they appeared. Additionally, the “remove self-loop” option was set to “yes” to exclude identical keyword relationships.

Our research team performed a centrality analysis to assess the impact of particular keywords on the entire network. This analysis focused on refined keywords and employed a mediation centrality approach to ascertain their positions within the network [13]. Furthermore, we conducted a word cloud analysis [14] to provide a visual representation of significant keywords at a glance. The word cloud was generated by considering documents with a term frequency-inverse document frequency (TF-IDF) of 0.1 or higher, specifically targeting refined keywords.

Topics were extracted using the latent Dirichlet allocation (LDA) topic modeling technique after text preprocessing [12]. LDA is the most widely employed document generation model in text mining analysis [9]. To obtain meaningful results through topic modeling, the number of topics is crucial and must be determined by the researcher [9]. Consequently, this study focused on 4 to 8 topics, and multiple analyses were conducted. Furthermore, after setting the alpha (α) value to 0.1, the beta (β) value to 0.01, and the number of sampling repetitions to 1,000, the number of topics was compared and analyzed.

In total, 3,324 words were extracted and refined from the 373 article abstracts. The simple appearance frequencies of the extracted words were as follows, in descending order: “women” (1,001 instances), “health” (424 instances), “nursing” (322 instances), “care” (259 instances), “cancer” (254 instances), “birth” (240 instances), “education” (233 instances), “depression” (228 instances), “life” (223 instances), “pregnancy” (200 instances), “mother” (198 instances), “HPV” (195 instances), “stress” (186 instances), “quality” (184 instances), and “nurse” (179 instances) (Figure 1).

A network connection structure consisting of 3,425 nodes and 12,589 links was confirmed by examining the relationships between words. The density of the analyzed network was 0.015, the average connection degree was 9.054, and the average connection distance was 4.587 (Figure 2). Next, according to the TF-IDF analysis, the top keywords with high importance were “women” (198 instances), “health” (197 instances), “nursing” (189 instances), and “care” (178 instances). “education” (was 178 instances), and “life” (was 171 instances) (Table 1). Degree and betweenness centrality were analyzed. The keywords with high degree centrality were “women,” “health,” “care,” “nursing,” and “pregnancy,” and those with high betweenness centrality were “women,” “intervention,” “stress,” “experience,” and “nursing” (Table 1). A word cloud analysis presenting these high-importance keywords is shown in Figure 3.

Six topics were identified and are presented in Table 2, along with their associated keywords and high allocation probabilities. Two researchers discussed and named the topics, taking into account the main keywords and connectivity for each topic extracted through topic modeling analysis. The six topics were named as follows: “gynecologic neoplasms,” “‘menopausal health,” “health behavior,” “infertility,” “women’s health in transition,” and “nursing education for women.” The proportions of each topic among the research papers were relatively even. Oncology-related research accounted for the highest proportion (33.3%), and education-related research exhibited the lowest proportion (10.7%) (Table 2). For the first topic (gynecologic neoplasms), the keywords were “permission,” “instrument,” “cm,” “tumor,” “modality,” “breast,” “survivor,” “education,” “mastectomy,” “indicator,” “chemotherapy,” “radiation,” “side effect,” “phase,” and “intrauterine.” The keywords that constituted the second topic (menopausal health) were “bone,” “obesity,” “mineral,” “height,” “menopause,” “dietary,” “cross,” “glucose,” “GDM,” “regularity,” “QOL,” “body composition,” “food,” “intake,” and “energy.” The keywords in the third topic (health behavior) were: “hygiene,” “drinking,” “smoking,” “exercise,” “diet,” “worker,” “hypertension,” “education,” “physical activities,” “water,” “incontinence,” “prevention,” “evaluation,” “women,” and “constipation.” The fourth topic (infertility) included the following keywords: “in vitro,” “IVF,” “failure,” “fact,” “laparoscopic,” “stigma,” “challenge,” “low,” “definition,” “format,” “choice,” “platform,” “welfare,” “decade,” and “primi.” The keywords found within the fifth topic were “recovery,” “translation,” “movement,” “caring,” “QOL,” “rest,” “nursing,” “orientation,” “behavioral,” “sexuality,” “functioning,” “reaction,” “medicine,” “midwifery,” and “flexibility,” and this topic was named women’s health in transition. Finally, keywords that constituted the sixth topic (nursing education for women) were “accuracy,” “education,” “educator,” “simulator,” “high fidelity,” “scenario,” “debriefing,” “childbearing,” “measuring,” “encouragement,” “context,” “professor,” and “facilitator.”