This poster analyzes the cognitive structures of Korean rural intellectuals from the late nineteenth to the early twentieth century through small-data-driven digital humanities methods. While previous scholarship on Korean modernity has primarily focused on state policies and elite discourse, this project foregrounds engagement with everyday experiential writing. By examining personal diaries written by non-elite male rural intellectuals, the study contributes to the theme of "Engagement" by analytically engaging with an underrepresented textual community. Although diaries have been discussed in Korean and English scholarship, their underlying cognitive structures, reconstructed through computational methods, remain largely underexplored. While recent Korean digital humanities research has begun applying network analysis to macro-historical concepts or elite kinship structures, applying lightweight computational semantic models to the everyday writings of non-elite rural intellectuals remains a novel frontier.

The dataset comprises two long-term Classical Chinese diaries: Sim Won-kwon's Diary (1870–1933) and the Gwallanjae Diary (1912–1948). Written without an external audience in mind, these diaries record the authors' daily practices, emotional reactions, and worldviews with immediacy. Although small in size compared to macro-scale colonial archives, their high density of repeated concepts makes them ideal for interpretive research. Sim's diary documents over sixty years of rural life, focusing on agriculture, weather, and local relations, revealing an empirical worldview. Chŏng's diary shares many thematic concerns but contains far more explicit reflections on colonial policy and social change, documenting emotional disorientation in the face of rapid modernization.

The central research question is how everyday experiential concerns were structured and reorganized in the language of rural intellectuals under late nineteenth- and early twentieth-century transformations. The analysis proceeds in three stages. First, drawing on my historical expertise, I developed a manual coding schema to classify daily diary entries into eight conceptual topics (e.g., Routine, Econ, Reflection) and distinct textual genres. This manually coded metadata is explicitly separated from the raw text to map conceptual domains accurately. Second, to reduce noise without over-complicating the model, I rely on raw word frequency after strictly applying a custom stopword list to filter out grammatical particles (e.g., 矣, 也) and simple movement verbs (e.g., 往, 來).

Third, following the semantic neighborhood approach developed by Gershkovich et al. (2025), the study examines the collocational environments of these keywords. Frequently co-occurring terms within a single day's entry are compared across the diary, enabling the reconstruction of recurring conceptual groupings. To validate these computational clusters, I cross-reference the automated outputs against my manually coded topic metadata. Diachronically, clear shifts emerge: In Sim's early entries, semantic neighborhoods concentrate on weather, bodily sensation, and routine movement, while later entries increasingly foreground agricultural labor and the quantified recording of rice prices. In contrast, the Gwallanjae Diary embeds economic terms within socially and ethically charged vocabularies, where moral evaluation and self-cultivation emerge as central hubs.

This study reveals micro-level cognitive shifts that remain difficult to observe through macro-scale approaches. Furthermore, it suggests how small but high-density historical datasets can inform AI language model evaluations and development. Contemporary LLMs trained on modern, large-scale corpora often struggle to capture the complex, culturally specific semantic networks of historical non-Western rural communities. Consequently, the dense, manually validated semantic networks extracted in this study can serve as highly specialized evaluation benchmarks to test the cultural and historical alignment of AI models, or as targeted fine-tuning data to mitigate Western-centric biases. By combining computational reading with expert qualitative coding, this project underscores the importance of studying the everyday writings of marginalized rural intellectuals to truly understand the lived experience of Korean modernity. Future work will expand this framework to additional rural diarists to map broader cognitive shifts across the region.