Daejeon, July 27–31
This paper focuses on the dynamics of authors' reputations, particularly long-term reappraisal after death. Posthumous fame is a popular concern in literary history (e.g., Hasumi 1988), often driven by obituaries or excavation of manuscripts. While critics claim the periodicity of reappraisal, ranging from 10, 30 to 60-year cycles (Nakamura 1963) (Karatani 1990) (Sayawaka 2014), quantitative verification has been difficult. However, digital archives for commons started to feed variable meta-data, now we could develop useful small data, and try applicable analysis.
The rapid and widespread adoption of generative AI has made it easier to explore and discover historical and classical works and led to substantial increases of work within an extremely short time. As the cycle of creation and reception accelerates, the dynamism of discourse surrounding literary works will undergo further change. Especially, the lifespan of intellectual property (IP) would be a common concern not only in digital humanities but also within the creative industries.
Then we sampled early modern writers in the Catalogue of the Collection of Japanese Classic Literature, leveraged the cross-sectional digital archive "Japan Search", created small datasets which are the volume of searchable materials over time, and applied time-series analysis.
Posthumous fame means the increase in the reputation of leaders, celebrities, or artists after their death (Allison / Eylon 2005) (Jones / Jensen 2005). The Death Effect denotes the price premium that arises for their death (Ekelund et al. 2000). Research focuses on prices and supplies exist for celebrities (e.g. painters, singers, athletes, heads of state) and products (e.g. artworks, memorabilia, toys), but studies examining over centuries are scarce, and seldom exist for writers.
Previous research revealed that death triggers the nostalgia effect (Matheson / Baade, 2004), increased media coverage (Steffens et al. 2017), and temporary declines in exhibitions, and price increases (Etro / Stepanova 2015) (Sahli / Cuntz 2025). The death effect occurs only to celebrities (Frick / Knebel 2007) and is also influenced by their age at death, and the accidentality of it (Maddison / Jul Pedersen, 2008) (Itaya / Ursprung 2016).
The diverse data had been used: bibliographic data such as Google Ngram (Braake / Fokkens 2015), Myspace comments (Brubaker / Hayes 2011), reactions to obituaries on Twitter (West et al. 2021), observational studies of online fandom (Yu 2021), and transaction data from the second-hand LEGO market (Pecchioli et al. 2025), and television, web, newspaper, and radio GRP (Gross Rating Point) (Mikuláš / Nemcova Tejkalova 2024). There is a large-scale discourse analysis using 38 million obituary articles from 1998 to 2024 (Markowitz et al. 2025). However, no studies dealing with fiction writers were found, apart from some case studies (Matthews 2004) (Zemanek 2010) (Chaghafi 2020). And, since these data have often restrictive conditions, it was hard to conduct long-term analyses rapidly with ease.
This paper sets three research questions: 1. How many documents remain about each author? 2. How do the authors' birth/death years influence their records? 3. Is any periodicity observable in the records?
We extracted research data from "Japan Search". This digital archive is operated by the National Diet Library, which aggregated 31,043,183 metadata records at the time (Cabinet Office Intellectual Property Strategy Promotion Office 2024). It contains from 67.6% of metadata from four archives (e.g. the National Diet Library Digital Collection), except non-compliant publishers, private collections, and records where API access is restricted. We compiled the list of 400 authors who lived in the 16th–19th period, by selecting from (Nichigai Associates 2005). Then we retrieved birth/death years, and annual material counts via SPARQL endpoints provided by Japan Search ( https://jpsearch.go.jp/static/developer/ja.html). If birth/death years were unavailable, we used Kotobank or Wikipedia to find them.
This study counted materials indexed by the author's name (excluding derivative works where the author is the subject). (a)Total records per author. (b)Time-series data: Material counts by year (b-1: raw values) and adjusted values (b-2: set birth year to 0). Then, we applied (c) a state-space model (Kalman filter) to the top-author to decompose time-series trends into levels, trends, and seasonality (testing 30 and 60-year cycles), and calculated RMSE (Root Mean Squared Error) and MAE (Mean Absolute Error).
(a)The total number of materials followed a long tail. Though top authors (Kyokutei Bakin)
A writer of popular fiction in the late Edo period; compared to his contemporaries, he had a long career and a large body of works, including his working journals. A series of historical tales had been intended for a readership of women and children, and was distributed from cities to rural areas through the rental book system. Publishers and literary figures born in the Meiji period often recalled that they had read these works during their childhood.
Regarding research questions: 1. The top author has an exceptionally high score, and the others are in a long-tail distribution. 2. The upward phase in raw values likely corresponds to periods when multiple authors were co-active or reappraisal (a sort of “golden age”). 3. Periodicity does not observe a top author, at least. Clear periodicity may be difficult to assume for others. Though only a few authors experience it, "posthumous fame" occurs not only immediately after death, but also over centuries. Contribution of 30-year cycle (17.7%), Contribution of 60-year cycle (26.5%) are estimated to contribute to the observed locations, but RMSE (262.7) and MAE (188.0) are high scores: this means the prediction error is large, because outliers have a high impact.
This research only focused on an author who got the highest score, and could not regard multiple peaks, outliners, details of specific incidents or environmental factors for their time series. Japan witnessed the spread of modern printing from the late 19th to the early 20th century, and throughout the 20th century, the ‘Complete Works’ of literature became accessible to the public. The data and resources used in this study rely on their academic and industrial achievements; however, they also have the limitation that unknown authors have not yet re-evaluated. We need to expand the sample size including derivative works and test a state-space model applicable to non-stationary time-series transitions with multiple peaks (e.g. Non-Gaussian State Space Representations), of multiple authors who have their own trends. We would aim to clarify whether writers born in the 19th–20th century show similar patterns or not. In this century, Japan has undergone several shifts in dominant media. If these have affected lifespans of writers' reputations, some changes will occur in long-term scores. Although this result alone does not permit causal inference, we hope that methods in this research will provide foundational findings for reappraisal of the achievements of literary studies to date.