NEOJAPAN Inc. (3921)

Overview Threads132 Thread groups125 News1 Papers12 Documents92 Company

Papers

Research linked through this company’s shared themes. These papers may study a broader subject.

1–12 of 12

Corporate governance and capital policyTechnical Research Group Materials, Japanese Society for Artificial Intelligence2026-10-06

Corporate Dividend Policy under Common Ownership: Examining Dividend Similarity Using HypergraphsHarada Haruki (原田春貴)Using data on 4,291 companies covering 43,800 firm-years from 2005 to 2023, this study represents common-ownership relationships with hypergraphs and examines similarity in companies’ dividends (DOE). It analyzes the relationship between common ownership and dividend similarity using hypergraphs, which can represent investors holding multiple companies simultaneously, and the Zhou and HTV statistics, rather than conventional pairwise analysis.

ThemesStable, Continuous Shareholder Returns and Dividend Increase Policy+1 more

Overall scoreThin

39/ 100

Strongest

Relevance to the Japanese Market67

Novelty67

All seven categories

Disclosure & Text AnalysisTechnical Research Group Materials, Japanese Society for Artificial Intelligence2026-10-06

Analysis of Risk Information Disclosures by Japanese Companies Using a Structural Topic ModelContent analysis of risk information disclosures using a structural topic modelAikawa Nobuhiro (相川延弘) · Yoshida Mitsuo (吉田光男)Risk disclosures by 1,829 Japanese companies over the 10-year period from 2015 to 2024, collected from EDINET and other sources, were analyzed using a 10-topic structural topic model (STM). The analysis shows that the composition and concentration of risk disclosure topics changed following the 2019 disclosure system reforms, with the share of certain topics, particularly IT-related risks, increasing.

ThemesDisclosure of no material changes to business and other risks+3 more

Overall scoreTypical

46/ 100

Strongest

Relevance to the Japanese Market100

Novelty67

All seven categories

Corporate governance and capital policyarXiv2026-08-26

The Dynamic Trade-offs of Dual-Class Shares: Founder Incentives and Private Benefits of ControlThe Dynamic Trade-Off of Dual-Class SharesHyunseob Kim · Doron Levit · Roni MichaelyUsing a 1971–2022 panel of U.S. dual-class companies, the study finds that firm value, measured by Tobin’s q, rises by 0.21 in the year following a change from single-class to dual-class shares, while the valuation gap nearly disappears after about 10 years. By contrast, innovation, measured by patent citations, increases persistently. In mature companies, private benefits of control expand, making investment less responsive to opportunities.

ThemesControlling Shareholder Structure Led by the Founder's Family+3 more

Overall scoreThin

41/ 100

Strongest

Novelty100

Empirical Rigor67

All seven categories

Market Structure and Investor BehaviorarXiv2026-08-17

Targeting Employment Support to Job Seekers’ Cognitive Biases: A Randomized Experiment in FranceTargeting Support Using Job Seekers' Biases: A Randomized ExperimentBruno Crépon · Aurélien Frot · Christophe GaillacA study of 52,465 job seekers registered with France’s public employment service compared recommendations encouraging them to broaden their occupational choices with a motivational intervention designed to increase job-search effort and desired wages. The latter was effective for pessimistic job seekers, increasing time spent searching by 0.9 hours per week and raising the probability of reemployment within 12 months by 2.2 percentage points. Occupational recommendations increased applications among optimistic job seekers, but did not clearly improve reemployment outcomes.

ThemesLabor Shortages and Strengthening Organizational Management Systems+2 more

SectorsServices

Overall scoreAlmost nothing

29/ 100

Strongest

Novelty67

Practicality40

All seven categories

Market Structure and Investor BehaviorarXiv2026-08-17

The Information Value of Skill Credentials in the Generative AI Era: Kaggle’s Evaluation System and Medal ObsolescenceStranded credentials: how a skill-signaling market absorbed generative AISong YaoUsing Kaggle’s 2010–2026 archive, the study examines how well medal credentials predict performance on subsequent hidden tests, based on 444,698 participation records. The predictive power of medals was concentrated almost entirely in the first year after they were earned. The information content of medal holdings in upload-based competitions fell by 82% in the AI era, but half to three-quarters of that decline was explained by medal obsolescence resulting from the discontinuation of the competition format, rather than by AI.

ThemesRisk of securing and developing AI and data talent+3 more

Overall scoreThin

35/ 100

Strongest

Empirical Rigor83

Novelty67

All seven categories

Disclosure & Text AnalysisarXiv2026-08-14

When Is Financial News Priced into Stock Prices? Evidence from U.S. Equities Using LLM-Based Event Classification and Placebo AdjustmentBuy the Rumor, Sell the News: When Is News Priced In?Alireza Kargarzadeh · Nariman Khaledian · Navid ParviniThe study classifies approximately 4.57 million financial news items into 17 event types and analyzes beta-adjusted abnormal returns before and after the news for 1.68 million stock-day events covering approximately 3,000 stocks. Directional stock-price movements concentrate around the publication date. When rumors are present, the news is almost fully priced in on the rumor date, while placebo-adjusted results show that quantitative fundamental news continues to have an effect and soft, story-driven news subsequently reverses.

ThemesDisclosure of Historical Sales and Profit Results+2 more

Overall scoreTypical

56/ 100

Strongest

Applicability to Systematic Strategies100

Novelty100

All seven categories

Portfolio Construction & Risk ModelsarXiv2026-07-28

Long-Memory GARCH-Type Volatility Model Based on a Two-Dimensional Markov ChainLong-memory GARCH via a two-dimensional Markov chainKyungsub Lee · Kennedy Titus KayakiA two-dimensional Markov-type GARCH model was proposed, comprising the variance level and memory scale, with the decay rate of past shocks changing endogenously. Theoretical stability was established, and the simulations reproduced long memory, but empirical tests across six markets did not confirm consistently superior forecasting performance to FIGARCH or HAR-RV.

ThemesMarket Price Risk on Held Financial Instruments+1 more

SectorsSecurities and commodity futures+2 more

Overall scoreTypical

45/ 100

Strongest

Macro Relevance75

Novelty67

All seven categories

AI and LLM Methods in FinancearXiv2026-05-29

Efficient Estimation of Longitudinal Treatment Effects Using a Covariate Transition Model in Randomized ExperimentsModeling Covariate Transition for Efficient Estimation of Longitudinal Treatment Effects in Randomized ExperimentsNaoki Chihara · Tatsushi Oka · Yasuko MatsubaraWe propose a regression-adjusted estimator that incorporates a transition kernel and forward integration to avoid the bias that can arise from simply adjusting for covariates that change over time after treatment. We establish asymptotic normality and semiparametric efficiency, and report that simulations and an A/B test conducted by a Japanese video-streaming service achieved lower estimation errors and standard errors than the conventional empirical estimator.

ThemesReducing Churn Through Enhanced Service Functionality and Customer Success+3 more

SectorsInformation and Communications

Overall scoreThin

36/ 100

Strongest

Novelty67

Empirical Rigor58

All seven categories

AI and LLM Methods in FinanceTechnical Research Group Materials, Japanese Society for Artificial Intelligence2026-03-18

An LLM-Generated Method for Extracting Unexpected Businesses from Securities Reports and Explaining Their Relevance to QueriesGenerating business descriptions that express the relevance between queries using unexpected businesses and the businesses themselvesKobayashi Yoshimasa (小林義正) · Sakai Hiroyuki (酒井浩之) · Nishina Kei (仁科慧)We proposed a method that extracts companies’ business activities and R&D activities from annual securities reports and classifies them as either core businesses or unexpected businesses based on their share of revenue and similarity to the core businesses. When the classification labels and an overview of the company’s business as a whole were provided to an LLM to generate business descriptions, the classification F1 score for unexpected businesses was 0.517, exceeding the 0.482 score achieved when only expanded core-business descriptions were used. The proposed method also outperformed the baseline in a comparative evaluation of the business descriptions.

ThemesExploring and developing new businesses+2 more

Overall scoreThin

44/ 100

Strongest

Novelty100

Empirical Rigor58

All seven categories

Disclosure & Text AnalysisTechnical Research Group Materials, Japanese Society for Artificial Intelligence2025-10-08

Building a Dividend Policy Dataset and Conducting Investment Analysis Using Annual Securities Reports, BERT and LLMsBuilding and Analysing Dividend Policy Data Using Annual Securities Report TextTakeshita Sora (竹下蒼空) · Takano Kaito (高野海斗) · Nishina Kei (仁科慧)A dataset covering 2015–2025 was constructed by using BERT and LLMs to extract and classify future dividend policies, such as dividend increases and the avoidance of dividend cuts, from Japanese companies’ annual securities reports. After evaluating classification accuracy, portfolios were created by dividend policy, and the dividend-increase and progressive-dividend groups showed higher information ratios than the no-dividend-cut group (0.973 versus 0.380 on an equal-weighted basis).

ThemesStable, Continuous Shareholder Returns and Dividend Increase Policy+1 more

Overall scoreTypical

48/ 100

Strongest

Applicability to Systematic Strategies80

Novelty67

All seven categories

Financial Performance & Fundamental ForecastingTechnical Research Group Materials, Japanese Society for Artificial Intelligence2025-10-08

Predicting Profit Changes at Japanese Companies Using Large Language Models and Out-of-Sample EvaluationAnalysis of Profit Forecasts Using LLMs and Out-of-Sample EvaluationShirai Yusuke (白井祐典) · Ichikawa Yoshihiko (市川佳彦) · Nakagawa Kei (中川慧)Using EDINET-BENCH, which covers Japanese listed companies, together with new earnings data that was unavailable when the benchmark was created, this study analyzed the accuracy of LLM-based profit-change forecasts and company characteristics. Claude 3.7 Sonnet achieved an out-of-sample ROC-AUC of 0.6327, exceeding the ROC-AUC of 0.6075 on the benchmark data.

ThemesUsing AI Text Analysis of IR Information for Investment+2 more

Overall scoreDetailed

60/ 100

Strongest

Novelty100

Relevance to the Japanese Market83

All seven categories

AI and LLM Methods in FinanceTechnical Research Group Materials, Japanese Society for Artificial Intelligence2024-10-15

Building and Benchmarking a Finance-Focused Large Language Model Using Financial Reports and Central Bank DocumentsBuilding and Validating a Large Language Model Specialized in FinanceHirano Masanori (平野正徳) · Imashiro Kentaro (今城健太郎) · Saito Shunta (齋藤俊太)A pretraining dataset for the financial sector was created from financial reports, central bank documents, company information and other sources, and continued pretraining was performed on rinna/nekomata-14b and Meta-Llama-3-70B. On a financial benchmark, the aggregate score improved from 0.4335 to 0.4716 for the nekomata-14b family and from 0.5658 to 0.5967 for the Llama-3-70B family. Instruction-following models created through model merging were also evaluated.

ThemesBusiness Transformation Through DX, AI and Big Data+2 more

SectorsInformation and Communications+4 more

Overall scoreThin

39/ 100

Strongest

Relevance to the Japanese Market83

Novelty67

All seven categories

Komori×Hone Systems

Impressed with Komori?

Komori is free because Hone Systems funds it.

We're engineers who love applied AI. We fit it into the way your company already works, and we stay after launch: keeping it monitored, keeping it sharp, and bringing you the next good idea before you have to ask.

Book 30 minutes→About Hone Systems

This company's documents

Filings

2026-09-11Semiannual Report
2026-04-27Annual Securities Report
2025-09-12Semiannual Report
2025-04-25Annual Securities Report
2024-09-11Semiannual Report

Timely disclosures

2026-10-15Received the highest-ranking “Leader” award in 5 categories at the “ITreview Grid Award 2026 Fall”; desknet's NEO recognized for the 30th consecutive period in the groupware category
2026-10-05desknet's NEO Version 9.7 Released, with Major Enhancements to AI Assistant Features Enabling Internal Deployment of the Latest LLM
2026-09-11Financial Results Presentation for the Second Quarter of the Fiscal Year Ending January 2027
2026-09-11Financial Results for the Second Quarter (Interim Period) of the Fiscal Year Ending January 2027 (Japanese GAAP) (Consolidated)
2026-09-09(Correction) Partial correction to the "Financial Results Presentation Materials for the First Quarter of the Fiscal Year Ending January 2027"
2026-09-01Notice of Nagoya Sales Office Relocation: Strengthening Support for Sales and Distribution Partners in the Chubu Region
2026-08-27Ranked No. 1 in the Information-Sharing Software/Services Category in the Nikkei Computer Customer Satisfaction Survey 2026-2027
2026-08-25Launch of the secure remote access service “moconavi Browser for desknet's”
48 more

Threads

2026-10-30Responding to Advances in AI Technology and Company-Wide AI Adoption
2026-10-15Groupware Division Ranked No. 1 in Customer Satisfaction for Two Consecutive Years
2026-10-05Developing New Products and Services, Raising Awareness, and Expanding Sales Channels
2026-09-11Expansion of the Cloud Services Market
2026-09-11Economic Instability, Yen Appreciation and Share Price Volatility
2026-09-11Growth in desknet's NEO Cloud Users and Migration from Applitus
2026-09-11Steady Performance of desknet's NEO Support Services
2026-09-11Revenue, Operating Profit, Ordinary Profit and Net Profit
124 more

Data sources

Filings37

Period2016 — 2026

Price15-min delay

The numbers come from filings and market data. Descriptions and extracted research are organized by AI; check the linked sources.