Yoshiko Arima

Prof. Dr. Yoshiko Arima

orcid

Research Focus & Metrics

Academic Career Length

9Years (2018-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
4 July 2018 in Scientific Reports
Verified
Journal Article
Open access
PDF file
YAYoshiko Arima
Yoshiko Arima
RYRyoji Yukihiro
Ryoji Yukihiro
YHYosuke Hattori
Yosuke Hattori
We conducted two experiments to investigate the effects of word list consistency and group collaboration on false memory. Using the DRM (Deese-Roediger-McDermott) paradigm, the first experiment (n = 121) examined false memory in a group; participants collaborated to select keywords using a between-subject condition of a consistent or randomized word list. The proportion of false responses was larger than that of error responses, especially for the consistent word list condition (p < 0.001), and group collaboration increased false recognition for both word lists (p < 0.001). Using an applied within-subject word list condition, the second experiment (n = 119) confirmed the results of Experiment 1. Furthermore, individual differences Euclidean distance model analysis revealed a difference in the shared cognitive dimension related to group false memory.


13 April 2023 in BMC Digital Health
Verified
Journal Article
Open access
PDF file
YAYoshiko Arima
Yoshiko Arima
This study explored physical activity during remote work, most of which takes place while sitting in front of a computer. The purpose of Experiment 1 was to develop a classification for body motion by creating a neural net that can distinguish among several kinds of chest movement. Experiment 2 examined the effects of chest movements on stress and performance on the Navon test to validate the model developed in Experiment 1. The procedures for this study were as follows. Experiment 1: Creation of the body movement classification model and preliminary experiment for Experiment 2. Data from five participants were used to construct a machine-learning categorization model. The other three participants participated in a pilot study for Experiment 2. Experiment 2: Model validation and confirmation of stress measurement validity. We recruited 34 new participants to test the validity of the model developed in Experiment 1. We asked 10 of the 34 participants to retake the stress measurement since the results of the stress assessment were unreliable. Using LSTM models, we classified six categories of chest movement in Experiment 1: walking, standing up and sitting down, sitting still, rotating, swinging, and rocking. The LSTM models yielded an accuracy rate of 83.8%. Experiment 2 tested the LSTM model and found that Navon task performance correlated with swinging chest movement. Due to the limited reliability of the stress measurement results, we were unable to draw a conclusion regarding the effects of body movements on stress. In terms of cognitive performance, swinging of the chest reduced RT and increased accuracy on the Navon task (β = .015 [-.003,.054], R2 = .31). LSTM classification successfully distinguished subtle movements of the chest; however, only swinging was related to cognitive performance. Chest movements reduced the reaction time, improving cognitive performance. However, the stress measurements were not stable; thus, we were unable to draw a clear conclusion about the relationship between body movement and stress. The results indicated that swinging of the chest improved reaction times in the Navon task, while sitting still was not related to cognitive performance or stress. The present article discusses how to collect sensor data and analyze it using machine-learning methods as well as the future applicability of measuring physical activity during remote work.

4 December 2023 in Research Square
Verified
Preprint
Open access
PDF file
YAYoshiko Arima
Yoshiko Arima
This study aimed to examine the appropriateness of text analysis in an experimental study to ascertain the common knowledge effect on the discussion process. Two mock jury experiments were conducted wherein 204 participants were given information about a fictitious murder case, choosing from verdicts of “guilty”, “not guilty”, and “presumed innocent”. Experiment 1 set conditions using the information distributed to each group member. Latent Dirichlet allocation topic model was conducted to explore topics in chatlogs that would be influenced by the information allocated to each member. ANOVA revealed that the experimental conditions, “sharing (high/low) × information (alibi/dummy)” had effects on topic ratios. Experiment 2 investigated false memory related to the topics identified in Experiment 1. The results of the Deese, Roediger, and McDermott experiments conducted pre and post-discussion revealed that false memories increased post-discussion. Mediation analysis suggested that a certain topic ratio correlated with false memory.

17 June 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
YAYoshiko Arima
Yoshiko Arima
SOSusumu Ohnuma
Susumu Ohnuma
How do humans form trust with AI agents? We address this question with a computational model of human trust game behaviour grounded in the Active Inference Framework (AIF), tested against two behavioural experiments with bot partners. The model introduces a parameter γ, a trial-by-trial weighting of social versus economic inference in the free energy computation; higher γ indicates an internal model in which both self and other are predicted to prioritize social over economic value. Participants’ γ was estimated from ob- served behaviour via gradient descent. Across both experiments, the model achieved high predictive accuracy for trust/betrayal choices (F1 > .80) under cooperative and betrayal conditions alike. Study 1, in which bots maintained a fixed cooperation or betrayal pattern, replicated the simulated belief trajec- tories with only a small effect of Bot condition on γ. Study 2 useda between- subjects design contrasting a Trust condition (continuous cooperation) with a Betray condition (cooperation–betrayal–cooperation). The simulation pre- dicted that γ would rebound once cooperation resumed; in human data it did not. Beliefs (posterior predictions of the partner’s next cooperation) re- covered partially in the final phase, consistent with precision convergence in Bayesian updating, but investment rate and γ remained suppressed. Thus humans update their predictions about a bot’s behaviour in line with Active Inference accounts, yet their underlying social–economic value weighting car- ries a longer trace of the betrayal. This asymmetry, observed only in humans, points to factors beyond the present model, including individual differences such as sex.

17 August 2026 in Computers in Human Behavior Reports
Verified
Journal Article
Open access
YAYoshiko Arima
Yoshiko Arima
SOSusumu Ohnuma
Susumu Ohnuma