International Journal of Systematic Innovation

NLP Development of AI- Driven Autonomous Socially Assistive Robots (SAR)

Demiral Akbar — 2026-02-26

This study focuses on addressing the growing need for localized language support in Socially Assistive Robots (SARs) due to rising labor costs and the limitations of human labor in developed countries. The research specifically aims to develop a Turkish Natural Language Processing (NLP) module to enhance SARs' social interaction capabilities and integration into smart living spaces. By leveraging advanced machine learning models such as Transformers and BERT, the study fine-tunes these models for the Turkish language. The system was tested within a university setting, achieving notable results, including a 95% accuracy in voice recognition, 82% in model response accuracy, and a 92% speech comprehensibility rating among native Turkish speakers. These outcomes highlight the potential of SARs with localized language support to improve user experiences in various public and educational settings in Turkey. The research also underscores the importance of integrating NLP into SARs to overcome language barriers and enhance their functionality in diverse linguistic environments. Future work is encouraged to refine these systems and explore their applications in other sectors, contributing to the broader field of AI and robotics.

Identifying mineral potentials related to geological structures using deep learning

hadi shahraki — 2026-02-26

Artificial intelligence is increasingly being used as a powerful tool in various industries, including earth sciences. Geological structures play an undeniable role in the formation of mineral potentials. Investigating these structures relies on image and satellite data observations, as well as expert opinions, which can be a time-consuming process. Artificial intelligence can serve as a useful tool to expedite this process and enhance the accuracy of mineral potential identification. This article presents a new model based on deep neural networks for identifying mineral potentials. The unique feature of the proposed method is the incorporation of morphological data alongside multi-spectral data to identify mineral potentials. To evaluate the effectiveness of the proposed method, ASTER satellite images from a region in the southeast of Iran were utilized. The obtained results demonstrate an improvement in the accuracy of the proposed method compared to similar approaches.

Variability in Gesticulation Patterns: A Robust Framework for Recognizing Self-Co-articulated Dynamic Gestures

Shweta Saboo — 2026-02-26

This paper presents a dynamic hand gesture recognition system which addresses the effects of variation in pattern hand gestures when gesticulated in different ways by various users. In the proposed system, a new set of features has been proposed which divides the gesture into two equal halves and feature extraction is done after removal of self-co-articulation. Efficiency of proposed system is checked on a new set of gestures recorded in ‘LNMIIT Dynamic Hand Gesture Dataset-4’ which consists of videos recorded according to different patterns. The performance of the proposed system is calculated with different features combined with individual as well as combination of classifiers like SVM, k-NN, Naïve Bayes, ANFIS and DA classifiers. It has been concluded that recognition accuracy of Naïve Bayes classifier comes out to be 93.13% which is best among all the classifiers. Recognition accuracy improves by about 10% with the increase in number of features.