International Journal of Artificial Intelligence and Soft Computing (5 papers in press)

Regular Issues

• An Integrated Hybrid Deep Learning Framework of Speech Emotion Recognition for Personalised Movie Recommendations  
  by Syed Azeem Inam, Ghulam Mustafa, Abdul Rahim 
  Abstract: The study investigates the intricate relationship between SER and CRS by utilising advanced machine learning. It proposes a recommendation mechanism that goes beyond the conventional outcome of a content-based system to extract specific emotional nuances from speech data. It examines the emotional subtleties and the genre of the movies by integrating movie datasets with emotional classification. It further evaluates the emotional impact of movies and provides recommendations to viewers through vectorisation, cosine similarity, and genre-specific emotional weight. Overall, the classifiers reflect the system's outstanding performance in mapping user preferences to suggested movies, indicating a major improvement in recommendation systems. With an F1 score of 93%, the proposed CNN (93%) is identified as the best integrated deep learning model among other classifiers that which includes MLP (73%), LightGBM (90%), XGBoost (91%), Naive Bayes (34%), AdaBoost (49%), k-NN (74%), logistic regression (43%), LSTM (84%), RF (88%), DT (82%), and SVM (48%).
  Keywords: Speech Emotion Recognition (SER); Content-Based Recommendation System (CRS); Movie Recommendation; CNN.
  DOI: 10.1504/IJAISC.2025.10075561
   
  
• A Cascade Decoder-Based Method for Complex Causal Relation Extraction  
  by Wanli Su, Bo Shen 
  Abstract: Benchmarking is the continuous measurement and comparison of ones business processes against comparable processes in leading organisations to obtain information to help the organisation identify and implement improvements. Selecting the right benchmarking partners is crucial for achieving benchmarking success. In addition to the existing methods of partner selection, this study explores the application of unsupervised machine learning clustering tools for competitive, functional, peer and hierarchical benchmarking. The Euclidean K-distance method can be used to select the best performer within a group for both competitive and functional benchmarking. DBSCAN is an ideal tool for determining the peer group for peer benchmarking. GMM combined with Euclidean K-distance can be a perfect tool for hierarchical benchmarking. This study is illustrated through an empirical example involving the research benchmarking of educational institutions, universities, and premium institutions.
  Keywords: Complex causal extraction; Cause-Effect mapping; Entity identification; Cross-Attention; Cascade Decoder.
  DOI: 10.1504/IJAISC.2025.10075563
   
  
• Benchmarking Partner Selection using Machine Learning Methods  
  by Geo George, Georgy Kurien 
  Abstract: Benchmarking is continuously measuring and comparing one's business processes against comparable processes in leading organizations to obtain information to help the organization identify and implement improvements. The selection of the right benchmarking partners is crucial for benchmarking success. In addition to the existing methods of partner selection, this study explores the application of unsupervised machine learning clustering tools for competitive, functional, peer and hierarchical benchmarking. The Euclidean K distance method can be used to select the best in the group for both competitive and functional benchmarking. DBSCAN is an ideal tool for determining the peer group for peer benchmarking. GMM combined with Euclidean K distance can be a perfect tool for Hierarchical benchmarking. This study is illustrated through an empirical example involving the research benchmarking of educational institutions, universities, and premium institutions.
  Keywords: Benchmarking; Competitive benchmarking; Functional benchmarking; Peer benchmarking; Hierarchical benchmarking.
  DOI: 10.1504/IJAISC.2025.10075669
   
  
• Automatic Number Plate Detection Using Deep Learning.  
  by Madhura Bhosale, Yogesh Angal 
  Abstract: Automatic number plate recognition (ANPR) is a critical technology in intelligent transportation systems, playing a vital role in autonomous driving, traffic law enforcement, and vehicle monitoring. Accurate and efficient ANPR is essential for applications such as over-speed detection and automated traffic violation management. This paper presents a robust ANPR system leveraging YOLOv8, a state-of-the-art deep learning-based object detection model, for precise number plate localisation. To enhance character recognition accuracy, we employ advanced optical character recognition (OCR) techniques, specifically EasyOCR and PARSeq, ensuring reliable extraction of alphanumeric data from detected plates. Experimental results demonstrate that our system achieves approximately 96% accuracy in number plate detection and over 91% accuracy in character recognition, making it highly suitable for real-world deployment. The pro-posed approach enhances traffic surveillance, automated toll collection, and law enforcement by providing a high-speed, scalable, and accurate ANPR solution.
  Keywords: Automatic Number Plate Detection; Object detection; Machine Learning; Arti-ficial Intelligence.
  DOI: 10.1504/IJAISC.2025.10075703
   
  
• Event Causality Identification via a Method of Integrating Context-Awareness and Knowledge-Enhancement  
  by Shaonan Liu, Bo Shen 
  Abstract: In the field of natural language processing, identifying causality between events is conducive to enhance the logical reasoning ability of language models. Due to the limited scale of datasets and the diversity of event causality expressions, mainstream representation-based methods exhibit poor generalization to unseen causality and struggle to identify implicit causality. To address these issues, we proposes a model named CAKE which integrating context-awareness and knowledge-enhancement for event causality identification.Specifically, the model constructs a context-aware reasoner that learns contextual semantics and event-independent global causal patterns through masking operation and adversarial perturbation mechanism, thereby improving the model’s generalization to unseen causality. Additionally, it builds a knowledge-enhanced reasoner by incorporating external knowledge, which aggregates one-hop neighbor information and models multi-hop relational paths to enrich event representations, enhancing the model’s capability to identify implicit causality. Experimental results demonstrate that the proposed model significantly outperforms baseline methods on two public datasets.
  Keywords: Event causality identification;Context awareness; Knowledge enhancement; Natural language processing.
  DOI: 10.1504/IJAISC.2025.10075705