Cosine Jaccard Distance (CJD): An Improved Similarity-Based Approach for Fake News Detection Using Machine Learning and Deep Learning Models
International Journal of Innovation in Science and Technology (IJIST) · 30.05.2026 · 1 (1) · 160-171
DOI: 10.64290/ijsat.v1i1.11 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Innovation in Science and Technology (IJIST), online erschienen am 30.05.2026. Verfasst von Aliyu Shuaibu und 2 weiteren Autor:innen.
Abstract (Originalsprache)
The continuous growth of digital communication technologies has accelerated the spread of fake news and misinformation, posing serious threats to information credibility, public trust, and decision making. Although machine learning and deep learning models have demonstrated effectiveness in fake news detection, their performance is often constrained by traditional similarity measures such as cosine similarity, which may not effectively capture relationships in sparse and high-dimensional textual data. This study proposes a modified cosine similarity framework by replacing the Euclidean distance component with Jaccard distance to develop the Cosine Jaccard Distance (CJD) approach for fake news detection. The proposed method enhances semantic and set-based feature representation, particularly for binary and discrete textual data. The study evaluated the performance of Support Vector Machine (SVM), Random Forest (RF), K-Nearest Neighbors (KNN), and a Deep learning Long Short-Term Memory (LSTM) classifiers under both the traditional cosine similarity framework and the proposed CJD framework. Experimental results from the baseline model revealed that SVM achieved the best overall performance with an accuracy of 98.7%, precision of 98.9%, recall of 98.4%, and F1-score of 98.6%, while also recording the lowest cross-entropy loss, false positive rate, and false negative rate. However, under the proposed CJD framework, the LSTM model emerged as the best-performing classifier, achieving the highest accuracy of 99.5%, recall of 99.8%, and F1-score of 99.6%, alongside the lowest cross-entropy loss (0.018), false positive rate (0.007), and false negative rate (0.002). KNN and RF also demonstrated strong and competitive performance, while SVM recorded comparatively lower results within the modified framework. Overall, the findings demonstrate that the proposed Cosine Jaccard Distance approach improves classification effectiveness and robustness compared to the traditional cosine similarity framework for fake news detection.
Zitieren
APA
Shuaibu, A., Idris, B. I. I., & Ahmad, Y. (2026). Cosine Jaccard Distance (CJD): An Improved Similarity-Based Approach for Fake News Detection Using Machine Learning and Deep Learning Models. International Journal of Innovation in Science and Technology (IJIST), 1(1), 160-171. https://doi.org/10.64290/ijsat.v1i1.11
Vancouver
Shuaibu A, Idris BII, Ahmad Y. Cosine Jaccard Distance (CJD): An Improved Similarity-Based Approach for Fake News Detection Using Machine Learning and Deep Learning Models. International Journal of Innovation in Science and Technology (IJIST). 2026;1(1):160-171. doi:10.64290/ijsat.v1i1.11
BibTeX
@article{Shuaibu2026,
author = {Shuaibu, Aliyu and Idris, Bello Ibrahim Idris and Ahmad, Yusuf},
title = {Cosine Jaccard Distance (CJD): An Improved Similarity-Based Approach for Fake News Detection Using Machine Learning and Deep Learning Models},
journal = {International Journal of Innovation in Science and Technology (IJIST)},
year = {2026},
volume = {1},
number = {1},
pages = {160--171},
doi = {10.64290/ijsat.v1i1.11}
}
EUVIMED