TY - JOUR
T1 - Enhancing Book Recommendation Accuracy through User Rating Analysis and Collaborative Filtering Techniques
AU - Chongwarin, Jirat
AU - Manorom, Paiboon
AU - Chaichuay, Vispat
AU - Boongoen, Tossapon
AU - Li, Chunqiu
AU - Chansanam, Wirapong
N1 - Publisher Copyright:
© 2024 Telecommunications Association Inc.. All rights reserved.
PY - 2024/9/30
Y1 - 2024/9/30
N2 - Since online electronic books have become popular, book recommendation systems have been invented and challenged to handle the high demand from users in the digital era. This study aimed to develop and evaluate a book recommendation model using data mining techniques through RapidMiner Studio. The datasets used were comprised of 981,756 user ratings. Before conducting the data analytics, the data was pre-processed to eliminate duplicates and retain only the highest ratings. Collaborative Filtering (CF) techniques, particularly k-Nearest Neighbours (k-NN) and Matrix Factorization (KF), were employed to elicit insightful information for development and to highlight their capabilities in handling enormous datasets. Furthermore, statistical analysis, visualization, elementary modelling, and model combinations were investigated to compare their performance. To reinforce creditability, modelling techniques and parameter adjustments were integrated to optimize the performance of the algorithms, since the results indicated that different model settings and data partitions impacted the effectiveness of the recommendation system. Additionally, these results demonstrated the potential of hybrid models in improving the accuracy and efficiency of recommendation systems and highlighted the trade-off between algorithmic approaches and dataset characteristics that interplay in optimizing the performance of recommendation systems.
AB - Since online electronic books have become popular, book recommendation systems have been invented and challenged to handle the high demand from users in the digital era. This study aimed to develop and evaluate a book recommendation model using data mining techniques through RapidMiner Studio. The datasets used were comprised of 981,756 user ratings. Before conducting the data analytics, the data was pre-processed to eliminate duplicates and retain only the highest ratings. Collaborative Filtering (CF) techniques, particularly k-Nearest Neighbours (k-NN) and Matrix Factorization (KF), were employed to elicit insightful information for development and to highlight their capabilities in handling enormous datasets. Furthermore, statistical analysis, visualization, elementary modelling, and model combinations were investigated to compare their performance. To reinforce creditability, modelling techniques and parameter adjustments were integrated to optimize the performance of the algorithms, since the results indicated that different model settings and data partitions impacted the effectiveness of the recommendation system. Additionally, these results demonstrated the potential of hybrid models in improving the accuracy and efficiency of recommendation systems and highlighted the trade-off between algorithmic approaches and dataset characteristics that interplay in optimizing the performance of recommendation systems.
KW - collaborative filtering
KW - data mining techniques
KW - k-Nearest Neighbours (k-NN)
KW - matrix factorization
KW - RapidMiner Studio
UR - http://www.scopus.com/inward/record.url?scp=85206096638&partnerID=8YFLogxK
U2 - 10.18080/jtde.v12n3.976
DO - 10.18080/jtde.v12n3.976
M3 - Article
AN - SCOPUS:85206096638
SN - 2203-1693
VL - 12
SP - 51
EP - 72
JO - Journal of Telecommunications and the Digital Economy
JF - Journal of Telecommunications and the Digital Economy
IS - 3
ER -