Authors :
Mustapha Maidawa; A. Y. Dutse; Aminu Ahmad; Abdulsalam Ya’u Gital; Ismail Zahraddeen Yakubu
Volume/Issue :
Volume 8 - 2023, Issue 11 - November
Google Scholar :
https://tinyurl.com/3m2xr3pb
Scribd :
https://tinyurl.com/5yydc9yf
DOI :
https://doi.org/10.5281/zenodo.10297550
Abstract :
AI allows for a higher quality of
recommendation than can be achieved by conventional
recommendation methods. This has ushered in a new era
for recommender systems, creating advanced
observations of the relationship between users and items,
presented an expanded understanding of demographic,
textural, virtual, and contextual data as well as more
intricate data representations. However, the challenge for
the recommendation systems is to solve the problems of
sparsity, scalability, and cold start. The existing capsule
networks take times in training making it a slow
algorithm. Also, ignoring the sparsity in the datasets have
result to reduction in prediction accuracy. Other works of
literature already in existence add column or row
meanings to such sparse values. Because the mean
disregards the underlying correlation in the data,
accuracy is compromised. Hence, this study examined the
existing framework and the need to provide a solution to
the problem by proposing the inclusion of business
intelligence component framework base on recommender
system. Therefore, to address these issues, this research
proposed a hybrid collaborative base recommendation
system using an improved SVD and self-organized map
neural network (SOM) to improve cold start, accuracy,
speed and sparsity issue of the current recommendations
by combining SOM clustering to cluster the dataset, a
better SVD to reduce dimensionality and increase
sparsity, and a cooperative strategy to address accuracy
and sparsity concerns. Experimental result shows that the
proposed model has consistently performed better than
all the three state-of-the-art methods including the
Capsule Neural Network CF algorithm, the KNN CF
algorithm and the SVD+SOM clustering base
recommender system. This study has proven that data
mining can helps companies and business managers to
visualize hidden patterns and trends in datasets that were
not visible before. Whatever insights are revealed, they
make clear decisions that benefit both the company and
the customers and the stakeholders they serve.
Keywords :
Recommender System, K-Neareast Neighbour, Jaccard Distance, Euclidian Distance and Cosine Distance.
AI allows for a higher quality of
recommendation than can be achieved by conventional
recommendation methods. This has ushered in a new era
for recommender systems, creating advanced
observations of the relationship between users and items,
presented an expanded understanding of demographic,
textural, virtual, and contextual data as well as more
intricate data representations. However, the challenge for
the recommendation systems is to solve the problems of
sparsity, scalability, and cold start. The existing capsule
networks take times in training making it a slow
algorithm. Also, ignoring the sparsity in the datasets have
result to reduction in prediction accuracy. Other works of
literature already in existence add column or row
meanings to such sparse values. Because the mean
disregards the underlying correlation in the data,
accuracy is compromised. Hence, this study examined the
existing framework and the need to provide a solution to
the problem by proposing the inclusion of business
intelligence component framework base on recommender
system. Therefore, to address these issues, this research
proposed a hybrid collaborative base recommendation
system using an improved SVD and self-organized map
neural network (SOM) to improve cold start, accuracy,
speed and sparsity issue of the current recommendations
by combining SOM clustering to cluster the dataset, a
better SVD to reduce dimensionality and increase
sparsity, and a cooperative strategy to address accuracy
and sparsity concerns. Experimental result shows that the
proposed model has consistently performed better than
all the three state-of-the-art methods including the
Capsule Neural Network CF algorithm, the KNN CF
algorithm and the SVD+SOM clustering base
recommender system. This study has proven that data
mining can helps companies and business managers to
visualize hidden patterns and trends in datasets that were
not visible before. Whatever insights are revealed, they
make clear decisions that benefit both the company and
the customers and the stakeholders they serve.
Keywords :
Recommender System, K-Neareast Neighbour, Jaccard Distance, Euclidian Distance and Cosine Distance.
