Page 211 - The-5th-MCAIT2021-eProceeding

Recommendation for Group of Users with LOD

a
b
c
Rosmamalmi Mat Nawi *, Shahrul Azman Mohd Noah , Lailatul Qadri Zakaria
a,b,c Centre of Artificial Intelligence Technology, Faculty of Technology and Information Science, The National University of Malaysia.
* Email: rosmamalmi@siswa.ukm.edu.my

Abstract

This paper presents a model-based movie recommender system by exploiting Linked of Data (LOD) technology to enrich
the item information. While numerous studies have been conducted on both conventional recommender systems and Linked
Data-enabled recommender systems, little effort has been made to investigate the potentials LOD in the context of group
formation. Our proposed framework adopts the LOD-based group recommender system model before the formation of
group to enhance the quality of recommendation by tackling the sparseness issue. The findings indicate that the proposed
model leads to a better result compared to the baseline. The GRS-LOD model improves prediction accuracy, as indicated
by the low RMSE and MAE errors, and exhibits good recommendation relevancy considering the high Precision, Recall,
and F1-Score metrics.

Keywords: Group Recommendation System; Linked Open Data; sparsity

1. Introduction

Recommender Systems (RS) are information filtering tools in the decision-making process that recommend
items to users. (El-Ashmawi et al., 2020). Although conventional research on RS has almost entirely focused
on offering recommendations to single users, there are several other scenarios where the system needs to provide
items to groups of users. Group Recommender Systems (GRS) has a significant challenge to provide
recommendations if insufficient data is received from group member rating (Ghazarian et al., 2014) since it
relies heavily on the available data sources. Furthermore, it is difficult to identify the most similar users when
there are insufficient data items ratings (Khusro et al., 2016).
Linked Open Data (LOD) datasets provide additional data that can be integrated into several domains (e.g.,
film) to augment item information. Hence, LOD may be beneficial when such a number or quality of a ready
dataset is insufficient. RS can effectively exploit LOD to deal with classical problems of cold-start and sparsity.
Even though it has been widely implemented in conventional RS from many perspectives (references LOD
untuk individual recommendation), the LOD technology approach in GRS is still underexplored. Thus, the
research presented in this paper proposed the use of LOD for group recommendation. The proposed model aims
to offer a method to tackle the sparseness issue prior to user group formation since limited rating affects
clustering.
In a nutshell, the contributions of our work are as follows:
 We proposed a GRS-LOD model that applies the LOD technology in enriching the item information to be
applied in GRS.
 We presented an approach for enhancing user information using similarity-based DBpedia attributes
('dbo:director' and 'dbo:starring').
 We illustrate how the model could be utilized to form a user cluster more efficiently by dealing with data
sparsity in group member profiles.

E- Proceedings of The 5th International Multi-Conference on Artificial Intelligence Technology (MCAIT 2021) [197]
Artificial Intelligence in the 4th Industrial Revolution