An Effective Recommender System Based on Clustering Technique for TED Talks

An Effective Recommender System Based on Clustering Technique for TED Talks

Faiz Maazouzi (University of Souk Ahras, Souk Ahras, Algeria), Hafed Zarzour (University of Souk Ahras, Souk Ahras, Algeria) and Yaser Jararweh (Department of Computer Science, Jordan University of Science and Technology, Irbid, Jordan)
DOI: 10.4018/IJITWE.2020010103

Abstract

With the enormous amount of information circulating on the Web, it is becoming increasingly difficult to find the necessary and useful information quickly and efficiently. However, with the emergence of recommender systems in the 1990s, reducing information overload became easy. In the last few years, many recommender systems employ the collaborative filtering technology, which has been proven to be one of the most successful techniques in recommender systems. Nowadays, the latest generation of collaborative filtering methods still requires further improvements to make the recommendations more efficient and accurate. Therefore, the objective of this article is to propose a new effective recommender system for TED talks that first groups users according to their preferences, and then provides a powerful mechanism to improve the quality of recommendations for users. In this context, the authors used the Pearson Correlation Coefficient (PCC) method and TED talks to create the TED user-user matrix. Then, they used the k-means clustering method to group the same users in clusters and create a predictive model. Finally, they used this model to make relevant recommendations to other users. The experimental results on real dataset show that their approach significantly outperforms the state-of-the-art methods in terms of RMSE, precision, recall, and F1 scores.
Article Preview
Top

1. Introduction

Since the advent of the Internet, digital information resources have become a means of communication, knowledge sharing and decision-making. These resources can influence the choice of the user by using information technologies and tools that help in decision-making. Nowadays, artificial intelligence replaces the intelligence of man. All intelligent agents such as recommender systems (RS) can obtain customized information. Their goal is to reduce the overload of information through a process of collecting, filtering and recommending information in a proactive manner. For example, recommender systems try to predict which products are best suited to users, based on their preferences collected in several ways, to perform the recommendation task. The recommendation must provide relevant objects that satisfy users rather than propose suggestions in relation to a commercial policy. We can encounter this situation in e-commerce applications. For example, to encourage the introduction of new clothing items from Amazon's partners, the RS suggests false recommendations (Wingfield & Pereira, 2002). These suggestions of the items provided by the RS to a user can be proposals of items to buy, news to read, music to listen, films to be seen or books to be read. The word “item” is the general term used to denote what the recommender system recommends to users. RS can be used to provide personalized information being primarily geared towards users who do not have sufficient skills to assess the immense number of items.

The initial idea in developing RS was simply to observe that the user tended to rely on the recommendations of other users for decision-making (McSherry & Mironov, 2009). Currently, recommendation engines rely on three paradigms (Ricci et al., 2011): RS based on content, RS based on collaborative filtering, and RS based on the hybrid method of recommendation, which is the combination of the two first paradigms (Kaššák et al., 2016). Collaborative filtering methods can be classified as memory-based or model-based approaches (Breese et al., 1998). They have been proven to be effective in the practice.

In the last years, several methods for collaborative filtering were proposed such as that proposed by Ha and Lee (2016), which suggested to use item-network-based collaborative filtering. Lee and Brusilovsky (2017) have used in their study the community membership information. In addition, data mining techniques were used in the recommender systems such as: clustering (Altingovde et al., 2013) and rough set-based association rule (Liao, & Chang, 2016).

In this paper, we focus on the model-based approach in our prediction strategy. The unsupervised classification (clustering) on matrix data, made up of users and items, makes it possible to form relevant and significant blocks. Model-based method constructs a prediction model, often probabilistic, based on a part of data, some of which use clustering techniques. Model-based clustering techniques have better scalability than the conventional collaborative filtering methods, because they make predictions in clusters, rather than across the entire database (Zhuang et al., 2013). The collaborative filtering approaches using model-based techniques attempt to provide more accurate results than memory-based systems.

In our study, the field of application is TED talks (www.ted.com), in which we use a dataset acquired from the TED website that is considered as a repository of lecture recordings given by prominent speakers (Hu, & Li, 2017; Pappas, & Popescu-Belis, 2015). Thus, this study has been developed following two steps. The first step consists of creating a TED user-user matrix based on PCC method; in the second step, we use our TED user-user matrix for creating a model. The method used to create this model is k-means clustering. Hence, the novelty of this paper is to design a new collaborative filtering recommendation algorithm based on clustering technique with the consideration of TED talk proprieties, which improves the accuracy of the recommendation.

Complete Article List

Search this Journal:
Reset
Open Access Articles: Forthcoming
Volume 15: 4 Issues (2020): 1 Released, 3 Forthcoming
Volume 14: 4 Issues (2019)
Volume 13: 4 Issues (2018)
Volume 12: 4 Issues (2017)
Volume 11: 4 Issues (2016)
Volume 10: 4 Issues (2015)
Volume 9: 4 Issues (2014)
Volume 8: 4 Issues (2013)
Volume 7: 4 Issues (2012)
Volume 6: 4 Issues (2011)
Volume 5: 4 Issues (2010)
Volume 4: 4 Issues (2009)
Volume 3: 4 Issues (2008)
Volume 2: 4 Issues (2007)
Volume 1: 4 Issues (2006)
View Complete Journal Contents Listing