Journal of AI and Data Mining Vol 8, No 3, 2020, 379-389. DOI: 10.22044/JADM.2020.7326.1871 Improving Accuracy of Recommender Systems using Social Network Information and Longitudinal Data B. Hassanpour1, N. Abdolvand2* and S. Rajaee Harandi2 1. Department of Electrical, Computer and IT Engineering, Qazvin Islamic Azad University, Qazvin, Iran. 2. Department of Management, Faculty of Social Sciences and Economics, Alzahra University, Tehran, Iran. Received 01 August 2018; Revised 21 July 2019; Accepted 31 March 2020 *Corresponding author: [email protected] (N. Abdolvand). Abstract The rapid development of technology, the Internet, and the development of electronic commerce have led to the emergence of the recommender systems. These systems assist the users in finding and selecting their desired items. The accuracy of the advice in recommender systems is one of the main challenges of these systems. Regarding the fuzzy system capabilities in determining the borders of user interests, it seems reasonable to combine it with social network information and the factor of time. In this work, for the first time, we try to assess the efficiency of the recommender systems by combining fuzzy logic, longitudinal data, and social network information such as tags, friendship, and membership in groups. Also the impact of the proposed algorithm for improving the accuracy of the recommender systems is studied by specifying the neighborhood and the border between the users’ preferences over time. The results obtained reveal that using longitudinal data in social network information in memory-based recommender systems improves the accuracy of these systems. Keywords: Recommender System, Social Network, Longitudinal Data, Fuzzy Logic, Tags, Membership. 1. Introduction and IMDb use the recommender systems in order In the recent years, with the growth of Internet to provide a good service for their customers [3]. users, many online networks have been Of course, not only the consumers and individuals developed, and thousands of people talk online use the recommender systems but the suppliers together [1]. Every day, the number of articles, and vendors will also benefit from these tools. music files, movies, books, and web pages on the Because of a large number of customers and Internet is increasing. In such an environment, goods, the recommender systems try to give users people do not know what to do with the enormous a better suggestion. Regarding the information amount of information. They are often unaware of requirements of these systems, the mechanisms the opportunities because of the high volume of that enable these systems to collect the required data and, in some cases, irrespective of the information are necessary. Another issue that is decision in this regard. In the past, the opinions of considered in some networks, and is also friends, classmates or colleagues were used to considered in this work, is the matter of time. A overcome this type of problem but today no one person may change or forget her/his interest over can offer various suggestions according to all the time and reminding them may change one's available information. This data growth and preferences and desires [4]. The time of rating massive data sources have created new fields of goods in social networks indicates the users’ data mining and pattern discovery [1]. The behavior over time. Therefore, using the time recommender systems were developed in dimension will improve advice estimation. Failing response to the changing needs of people and the to pay attention to the time people spend on number of choices in every field that help users to networks and neglecting to update the find and select items [2]. Today, many websites considerations of network requirements can lead including Amazon, CDNOW, Barnes & Noble, to a potentially significant loss of information [5]. Abdolvand et al./ Journal of AI and Data Mining, Vol 8, No 3, 2020. Moreover, the information explosion may not This work begins with a review of the literature, necessarily improve the quality of people’s lives, followed by a description of the research method and finding facts and knowledge in the wealth of used and proposed algorithm. Finally, the information available can be time-consuming and implications and conclusions will be explained. even frustrating. Therefore, having an intelligent system capable of automatically learning the 2. Literature review users' interests, filtering unrelated interests, The recommender systems have become an offering the relevant information in a limited essential research area. Their main purpose is to time, and helping users with product-selection identify the users’ neighbors based on profile decisions is essential. The recommender systems similarity and then suggesting something that the counter the problem of information overload and neighbors had already liked [10]. The assist with the problem solving by recommending recommender systems are intelligent systems that items to users [6]. Although several memory- provide appropriate recommendations to everyone based recommender algorithms have been and help the users to find and select their required proposed, an algorithm that uses the social items by finding and then analyzing their data [2, network data and addresses the time factor has not 11]. Usually, these systems are not able to offer been introduced [7]. Furthermore, many of the suggestions without an accurate information about current approaches in the recommender systems the users and their desired items (e.g. movies, focus on offering users the most relevant music, books) [12]. According to Liang et al. recommendations and overlook the background [11], the recommender system is a subset of the information such as time, place, and people decision support systems and is defined as the participating in a given action. In other words, the information system with the ability to analyze the traditional recommender systems use only the two past behavior and make recommendations for entities of users and elements in the process of current issues. Moreover, the recommender providing advice. Thus the time series analysis in systems are algorithms that provide the best and combination with other features of social most accurate recommendations through the networks can provide a good measure of suitable exploration of users’ associated information from recommender systems [8]. Given the different the relevant databases. Such systems find patterns preferences of users at different times, the in the users’ data by examining their past choices recommendations users receive should be based and displaying appropriate recommendations on these preferences and should not be based on them [11]. Therefore, one of the independent from the recommendations received essential goals of such systems is to collect data at other times [9]. This raises an interesting related to the users’ interests and items in the question: given the importance of time and the system such as videos [12]. In a general influence of friends and people in decision- classification, the recommender systems are making, how can a time component be added to divided into two categories: traditional social network information to improve the recommender systems and computational efficiency of the recommender systems? Because intelligence methods [13]. The recommender of the acceptable results of the Jin et al. [4] systems based on computational intelligent method, in comparison with other methods, methods use artificial intelligence tools such as especially techniques based on social networks, it Bayesian techniques, neural networks, clustering is used as a reference method in this work. Hence, techniques, genetic algorithms, and fuzzy theory in this work, we aimed to design memory-based to build the proposed model. The traditional recommender systems through the integration of recommender systems are divided into three longitudinal data and social network information. general categories: collaborative filtering, Therefore, a recommender system is designed by content-based recommendation, and hybrid combining fuzzy logic, longitudinal data, social approaches [2,13]. The content-based network information such as tags, friendship, and recommendation method chooses the membership in groups, and by using the impact of characteristics of the items with the highest long-standing interest, although with variable popularity among celebrities like directors and coefficients so that the proposed method can actors to offer them to others. Hybrid approaches improve the efficiency of previous recommender make recommendations by combining the methods. Overall, this system has a huge impact collaborative filtering and content-based on improving the efficiency of the proposed recommendation approaches. method. 380 Abdolvand et al./ Journal of AI and Data Mining, Vol 8, No 3, 2020. Collaborative filtering is the most widely used that were as accurate as those produced through a technology in the recommender systems and CF approach but with a better efficiency. makes suggestions based on the ranking of active The ability to assign a tag to an item by the user is users compared to their neighbors [13,14]. one of the features of most social recommender Collaborative filtering is divided into two systems. Therefore, the users can allocate tags to categories: memory-based and model-driven each product based on their preferences. The approaches. Memory-based algorithms operate on social tagging systems have been through the