Div-clustering: Exploring active users for social collaborative recommendation
Introduction
Networks and computer applications allow people to communicate with one another beyond spoken communications. However, countless opinions cause difficulty in judging the reliability of information. People are faced with hard decisions, including books to read, restaurants to visit for lunch, and movies to watch. Under such circumstances, the recommended solution is to refer to the opinions of classmates, friends, and colleagues. This solution does not only save time but also helps determine the most reliable recommendation. In addition, companies want their products to attract more customers. Loss of credibility in a product implies a greater loss of customers than expected, considering the associations of customers with other people. These mutual considerations can be solved by a recommender system (McDonald and Ackerman, 2000, Resnick and Varian, 1997). A recommender system is one of the most interesting research areas in addressing information overload. The system is an information filtering system subclass that seeks to predict what interests users according to their behavior, or consult experts for help.
A recommender system produces a list of recommendations either through collaborative filtering approach or content-based filtering approach. A collaborative filtering approach builds a model according to a user's past behavior, and uses that model to predict the items that this user may also have interest in Melville and Sindhwani (2010). A content-based filtering approach utilizes a series of discrete product properties to recommend additional items with similar characteristics (Bollen et al., 2007, Pazzani and Billsus, 2007). Sometimes, these two approaches combine, called a hybrid recommender system (Sarwar et al., 2001, Burke, 2007). Collaborative filtering recommendation (also called collaborative recommendation, or CR) has been widely used (Schafer et al., 2007, Konstas et al., 2009, Anaya and Boticario, 2011) because products are not limited in scope and adapt to users in detail. (Kobsa and Wahlster, 1989) seems to be the first author applying user-adaptive modeling techniques during runtime in the expert system, which is an advanced form of CR. Kautz et al. (1997) first combined CR approaches with social networks, giving these approaches an important role in recommender systems. Inspired by their seminal research, numerous application systems were subsequently developed in social networks. Golbeck and Hendler (2006) developed FilmTrust, a website that integrates the Semantic Web-based social networks, augmented with trust, to create predictive movie recommendations. Zhang and Ackerman (2005) established their own expert locating recommender system, detailing the work necessary in fitting expert recommendations to a work setting to decrease an overload and to support people confronted with difficult problems. Yimam-Seid and Kobsa (2003) explored the expert finding problem in depth, reviewed and analyzed existing systems, and suggested a domain model that can serve as a framework for design and development decisions. Ge et al. (2011) enabled a CR system to collect massive amounts of travel and tours data and use these data to provide real-time decision making regarding travel and tours recommendations. Dong et al. (2011) used a service concept recommendation system to find the connectivity between a customer's request and a concept. We can find collaborative mechanisms even in mobile communities (Rodriguez-Covili et al., 2011). However, many newly established CR systems in social networks have a cold-start problem, because the Internet contains information on millions of users and products, whereas information on their relationships e.g., likes/dislikes is rare. Users may be sensitive to privacy issues (Kobsa and Schreck, 2003); thus, they might only provide their name (name is the ID a system-registered user, e.g., username or pen name) and age (age is the information that a user would most likely disclose compared with other information) as typical data. The issues described above all contribute to the cold-start problem (Schein et al., 2002, Ahn, 2008), which is a major recommender system drawback that affects user confidence, and only a few effective researches have been conducted regarding this issue in social networks.
In this paper, we propose a modified social collaborative recommendation method based on complex clustering to cluster both users and items. This novel modified method is referred to as div-clustering, which aims to provide collaborative recommendations suited to user interests with the help of active users. We hypothesize that users in a cluster will have similar interests; thus, if an item is attractive to one user, then the other items in the same cluster will also be suitable. We also select active users from each user cluster to accelerate the performance of our recommendation system. To attract users to take an active part in the recommendation system, we set up an anonymous platform. Clustering users and items leads to more accurate recommendations, and the preservation of anonymity during registration frees users of privacy concerns. Furthermore, we conducted experiments to test the performance of div-clustering in offline and online modes, and we compared our proposed collaborative method with a baseline collaborative method that lacked the data clustering step. The test datasets were movies and academic papers. The former were collected from two famous online movie databases, MovieLens and IMDB, and the latter were mainly derived from academic websites related to conferences and journals, e.g., Elsevier, IEEE, Springer, which had already been crawled previously. The experimental results indicated that our proposed div-clustering method performed better in terms of accuracy compared with the baseline method.
The remainder of this paper is organized as follows. In Section 2, the basic methods and flaws of the current recommender systems are provided. Section 3 introduces the graphical model of div-clustering in our recommender system. The core concept of div-clustering and the rigorous process used for finding active users are discussed in Section 4. Section 5 presents the experiments conducted to test the accuracy of div-clustering in offline and online environments and the comparison between our proposed recommendation system and a baseline system in testing accuracy and robustness. The experimental results and discussion are detailed in Section 6. Conclusions are provided in Section 7.
Section snippets
Related works
The development of new types of recommendation systems has accelerated the speed of social networks and information retrieval. A collaborative recommendation has gradually increased in popularity because of its high recommendation accuracy, and many recent papers have focused on this approach. This approach focuses on providing content that a user has not yet viewed previously that may be suited to their interests. If two users share similar interests, one user might also be interested in the
Graphical model for div-clustering
The entities found in social CR systems can be depicted as a graphical model for div-clustering (Fig. 1), showing the overall elements in the Web and their mutual relationships. This model can be represented as a network G=(V, E), where V represents a set of vertices and each vertex can either represent a user or an item e.g., a type of newly-released phone on the market, a news headline, a Chinese restaurant, or even a user who has just tweeted, and E represents the set of edges connecting
Div-clustering framework
We clustered entities, including users and items, to enhance the performance of the collaborative recommendation. In this section, we improved the most popular traditional clustering approach, that is, the k-means clustering, because this approach is concise and efficient. The algorithm is described as follows.
Experiment and evaluation
The current study evaluated a recommender system that recommends profile items such as movies and academic papers on the basis of popularity. The aforementioned methods were implemented in Java with Eclipse 3.7.2, and the datasets were stored in a MYSQL Sever 5.0 with Intel R 2.93 GHz CPU and 4 GB memory running on a Windows 7 operating system.
Experimental results and discussion
In the first part, the initial performance of DCRM was compared with that of BCRM. Volunteers were not required to finish any amount of work in the recommender system and had the option to search freely. The two recommender systems were used on March 23, 2012. The results for the first month showed no difference because the statistics in the experiment were not apparent initially. The experiment ran from April 30, 2012 to June 30, 2012 for a total of 60 days. User click times were monitored and
Conclusions
This study proposed a modified CR based on the div-clustering method, which clusters both users and items to find active users in each cluster accurately and solve the cold-start problem. The proposed system enhanced recommendation accuracy and speed. The experimental results show that the proposed method obtained more accurate results than those of the baseline system. This study accomplished a number of detailed works in the use of clustering to construct the recommender system and completed
Acknowledgments
This work was supported by the National Key Technologies R&D Program under grant 2012BAH54F04, the National Natural Science Foundation of China under grant 61003051, the Natural Science Foundation of Shandong Province of China under grants ZR2010FM033 and 2009ZRB019RW, and Shandong Distinguished Middle-aged and Young Scientist Encouragement and Reward Foundation under grant BS2009DX040.
References (35)
A new similarity measure for collaborative filtering to alleviate the new user cold-starting problem
Information Sciences
(2008)- et al.
Usage derived recommendations for a video digital library
Journal of Network and Computer Applications
(2007) - et al.
A service concept recommendation system for enhancing the dependability of semantic service matchmakers in the service ecosystem environment
Journal of Network and Computer Applications
(2011) User modeling via stereotypes
Cognitive Science
(1979)- et al.
A communication infrastructure to ease the development of mobile collaborative applications
Journal of Network and Computer Applications
(2011) - et al.
Content-free collaborative learning modeling using data mining
User Modeling and User-Adapted Interaction
(2011) Hybrid web recommender systems
Lecture Notes in Computer Science
(2007)- Ge Y, Liu Q, Xiong H, Tuzhilin A, Chen J. Cost-aware travel tour recommendation. In: Proceedings of the 17th ACM SIGKDD...
- Golbeck J, Hendler J. FilmTrust: movie recommendations using trust in web-based social networks. In: Proceedings of the...
- et al.
Data mining: concepts and technologies
(2001)
Evaluating collaborative filtering recommender systems
ACM Transactions on Information Systems
Refferral web: combining social networks and collaborative filtering
Communications of the ACM
Privacy through pseudonymity in user-adaptive systems
ACM Transactions on Internet Technology
User models in dialog systems
Cited by (14)
Adaptive knowledge push method of product intelligent design based on feature transfer
2024, Advanced Engineering InformaticsA microservice recommendation mechanism based on mobile architecture
2020, Journal of Network and Computer ApplicationsCitation Excerpt :An information filtering method is first offered to identity proper R&D projects as a candidate set (Xu et al., 2016). Literature (Wu et al., 2013) modified CR method called div-clustering is presented to cluster Web entities in which the properties are specified formally in a recommendation framework, with the reusability of the user modeling component considered. Literature (Zhu et al., 2017) proposes a semantical pattern and preference-aware service mining method called SEM-PPA to make full use of the semantic information of locations for personalized POI recommendation.
Social network data to alleviate cold-start in recommender system: A systematic review
2018, Information Processing and ManagementCitation Excerpt :It is created from both history of the old users of the system and the type of relationship presumed between them. Wu, Wang, Peng, and Li (2013) develops a RS based on the div-clustering method, which groups users and items according to their similarity, using the K-means algorithm to solve the cold-start issue. The recommendations are made considering the characteristics of the cluster which the user belongs to and the characteristics of the clusters associated with the items the user preferred in the past.
DDSE: A novel evolutionary algorithm based on degree-descending search strategy for influence maximization in social networks
2018, Journal of Network and Computer ApplicationsCitation Excerpt :Social networks can generate massive data so that we need the big data technologies (Lv et al., 2017) to handle the issues including data types, storage models, data privacy and security. More and more researchers are focusing on using machine learning techniques (Meng et al., 2016) to analyze social network data and discover models and patterns (Zhang et al., 2017) of the people's behaviors in social networks, such as crowd-sensing(Sun et al., 2014) and friends or purchase recommender systems (Wu et al., 2013; Wang et al., 2015). Previous researches show that the “word-of-mouth” effect (Bass, 1976; Brown et al., 1987; Domingos et al., 2001; Goldenberg and Libai, 2001; Goldenberg et al., 2001; Mahajan et al., 1990; Richardson et al., 2002) plays a very important role in spreading innovations and ideas in social networks.
Collaboration computing technologies and applications
2013, Journal of Network and Computer ApplicationsApproaches and algorithms to mitigate cold start problems in recommender systems: a systematic literature review
2022, Journal of Intelligent Information Systems