Neighborhood-user profiling based on perception relationship in the micro-blog scenario
Introduction
Nowadays, as a newly emerged communication tool and public medium on the Internet, micro-blog spreads popular hot topics from one user to millions of individuals just in a few minutes, which allows the user to receive desired information anytime and anywhere. Meanwhile, searching for personalized interests and feelings posted to the multi-source information platforms, such as micro-blog systems like Twitter, social network sites including Facebook and LinkedIn, and personal homepages and blogs as well as many others [1], [2], is an interesting yet challenging task. Especially, in the micro-blog system platforms, people repost a lot of short messages about their daily activities and feelings so as to maintain latest interests or friendship.
Many researchers have successfully tested feasibility of applications in many areas including interesting topics [3] and micro-blog environments [4]. However, personal User Profile (UP) is a customized model of interest representing and reasoning for a user, which is implicitly contained and generated from one’s behaviors, browsing contents, or feedbacks [5], [6], [7]. That is, how to fulfill personalized activities and information requirements with one’s micro-blog user profile is an important yet challenging issue. Very little research, however, has been done on this issue.
In the micro-blog scenario, each micro-blog is short and lacks sufficient information for user profile construction. As is expected, a user profile is not only generated from individual short messages, but also profits from existing interactions of friends [8], [9], [10]. With more than 215 million users and more than 175 million postings per day in 2012, Twitter is one of the most prominent micro-blog services on the web [11]. In particular, most of users are used to forwarding tweets for communication, instead of directly posting. Hence, followees make an important role in the propagation and spread of personalized interests. Traditional user profiles capture personal interests over one’s own knowledge [12], [13], [14], [15], which are not holistic for discovering diverse information. In this case, items and products in user profiles could not reflect currently concerned subjects and socially propagative topics thoroughly.
In many scenarios, traditional collaborative filtering (CF) strategy provides users a lot of valuable information on the basis of mutual understanding and knowing. In social communities such as Facebook, LinkedIn and Twitter, the solution of CF is challenging. First, sparse data derived from short text is insufficient to capture enough similar users to recommend desired items, which hurt both the precision and recall of recommender systems. Second, pluralistic society makes people generate diverse interests. Not only are users restricted to daily monotonous interesting item, but they may be interested in diverse items posted by their friends [16]. A vast amount of diverse data enables similarity between users is small, which also leads the capture of similar users is hard. Meanwhile, a small amount of similar users is difficult to discover the items of the high correlation, which cannot effectively be applied into CF strategy [9]. Actually, when we follow what the followees have written, we can reflect our interests in a tracking way; and when we glimpse other followees or communities, we can realize where the interests come from. Follow friends’ knowledge is a kind of effective collective wisdom, which could extend personal interest to other latent but relevant subjects. Additionally, follow relation is a new back-to-back linkage, which can supply the target user diverse interests from collaborative users [9], [17]. It is reasonable that these follow friends contain a group of intimate interest users, named as neighborhood. Therefore, neighborhood with sufficient knowledge could help an individual user build the Neighborhood User Profile (NUP), addressing the problem of information shortage in representing personal interest.
In this work, we, using Sina micro-blog data source, constructed novel neighborhood user profile based on the collective knowledge. First, taking into account roles of followee friends in the interest propagation of the target user, we investigated the follow perception relationship and resource perception relationship. Furthermore, by adjusting the importance of two kinds of relationships, we discovered the neighborhood of a user. Lastly, the NUP relying on neighbor interests is proposed. In addition, the proposed NUP is evaluated by comparing against the existing personal UP and CF recommendation methods through experiments on large amounts of data from the Sina micro-blog platform.
Our experimental results show that the proposed NUP approach outperforms other methods in both precision and recall but with a relatively higher time complexity. By analyzing the expanded interests by NUP, we have observed that the recommendations based on NUP can accelerate the diffusion of the user interest, especially some semantically related interest between friends. We introduce the idea of neighborhood to solve the problem of acquiring behavior interest of social users. In particular, with the consideration of both the roles of followees’ friends and resource perception relationship equally, the selected neighborhood could expand semantic interest efficiently. When the neighborhood only includes oneself, the NUP becomes a conventional individual user profile. However, the zooming size of neighborhood is an important issue for interest supplement related to social networks and social Webs, which needs to leverage the adaptive diversification fusion algorithm for zooming-in and zooming-out of the neighborhood.
The remaining part of this paper is organized as follows. Section 2 briefly discusses the works related to user profile. In Section 3, an overview of our recommendation framework based on neighborhood user profile is presented. In Section 4, we introduce personal interest acquiring method. Concepts of neighborhood and detailed descriptions of interest extending in neighborhood user profile are presented in Section 5. In Section 6, we demonstrate the application of our system as well as our experiment results along with discussions on strength and limitations. Finally, we conclude in Section 7 directions of our future work.
Section snippets
A brief review of UP works
In the scenario of user profile construction, how to convert the raw micro-blog documents into user’s interesting subjects is usually challenging. To exactly recommend appropriate products to the user, many researchers have published their works in discovering demonstrated ways to build user profiles [18], [19], [20], [12], [13], [14], [10], [21], [22]. In this section, we will briefly review some popular works related to user profile.
Recommendation framework based on NUP
The NUP aims to discover a user’s personal interest and discover his/her neighbor interests from followee friends and their similar resources. Illustrated in Fig. 1 is the overview of recommendation based on neighborhood user profile. As shown in Fig. 1, the personal user profile is obtained by integration of content interest and semantic interest. Then, personal interests can be accumulated in terms of semantics of the ontology category structure in order to form Expanded User Profile (EUP).
Personal interest extraction
In this section, we first model the personal content interest with TF–IDF weighting mechanism from micro-blogs. Then, we identify personal UP by considering the semantics of subjects and expand latent subjects in terms of ontology category structure.
Interest extraction based on NUP
In this section, we present a novel approach to discover neighborhood of the target user, and the NUP is modeled by integrating neighbor EUPs into a personal EUP. Afterwards, characteristics of NUP are analyzed.
Application and experiment evaluation
In this section, we present some recommendation applications using the proposed NUP, and evaluate the performance of the NUP by comparing with personal UP and CF recommendations. Simultaneously, we share some insights from our observations and analysis of the NUP-based subject recommendation.
Conclusions
This paper investigates how neighborhood of a user can be used to help build a novel NUP in the micro-blog scenario by addressing the drawbacks of existing UP approaches immersing in use of personal knowledge. First, personal UP is constructed based on integrating the content interest and semantic interest, and then the improved EUP is created. Second, taking into account the roles of RPR and FPR, neighborhood of a user is discovered. Lastly, we propose a NUP modeling approach based on neighbor
Acknowledgments
We would like to thank all of the anonymous reviewers for their insightful and constructive comments and useful suggestions that have significantly improved the quality of the manuscript. This work was partially supported by the National Natural Science Foundation of China (Nos. 61303096, 61103067), and the Natural Science Foundation of Shanghai (13ZR1454600).
References (43)
- et al.
User interests modeling based on multi-source personal information fusion and semantic reasoning
Active Media Technol.
(2011) - et al.
Recommending multimedia web services in a multi-device environment
Inform. Syst.
(2013) - et al.
Keyword clustering for user interest profiling refinement within paper recommender systems
J. Syst. Softw.
(2012) - et al.
A diffusion mechanism for social advertising over microblogs
Decis. Support Syst.
(2012) - et al.
Extracting multilayered communities of interest from semantic user profiles: application to group modeling and hybrid recommendations
Comput. Human. Behav.
(2011) - et al.
Linked knowledge sources for topic classification of microposts: a semantic graph-based approach
Web Semant.
(2014) - et al.
An approach to providing a user of a “social folksonomy” with recommendations of similar users and potentially interesting resources
Knowl.-Based Syst.
(2011) - et al.
How is the semantic web evolving? a dynamic social network perspective
Comput. Human. Behav.
(2011) - et al.
A new temporal and social PMF-based method to predict users’ interests in micro-blogging
Decis. Support Syst.
(2013) Translation approach to portable ontology specifications
Knowl. Acquis.
(1993)
Learning user’s preferences by analyzing web browsing behaviors
Artificial Intelligence
Towards a tag-based user model: how can user model benefit from tags?
User Model.
Efficient personalization of e-learning activities using a multi-device decentralized recommender system
Comput. Intell.
Cited by (10)
Overlapping community detection in heterogeneous social networks via the user model
2018, Information SciencesCitation Excerpt :The approach of recommending microblogs based on the user model [27] does not provide insights into the social relations among user models. The research on theme recommendation in microblogging scenarios based on neighborhood-user profiles [37] merely binds the semantic interest expansions and the social relations of a user model. Twitter user profiling [18] adopts the community information only when the group demonstrates distinct characteristics, and detects clusters in the multi-hop network of the target user instead of the entire network, leading to loss of exact community information.
Attribute inference by link strength modeling in online social networks with user tags
2020, Journal of Internet TechnologySocial Profiling: A Review, Taxonomy, and Challenges
2019, Cyberpsychology, Behavior, and Social NetworkingFriend recommendation in social networks based on multi-source information fusion
2019, International Journal of Machine Learning and CyberneticsSense-Based Topic Word Embedding Model for Item Recommendation
2019, IEEE Access