Elsevier

Knowledge-Based Systems

Volume 24, Issue 8, December 2011, Pages 1277-1296
Knowledge-Based Systems

An approach to providing a user of a “social folksonomy” with recommendations of similar users and potentially interesting resources

https://doi.org/10.1016/j.knosys.2011.06.003Get rights and content

Abstract

“Social folksonomies” can be regarded as new-generation folksonomies obtained by empowering classical folksonomies with some features typical of social networks. Currently, they represent the natural evolution of traditional folksonomies since the presence of social network features in them could lead to an enormous increase of their performance. In order to efficiently and effectively handle this kind of folksonomies, new approaches appear compulsory because the simple extension of the ones operating on classical folksonomies appears incapable of fully capturing the new potentials of these entities. In this paper, we first illustrate a new approach that fully exploits the “social” features of these folksonomies to provide a user with recommendations of similar users and resources. Then, we present some experiments devoted to measure its performance. Finally, we compare it with various related ones already proposed in the literature.

Highlights

► Investigation of the role of the friendship relationship in folksonomy. ► Introduction of the concept of social folksonomy. ► A graph-based model to represent a social folksnomy. ► A technique to find and recommend similar users in a social folksonomy. ► A technique to find and recommend potentially interesting resources in a social folksonomy.

Introduction

Folksonomies and social networks are two of the most relevant phenomena that are currently characterizing the Web 2.0 scenario [31].

The term “folksonomy” is derived from the combination of the terms “folk” and “taxonomy”. It is used to denote a classification system obtained by collaboratively creating and managing tags to annotate a set of resources [46]. A folksonomy allows for both the categorization and the retrieval of available resources [14], [27], [34], [39], [44], [40]. Popular examples of folksonomies are Flickr [4] (which allows its users to annotate their photos), Del.icio.us [2] (which allows its users to store and share their Web bookmarks), Bibsonomy [1] (which allows its users to share bibliographic data on scientific papers), CiteULike (which allows its users to share, organize and store scholarly papers by assigning tags), Last.fm [5] (which allows its users to tag artists, songs or albums). Folksonomies are gaining a wider and wider popularity not only on the Web but also in large organizations and businesses [18], [37].

The term “social network” is used to denote a social structure formed by individuals connected to each other by one or more kinds of interdependency, such as friendship, common interests, knowledge, trustworthiness, and so forth. Generally, a social network can be represented through a graph whose nodes denote the individuals and whose edges denote the relationships between them. A social network allows its users to set up their profiles (generally storing both their biographical data and their interests) and to engage a broad range of activities, such as interactions with other users, resource posting, access and evaluation, and so on. Some popular social networks are Facebook [3] and MySpace [7] (which are general purpose ones), LinkedIn [6] (which is devoted to job search), ResearchGate [8] (which allows researchers to interact with each other, to share their results, etc.). Social networks are becoming more and more pervasive on the Web. A European Commission’s report [42] indicates that, from June 2007 until June 2008, the number of social network users in Europe grew by 35%, and in March 2008 the 57% of Internet users were members of at least one social network.

At the beginning folksonomies and social networks arose quite independently each other; however, they are becoming more and more strictly related. Indeed, currently, many folksonomies tend to favor the interaction among users and, vice versa, many social networks tend to favor the annotation of available resources. As a matter of fact, some social media, such as YouTube [12] and Flickr itself, can be considered as special cases of folksonomies and, at the same time, as special cases of social networks.

The term “social folksonomy” is starting to be used to represent these new forms of social sites. Specifically, a “social folksonomy” can be intended as a folksonomy in which users are connected to each other by one or more kinds of interdependency; these last ones are those generally considered for social networks. The most common form of interdependency is friendship.

The integration of features typical of social networks into a folksonomy can be highly beneficial for this last one. In fact, it allows a quicker information exchange and dissemination, and this leads to the construction of more refined user and resource profiles, and, ultimately, to an improvement of the folksonomy usage.

Clearly, traditional techniques typical of social networks cannot be directly applied to a folksonomy; on the other hand, traditional techniques typical of folksonomies appear not well suited to be directly extended to this new application scenario. For this reason, the development of new techniques appears compulsory.

This paper provides a contribution in this setting; in fact, it proposes: (i) an approach to providing a social folksonomy user with recommendations of similar users and potentially interesting resources and (ii) a support model which represents and organizes the information concerning a social folksonomy (which is the classical one represented by models operating in folksonomy and social network application contexts) in such a way as to facilitate the execution of the activities performed by our approach.

Our social folksonomy model is graph-based. Specifically, it represents a social folksonomy as a hypergraph. The nodes of this hypergraph represent users, resources and tags involved in the social folksonomy. Its hyperedges represent various forms of relationships among the involved actors. Specifically, there are six kinds of hyperedge.

The first four kinds correspond to the actions that users typically perform inside folksonomies; these actions are resource posting, resource labeling, resource access and folksonomy querying. The fifth kind represents user friendship; this information plays a key role in social networks and, as discussed above, in social folksonomies. Finally, the last kind indicates tag synonymies. The management of semantic relationships is a further important characteristic of our model which is provided with mechanisms to store and handle synonymies and homonymies possibly existing among tags. These mechanisms are extremely important in the folksonomy application field [27], [22], [43], [49], [23].

The proposed approach proactively provides a social folksonomy user ui with recommendations of similar users and potentially interesting resources. For this purpose, it first exploits information stored in the underlying model to construct user and resource profiles. The profile of a user ul consists of both a “direct” and an “indirect” component. The former can be derived by the examination of the actions performed by ul in the social folksonomy or of the actions performed by other users on the resources posted by ul. The latter can be derived by the examination of the profiles of the users belonging to the neighborhoods of ul; the closer these users, the higher the weight of their contribution. This last intuition represents the application of the notion of regular equivalence [32] in this context. This notion is largely accepted in the literature to detect the similarity of a pair of objects [15], [29], [32]. The profile of a resource rj is constructed by applying an analogous reasoning.

Once all user and resource profiles have been constructed, our approach applies a particular operator to compute the similarity between each of them and the profile of ui. Then, it recommends to ui the most similar users and resources.

Provided recommendations can be suitably exploited by ui, if he considers them relevant. For instance, he could contact recommended users to share resources, to start a friendship relationship, and so on; analogously, he could access or label recommended resources, and so forth. All activities performed by ui are registered in the underlying social folksonomy model; as a consequence, both his profile and the profiles of the involved users and resources are improved.

The plan of this paper is as follows: in Section 2 we illustrate our social folksonomy model; in Section 3 we present our recommendation approach; in Section 4 we illustrate some experiments we have carried out to test the performance of our approach; in Section 5 we examine related approaches and highlight the similarities and the differences between each of them and our own; finally, in Section 6, we draw our conclusions and examine some possible future developments.

Section snippets

The proposed social folksonomy model

As pointed out in the Introduction, we have defined a support model to represent and organize the information concerning a social folksonomy. Our model uses the “triadic” representation of a folksonomy (i.e., it considers relationships among users, tags and resources). Specifically, it represents a set of actions performed by users, as well as user friendships, typical of the social network context, and semantic information concerning tags.

It is worth pointing out that part or all of this

Recommendation of similar users and potentially interesting resources

As pointed out in the Introduction, the main purpose of our approach is to provide a user ui of a social folksonomy SF with recommendations of similar users and potentially interesting resources.

Our approach is based on the construction of “enhanced” profiles of involved users and resources and on the next comparison of them. As a consequence, in the next subsections, first we illustrate the strategy to construct “enhanced” user profiles, then we extend this strategy to the construction of

Experiments

In this section, we illustrate the experiments we have performed to evaluate our approach. These have been conceived to:

  • Analyze the role of the direct and the indirect components of user and resource profiles in the behavior of our approach;

  • Analyze the role of user friendship in the folksonomy building;

  • Analyze the capability of our approach of producing high quality recommendations.

In order to carry out these analyses, we built a prototype in Java and MySQL. We carried out all the experiments

Related work

In this section we examine some approaches conceived to produce recommendations to folksonomy users and highlight their similarities and differences with respect to our approach. An interesting analysis about classical approaches concerning folksonomy-based recommendation (with a particular emphasis on Information Retrieval based approaches) can be found in [39].

In [33] the authors propose an approach to increase the search precision in Flickr by exploiting the corresponding user-added

Conclusions

In this paper we have proposed a new approach to providing a social folksonomy user with recommendations of similar users and potentially interesting resources and a new social folksonomy model to support this approach.

We have seen that social networks and folksonomies, which are two of the most important phenomena of Web 2.0, are rapidly converging and are giving rise to a new kind of social system. As a matter of fact there already exist some examples of social media, like YouTube and Flickr,

References (51)

  • Last.fm, 2011....
  • LinkedIn, 2011....
  • MySpace, 2011....
  • Research Gate, 2011....
  • Simple Knowledge Organization System (SKOS), 2011....
  • The Friend of a Friend (FOAF) Project, 2011....
  • The Semantically-Interlinked Online Communities (SIOC) Project, 2011....
  • YouTube, 2011....
  • S. Amer-Yahia et al.

    From Del.icio.us to X.QUI.SITE: recommendations in social tagging sites

  • S. Bao et al.

    Optimizing Web search using social annotations

  • V.D. Blondel et al.

    A measure of similarity between graph vertices: applications to synonym extraction and web searching

    SIAM Review

    (2004)
  • P. Carrington et al.

    Models and Methods in Social Network Analysis

    (2005)
  • B. Cripe, Folksonomy, Keywords, and Tags: Social and Democratic User Interaction in Enterprise Content Management,...
  • P. De Meo et al.

    Finding reliable users and social networks in a social internetworking system

  • P. De Meo et al.

    Recommendation of reliable users, social networks and high-quality resources in a social internetworking system

    AI Communications

    (2011)
  • Cited by (18)

    • Characterizing context-aware recommender systems: A systematic literature review

      2018, Knowledge-Based Systems
      Citation Excerpt :

      Based on the assertion that the relevance of a recommended item depends not only on its correctness, but also on its novelty. Nocera et al. [66] define an ad-hoc measure that takes into account whether the recommended items were already known to the user (e.g. accesed in the past). This property measures the proportion of items that the system recommends from the universe of available items.

    • Neighborhood-user profiling based on perception relationship in the micro-blog scenario

      2015, Journal of Web Semantics
      Citation Excerpt :

      Interactions among users in social networks also attracted a lot of attention [16]. By exploiting the “social” features of social folksonomies, Nocera [31,32] provided an “enhanced” user with recommendations of similar users and potentially interesting resources based on involved users and resources. Taking into account relations between friends in a large-scale real-world social network, Zhou [33] found that semantic web community evolves in a speckled fashion which is a highly distributed form.

    • A smart TV system with body-gesture control, tag-based rating and context-aware recommendation

      2014, Knowledge-Based Systems
      Citation Excerpt :

      Second, we compare different computational methods for making personalized recommendations on multimedia items. Following the current trend of community-based information sharing [11–15], we also integrate social tags to annotate multimedia items to improve recommendation performance. The experimental results show that the tag-based method outperforms other conventional methods.

    • A collaborative filtering approach to mitigate the new user cold start problem

      2012, Knowledge-Based Systems
      Citation Excerpt :

      CF based RS allow users to give ratings about a set of elements (e.g. hotels, restaurants, tourist destinations, etc. in a CF based website), in such a way that when enough information is stored on the system we can make recommendations to each user based on information provided by those users we consider to have the most in common with them. Movie recommendation websites are probably the best-known cases to users and are without a doubt the most thoroughly studied by researchers [25,2], although there are many other fields in which RS have great and increasing importance, such as e-commerce [23,57,52,36], e-learning [12,4,21], music [15,31], digital libraries [40], playing games [32], social networks based collaborative filtering [13,35], etc. A key factor in the quality of the recommendations obtained in a CF based RS lies in its capacity to determine which users have the most in common (are the most similar) to a given user.

    View all citing articles on Scopus
    View full text