Dynamic Social Network Mining: Issues and Prospects

Introduction

During recent years, the steadfast popularity increase of social networks and online communities has caught the attention of the data mining research community. Since their introduction, social network sites have attracted millions of users, many of whom have integrated these sites into their daily practices. The possibility to easily share multimedia content, thoughts, and electronic documents opens room to a novel form of communication and gives access to a huge amount of shared information.

Since social data coming from social networks, blogs, twikis, and bookmarking sites has become a powerful source of knowledge, the in-depth analysis of both the social network structure and the published content may trigger many insightful actions. For instance, the study of common social user behaviors may significantly improve the effectiveness of e-commerce services. Moreover, the analysis of the annotations made by Web users may drive product recommendation as well as ease Web resources’ retrieval.

The application of Data mining and Knowledge Discovery (KDD) techniques to social data has become more and more attractive for research purposes. For instance, Heymann et al. (2008) discovered relevant tag associations to model the metadata associated by Web users to public resources. The discovered tag associations are also used to drive the annotation process of partially annotated resources. In parallel, social tagging systems have also been used to classify videos and photos in order to improve the performance of search engines (Yin et al., 2009). At the same time, Wikipedia content has been used to infer semantics-based models, such as ontologies and taxonomies (Suchanek et al., 2008).

Data mining approaches entail either modeling peculiar characteristics of the analyzed data or predicting the label/value to assign to a given object/variable given a set of labeled data. The large variety of different data mining techniques which may be applicable to social data prompts the need of better understanding (i) the objective of the most recently proposed algorithms, (ii) their suitability for being applied in dynamic contexts, and (iii) the usefulness of the discovered knowledge for in-depth analysis. Peculiar aspects that should be taken into account by data mining analysts are, for instance, the algorithm efficiency in terms of computational time, the average prediction accuracy, and the succinctness and readability of the generated models. Since these properties actually depend on the social context under analysis, choosing the most suitable data mining approach is a challenging task. Furthermore, since social data continuous evolve over time, data mining algorithms are often required to be easily adaptable to dynamic contexts.

Interested readers may look into the large body of survey papers or books regarding social network analysis and mining (Liu, 2007; Scott & Carrington, 2011; Bonchi et al., 2011). They describe the characteristics of real-life social Web systems (Scott & Carrington, 2011), the applicability of intelligent data mining and information retrieval techniques (Liu, 2007), as well as they provide an in-depth insight into the most challenging research directions in social data analysis (Bonchi et al., 2011). However, a systematic classification and overview of the most recent data mining approaches applied to data coming dynamic social networks and online communities are still missing.