Community Structure in Audio Clip Sharing

Abstract

Abstract-This paper describes the application of several network analysis techniques to the networks of users in audio clip sharing sites. Two kinds of networks are described: one based on the file downloading activity, and the other based on semantic similarity in the use of free tags to annotate sounds. The properties of these networks as well as their community structure are analyzed using data from Freesound, a very popular audio clip sharing site. Both are characterized as "small worlds". The component structure of the file downloading network provides a general overview of the user community which resembles the structure of the world wide web. On the other hand the analysis of the community structure of the semantic network using modularity optimization appears to be a good way to identify the different semantic areas in the site.

Small world networks and creativity in audio clip sharing

“…For the rest of networks, only active users appear as nodes in the network. This contrasts with previous analysis of these networks where we have statically considered the networks accumulated up to a certain point in time (Roma and Herrera, 2010). For the purposes of this paper, our experiments with 'accumulated' networks (where all past events are considered at each sample) yielded similar results at much higher computational cost, but the extracted measures suffered from high multi-collinearity, which made them unfeasible for regression analysis.…”

Extending Sound Sample Descriptions through the Extraction of Community Knowledge

“…The search process is either based on raw tags attached to sound samples (without any post-processing for "cleaning" or classifying these tags) or in acoustic audio similarity (using automatic feature extraction). Freesound does not support the explicit grouping of users into communities, nevertheless it has been observed that different implicit communities exist related to the interests of users [8]. As a consequence, Freesound turns out to be a highly heterogeneous database in types of sounds, their descriptions and user interests.…”

“…As a first step, we will focus in one of the implicit communities of interest found in Freesound [8]. Applying the Actor-Concept-Instance model described in [7], we expect to come upon a lightweight ontology emerging from the community and resembling the most important used tags -and their relations -for the description of sounds.…”

Characterization of the Freesound online community

“…We have published some articles on various aspects of Freesound [8] [9] but we have never performed a general characterization of the community with which to identify its strengths and weakness, and with which to identify appropriate future directions for design and development.…”

“…In parallel with the core-periphery structure, both networks exhibit high modularity as measured using the "louvain" method [18] (0.49 for the forum network, 0.57 for the comments network), meaning that user interactions tend to form more densely connected sub groups related with different interests [8].…”