The new age of internet social media provides us often with opportunities for collection and analysis of online social network data. Until recently, most social network data were analysed by classical social network analysis techniques originally designed for aggregated “snapshot-like” network data. The advent of time-stamped network-link data called for the development of statistical methods designed specifically for analysis of evolution of social networks. Here we provide a review of the available approaches for statistical analysis of evolution of social networks, and elaborate on our selection of one of these, namely the Tnet R framework, as our choice for our present and future dynamic network analyses and development. In this paper we present our preliminary research results from our study of network links of Australian advocacy groups on Twitter, a research funded by an Australian Research Council project. We first present the results of our study of the network/link evolution by taking into account structural network effects but also the attributes of network nodes (advocacy groups). Thereafter, we present preliminary results of our study of sentiments of these textual links.
In the first part we use a conditional logistic regression approach to discern processes that are acting to create the advocacy groups mention and reply Twitter networks. We study traditionally important structural effects of reciprocity and transitivity, but also those of the reinforcement, in-degree and in-strength effects, additionally incorporating effects for homophily based on attributes of advocacy groups. Our preliminary results indicate strong reciprocity, reinforcement and homophily effects. We also discovered significant closure and in-degree processes.
The second part of our research revolves around attributes of the network textual links themselves. We begin by classifying our network links as positive, negative or neutral, based on the text content of the links/tweets. In order to accomplish this we implement an algorithm for the sentiment analysis of tweets based on widely used opinion/sentiment dictionaries developed by (Hu and Liu, 2004), elaborating on the effectiveness of the algorithm’s coding of sentiments. We analyse tweet sentiments longitudinally by aggregating the sentiments on a daily basis. We also compare sentiments between different group-categories (attributes) of Australian advocacy groups.