A review of features for the discrimination of twitter users: application to the prediction of offline influence

Abstract

Many works related to Twitter aim at characterizing its users in some way: role on the service (spammers, bots, organizations, etc.), nature of the user (socio-professional category, age, etc.), topics of interest, and others. However, for a given user classification problem, it is very difficult to select a set of appropriate features, because the many features described in the literature are very heterogeneous, with name overlaps and collisions, and numerous very close variants. In this article, we review a wide range of such features. In order to present a clear state-of-the-art description, we unify their names, definitions and relationships, and we propose a new, neutral, typology. We then illustrate the interest of our review by applying a selection of these features to the offline influence detection problem. This task consists in identifying users which are influential in real-life, based on their Twitter account and related data. We show that most features deemed efficient to predict online influence, such as the numbers of retweets and followers, are not relevant to this problem. However, we propose several content-based approaches to label Twitter users as Influencers or not. We also rank them according to a predicted influence level. Our proposals are evaluated over the CLEF RepLab 2014 dataset, and outmatch state-of-the-art methods.

Automatic detection of influencers in social networks: Authority versus domain signals

“…The vocabulary in the tweets is enough to learn domain and authority signals. Our best result was obtained with L2R over domain and authority signals learned from the training data (0.74), which also outperforms the best published result on the data set (0.71; Cossu et al, 2016). The role of the domain signal, however, is only marginally relevant: the authority signal alone provides 0.73 (−1.3%) without using any machine learning (ML) algorithm.…”

“…Our best result was obtained with L2R over domain and authority signals learned from the training data (0.74), which also outperforms the best published result on the data set (0.71; Cossu et al, 2016). The vocabulary in the tweets is enough to learn domain and authority signals.…”

“…• Our best result was obtained with L2R using the output of our domain and authority topic models (0.74), which also outperforms the best published result on the data set (0.71; Cossu et al, 2016). The number of followers might be an initial filtering criterion (no one can be influential without an audience), but for users with more than 1,000 followers (all candidates in the data set meet this criterion), the textual content signal is significantly more powerful than the number of followers and any other Twitter authority indicators (number of retweets, favorites, ratio of followers/followees, and so on).…”

“…In Cossu, Labatut, and Dugué (2016), seven categories where the features can be grouped are introduced: user profile, publishing activity, local connections, user interaction, lexical aspects, stylistic traits, and external data. The first one is text sparsity due to Twitter's length restriction (140 characters) appearing in many infrequent terms, some of them being considered noise.…”

“…4. Best result published to date that uses the RepLab 2014 data set (Cossu et al, 2016). Table 1 summarizes the results (in terms of MAP) of all experiments using a simple approach to handle textual content (textual features are a BoW extracted from the author posts).…”