A Language-Independent Technique for Assessing Tweet Success: An Experience Report

Twitter is a very active social network with hundreds of millions of users. This huge number of users makes it a very important market, where companies need to participate in order to improve their business opportunities. In order to analyze data and promote contents many studies apply natural language approaches, which require libraries that are only available for widely spoken languages. However, it is not easy to adapt the results obtained to different products and contexts, since each culture has specific features that make them unique. For these reasons, a language-independent way to train systems to detect the main features required to write successful tweets in different contexts would be useful. In this paper, we propose five definitions for successful tweets. Once we have identified successful tweets with respect to these definitions we apply machine learning to build predictive models and extract those features that characterize them, so we can present a recipe for writing successful tweets following the most appropriate definition in each case. We have applied this approach to a data set of tweets obtained during the political events in Catalonia in October, 2017. Although the results are not completely satisfactory, we have been able to build good predictive models for one of the success definitions and extract from them some candidate features that make a successful tweet. Moreover, we identify the main problems with the rest of the definitions and discuss some improvements, so future research lines can take them into account.