Hybrid Weighted Retrieval of Twitter Users for Temporally Relevant Full-Text Querying in the Media Industry

Abstract

Barriers to the delivery of journalistic content to suitable media outlets present difficulties to both journalists and publishing houses. These may take the form of barriers to the identification of key individuals to whom the content is relevant, and who may be influential within their domain topic. To address this problem, we present a methodological approach to automated recommender systems for journalists or contentwriters, which enables the identification of social media accounts, based upon information retrieval techniques. Firstly, this work investigates three concepts, which to our knowledge have yet to be applied to such a domain; firstly, how user-defined descriptions, commonly associated with user profiles on the Twitter platform, may impact upon the perceived accuracy of recommendations. Secondly, the application of a full-text query facility to accept journalistic-content queries. Finally, an investigation into timedecay, as part of the recommendation-ranking methodology. This results in a proposed novel hybrid weighting methodology, which provides advantages to baseline information retrieval techniques, and demonstrates the influence of temporal decay ranking upon model performance. As part of this work, a vectorised tweet dataset is provided, entailing 390,750 unique tweets, for use in wider research. Finally, we demonstrate an implementation in a real-world commercial setting, and provide an anonymised data set to assist in future research into this domain.