Twitter as a Research Tool and Object in Psychology: Applications in Scientometrics and Distributional Semantics

Background: The increasing use of digital services such as social media platforms has led to the emergence of new data sources for psychology research. In this context the micro-blogging site Twitter has gained prominence, both, as a research tool and object. Since the last decade, Tweets have been analyzed for various psychological research questions, predominantly from the fields of media psychology, clinical psychology, and educational psychology. However, the additional value of utilizing Twitter for getting a bird’s- eye view of psychological science or its potential to differentiate human meaning construction in the wake of recent events has not yet been addressed. Aim: In this talk, we aim at expanding prior application fields by presenting two current projects from our research group that employ Twitter as a scientific tool and object. As an example of big-data-enriched scientometrics, the first application leverages tweets for an early detection of trends in bibliometric data. The second application demonstrates the superiority of word embeddings as measures for differentiating and characterizing semantic meaning in tweets. The overarching goal of this talk is to gauge possibilities of Twitter-based research in the light of current technical restrictions and ethical considerations. Application 1 (Twitter as a research tool): “Detecting Psychological Research Trends using Twitter Mining”. For identifying psychological hot topics, traditional bibliometric analyses suffer from a publication delay. Thus, only topics at the very end of the research process (i.e., publication) can be detected. In contrast, Twitter is used by the research community in all phases of the research cycle. A recent proof-of-concept study gave evidence that text mining of tweets can indicate future publication trends earlier than using literature databases alone. Building on this, we present research-in-progress that examines predictive features of Twitter topics, aiming at closing the gap between the early detection of emerging topics and an implementation in research monitoring tools such as PsychTopics (https://psyndex.de/en/trends/psychtopics/). Application 2 (Twitter as a research object): “Using Word Embeddings to Understand Human Meaning Construction”. We present a conceptual and practical framework for exploiting embedding models (e.g., GloVe algorithm) as measures of semantic meaning. This is based on a recent work-in-progress in which we applied word embeddings to gain insights on the dynamics of user behaviour and psychological content of conspiracy-focused tweets on Twitter. The presentation discusses both the advantages and shortcomings of using traditional approaches like dictionaries or other vector models and proposes word embeddings as an option that overcomes several limitations. Finally, we present the main takeaways of how to use these distributed representations in applied work. Discussion: Against the backdrop of the prevalence of Twitter as data source, we highlight both opportunities and pitfalls of current Twitter-based research in psychology. In particular, we address challenges and biases in collecting and processing of tweets, alongside privacy and data sharing issues.