Facial sentiment and verbal content in harmony: theory and evidence

Identifying the sentiment in policy announcements helps analyze shifts in underlying stances and signals embedded in policy announcements. Previous studies have mainly focused on either text-based approaches or non-verbal cues in relation to housing policies, resulting in a limited comprehension of interactions between them. This study aims to examine the relationship between the facial sentiment of housing policy announcers and the quantitative features of their speech transcripts. We first utilize a deep learning model to extract the facial sentiment embedded in housing policy announcers. Subsequently, we calculate the scaling exponent and employ the controlled growth process model to extract a text cohesion from policy announcers’ transcripts. Lastly, we examine how both verbal and non-verbal cues interact with each other in terms of housing policy announcements. This study exhibits three notable findings: (1) the facial sentiment of policy announcers varies in accordance with the heterogeneity in policy stances, (2) the quantitative features of policy transcripts also reflect the heterogeneous policy stances; and therefore, (3) there is a macroscopic interaction between housing policy announcers’ facial sentiment and their verbal contents. Agents can benchmark the research framework to establish a structured approach to effectively analyze the far-reaching effects of housing policies. Policymakers can utilize our approach to quantitatively assess current policy stances by exploring the paths of implemented policies. Additionally, market analysts can leverage these insights to track policy consistency over time, improving their understanding of regulatory trends and enhancing their comprehension of the government’s policy trajectory.