Expanding upon our prior research endeavors, our study delves into the dynamics of communal support towards migrants and refugees in the wake of transformative events, focusing specifically on the period before and after the emergence of the COVID-19 pandemic. We achieve this through the meticulous collection and examination of an extensive and original dataset composed of tweets associated with migration. Our primary objective entails the evaluation of alterations in social cohesion and solidarity expressed towards migrants during these distinct temporal phases. To initiate this investigation, we meticulously assess a corpus of more than 2000 tweets, deciphering the presence of either supportive or adverse sentiments towards immigrants. This is accomplished by employing two distinct approaches for annotation: one reliant on the expertise of individuals and the other soliciting contributions from the general public. Building upon these annotated tweets, we develop a Long Short-Term Memory (LSTM) model, enriched by a multitude of data augmentation strategies. Impressively, the performance of this model approaches the upper echelons of human accuracy. This finely-tuned model serves as the foundation for the subsequent automated labeling of over 240,000 tweets spanning the period from September 2019 to June 2021. Through a meticulous analysis of these automated labels, we elucidate the evolving landscape of migrant- oriented sentiments over this critical period, encompassing both the prelude and the aftermath of the COVID-19 outbreak. Notably, our findings underscore the escalating prominence and contentiousness of expressions of solidarity towards migrants during the initial phases of the pandemic. However, as the timeline progresses, this solidarity seems to recede in significance, with a slight dip in tweet volume below levels observed in more standard contexts by the summer of 2021. Interestingly, a subset of tweets linked to the COVID-19 crisis displays an elevated proportion of sentiments opposing solidarity. In a significant contribution to the field, we also address the potential pitfalls associated with scrutinizing trends in social cohesion. For instance, we underscore how the balance between solidarity-affirming and anti- solidarity sentiments can be influenced by various factors, such as the choice of sampled tweets, the linguistic characteristics of these tweets, the national identification of Twitter users (whether known or anonymous), and the meticulous selection of pertinent tweets.

Keywords : Solidarity, Crisis, Social Media, Long Short- Term Memory(LSTM), Random Forest, Social Dynamics, Trend Analysis, Societal Responses, Anti-Solidarity.