Executive Summary
Key findings
• Refugee-host community issues are widely discussed on social media in Lebanon and analysis of these discussions provides significant insight that can complement offline monitoring.
Social media has become a ‘magnified mirror’ of societal tensions in Lebanon and appears to have a real impact on perceptions of refugeehost community issues.
• Key methodological challenges to social media analysis of refugee-host community tensions include access to APIs, Arabic language in automated data collection and sentiment analysis. The report proposes ways to manage these challenges.
• Facebook and Twitter are both popular platforms for discussions, although there are important differences between them. Facebook reaches a broader subset of the Lebanese population, while Twitter serves predominantly political and intellectual elites. Facebook is a space for less filtered commentary and greater interactions between users, whilst Twitter is used predominantly for news bites, uses more formal language and is the site of less direct conversations between users.
• Across these conversations, an ongoing general discourse about Syrian refugees can be readily identified. Issues of return, assistance and crime and violations emerge as key topics. On the whole, assistance provoked more supportive sentiment, whilst return provoked a greater spread of supportive and antagonistic comments.
Crime and violations can provoke both reactions, depending on the nature and direction of the crime and violations.
• Whilst coding for sentiment directed towards and from the Syrian refugee community, the majority of data contained attitudes expressed by Lebanese towards Syrian refugees. This reflects the nature of the conversation more broadly, where sentiments expressed by Syrians towards Lebanese appear less often in public fora and are predominantly restricted to private WhatsApp conversations.
• Sentiment on social media appears to spike in response to 1) overall political discourse (eg a Tweet or a public statement by a major political figure); 2) macro-level events that affect the refugee community (eg Storm Norma) and 3) smaller scale incidents involving Syrian individuals (eg an individual crime).
• Where spikes in sentiment (supportive or antagonistic) do occur, they do not represent the absolute level of sentiment among social media users. Rather, they reflect an awakening of individual users’ pre-existing sentiments, triggered by the three causes above.
• Whilst responses to events provide interesting insights into sentiment discourse, it is not enough to understand only conversations responding directly to events. To gain a full picture, it is necessary to see the whole range of conversations happening in a given time period. For example, whilst direct responses to Storm Norma were overwhelmingly supportive on Twitter, antagonistic sentiment relating to broader issues of return spiked in the immediate aftermath of the Storm, affecting the overall discourse.
• Whilst this data provides an accurate temperature gauge of offline sentiments, it is very difficult to draw predictive conclusions from the data. In this period, online conversations served to mirror offline events, rather than to trigger them.
• Amid the Twitter conversations, five distinct communities emerged. These communities each operate within their own sphere of influence. The English-speaking community, which includes the majority of UN agencies and INGOs, does not exert much influence over the domestic Lebanese audience.
• Polarization between several Twitter communities appears to be occurring on the basis of political leaning, rather than on the basis of attitudes towards refugees.
• Each of these communities responded differently to different events, demonstrating distinct interests.