Human-generated information underlies nearly every aspect of our lives. With the rise of affordable handheld and embedded IoT devices with new connectivity capabilities and the ever-increasing integration of many technologies with social media platforms, most social problem-solving processes are increasingly relying on online conversations and textual communication. However, unlike in a face-to-face situation, where we can fairly recognise which conversations are co-occurring around us and their participants, automatically identifying cohesive conversations online is a challenging task. The aim of this talk is to discuss how we can combine network-analytic methodologies with textual data to better infer conversational structures in social media. Rather than providing direct answers to this problem, we would like instead to use this opportunity to report our current strategies and efforts to disentangle multiple intertwined conversations around similar topics.