Help Me Understand This Conversation: Methods of Identifying Implicit Links Between CSCL Contributions

Multi-participant chat conversations are one of the most frequently employed Computer Supported Collaborative Learning tools due to their ease of use. Moreover, chats enhance knowledge sharing, sustain creativity and aid in collaborative problem solving. Nevertheless, the manual analysis of multi-participant chats is a difficult task due to the mixture of different topics and the inter-twinning of multiple discussion threads during the same conversation. Several tools that employ Natural Language Processing techniques have been developed to automatically identify links between contributions in order to facilitate the tracking of topics and of discussion threads, as well as to highlight key contributions in terms of follow-up impact. This paper proposes a novel method for detecting implicit links based on features computed using string kernels and word embeddings, combined with neural networks. This method significantly outperforms previous results on the same dataset. Due to its smaller size, our model represents an alternative to more complex deep neural networks, especially when limited training data is available as is the case of CSCL chats in a specific domain.