Interest in emotion recognition in conversations (ERC) has been increasing in various fields, because it can be used to analyze user behaviors and detect fake news.Many recent ERC methods use graph-based neural networks to take the relationships between the utterances of the speakers into account.In particular, the state-of-the-art method considers self-and inter-speaker dependencies in conversations by using relational graph attention networks (RGAT).However, graph-based neural networks do not take sequential information into account.In this paper, we propose relational position encodings that provide RGAT with sequential information reflecting the relational graph structure.Accordingly, our RGAT model can capture both the speaker dependency and the sequential information.Experiments on four ERC datasets show that our model is beneficial to recognizing emotions expressed in conversations.In addition, our approach empirically outperforms the state-ofthe-art on all of the benchmark datasets.

Every human processes a set of mental schemas for problem solving. We develop and improve these schemas by reflecting on our experiences with errors, which is a type of metacognition (Kayashima, 2008). In this study, we proposed a cognitive model of this "self-reflection" process based on Kayashima's two-layer working memory model, and developed a food recommender system using our cognitive model. In the test simulation, the users were satisfied with the foods that the system recommended, although the recommendation results were unexpected to the users. This implied the system practically worked to satisfy the user's expectation. On the other hand, the candidate recommendations which the system selected as its final output were different from those provided by the users. This suggests that the cognitive model needs improvement in terms of psychological reality.