Towards Complex Scenarios: Building End-to-End Task-Oriented Dialogue System across Multiple Knowledge Bases

With the success of sequence-to-sequence model, end-to-end task-oriented dialogue systems (EToDs) have obtained remarkable progress. However, most existing EToDs are limited to single KB settings where dialogues can be supported by a KB, which is still far from satisfying requirements some complex applications (multi-KBs setting). In this work, we first empirically show that single-KB fail work on multi-KB require models reason across various KBs. To solve issue, take step consider multi-KBs scenario in and introduce KB-over-KB Heterogeneous Graph Attention Network (KoK-HAN) facilitate model over multiple The core module triple-connection graph interaction layer different granularity levels information KBs (i.e., intra-KB connection, inter-KB connection dialogue-KB connection). Experimental results confirm superiority our for reasoning.