GKS: Graph-based Knowledge Selector for Task-oriented Dialog System

In previous research, knowledge-selection tasks mostly rely on language model-based methods or knowledge ranking. However, while approaches that rely on the language models take all knowledge as sequential input, knowledge does not contain sequential information in most circumstances. On the other hand, the knowledge-ranking methods leverage dialog history and each given knowledge snippet separately, but they do not consider information between knowledge snippets. In the Tenth Dialog System Technology Challenges (DSTC10), we participated in the second Knowledge-grounded Task-oriented Dialogue Modeling on Spoken Conversations. To deal with the problems mentioned above, we modified training methods based on state-of-the-art (SOTA) models for the first and third sub-tasks. As for the second sub-task of knowledge selection, we proposed Graph-Knowledge Selector (GKS), utilizing a graph-attention base model incorporated with the language model. GKS makes knowledge-selection decisions in the dialog by simultaneously considering each knowledge embedding generated from the language model without sequential features. Moreover, GKS leverages considerable knowledge in decision-making and takes relations across knowledge as part of the selection process. As a result, GKS outperforms several SOTA models proposed in the data-set on knowledge selection from the Ninth Dialog System Technology Challenges (DSTC9).

Towards Generating Citation Sentences for Multiple References with Intent Control

Machine-generated citation sentences can aid automated scientific literature review and assist article writing. Current methods in generating citation text were limited to single citation generation using the citing document and a cited document as input. However, in real-world situations, writers often summarize several studies in one sentence or discuss relevant information across the entire paragraph. In addition, multiple citation intents have been previously identified, implying that writers may need control over the intents of generated sentences to cover different scenarios. Therefore, this work focuses on generating multiple citations and releasing a newly collected dataset named CiteMI to drive the future research. We first build a novel generation model with the Fusion-in-Decoder approach to cope with multiple long inputs. Second, we incorporate the predicted citation intents into training for intent control. The experiments demonstrate that the proposed approaches provide much more comprehensive features for generating citation sentences.