HyKnow: End-to-End Task-Oriented Dialog Modeling with Hybrid Knowledge Management

Abstract

Task-oriented dialog (TOD) systems typically manage structured knowledge (e.g. ontologies and databases) to guide the goal-oriented conversations. However, they fall short of handling dialog turns grounded on unstructured knowledge (e.g. reviews and documents). In this paper, we formulate a task of modeling TOD grounded on both structured and unstructured knowledge. To address this task, we propose a TOD system with hybrid knowledge management, HyKnow. It extends the belief state to manage both structured and unstructured knowledge, and is the first end-to-end model that jointly optimizes dialog modeling grounded on these two kinds of knowledge. We conduct experiments on the modified version of MultiWOZ 2.1 dataset, where dialogs are grounded on hybrid knowledge. Experimental results show that HyKnow has strong end-to-end performance compared to existing TOD systems. It also outperforms the pipeline knowledge management schemes, with higher unstructured knowledge retrieval accuracy.

End-to-End Task-Oriented Dialog Modeling with Semi-Structured Knowledge Management

“…The structured query can facilitate the unstructured document retrieval by helping to identify the relevant entity and narrow down the document candidates. This is different from our prior work HyKnow [1] where structured and unstructured knowledge are operated independently of each other.…”

“…However, HyKnow operates the structured and unstructured knowledge independently of each other, and it implements the Seq2Seq modeling based on traditional GRU [13]. In our experiments, we test HyKnow's performance under the single-decoder belief state decoding scheme, denoted as HyKnow (Single) in our prior work [1].…”

“…It extends the belief state to handle TODs grounded on semi-structured knowledge, and further uses the extended belief state to perform both structured query and document retrieval, whose outputs are thereby used to generate the final response. As a further development of our prior work HyKnow [1], SeKnow fuses the management of structured and unstructured knowledge via their shared domains and entities. Through the knowledge management fusion, the query of structured data can facilitate the retrieval of documents.…”

“…Both implementations are in end-to-end frameworks, where dialog modeling grounded on structured and unstructured knowledge can be jointly optimized to get overall better performance. In SeKnow-S2S, following our prior work HyKnow [1], we apply two schemes of extended belief state decoding to investigate the correlation of structured and unstructured knowledge management. However, different from HyKnow which uses GRU [13] as backbone, we implement SeKnow-S2S based on Universal Transformer [14] instead, which yields better TOD modeling performance.…”