Datasets and Benchmarks for Task-Oriented Log Dialogue Ranking Task.

Although the data-driven approaches of some recent bot building platforms make it possible for a wide range of users to easily create dialogue systems, those platforms don't offer tools for quickly identifying which log dialogues contain problems. Thus, in this paper, we (1) introduce a new task, log dialogue ranking, where the ranker places problematic dialogues higher (2) provide a collection of human-bot conversations in the restaurant inquiry task labelled with dialogue quality for ranker training and evaluation (3) present a detailed description of the data collection pipeline, which is entirely based on crowd-sourcing (4) finally report a benchmark result of dialogue ranking, which shows the usability of the data and sets a baseline for future studies.