We release Douban Conversation Corpus, comprising a training data set, a development set and a test set for retrieval based chatbot. The statistics of Douban Conversation Corpus are shown in the following table.

Train Val Test
session-response pairs 1m 50k 10k
Avg. positive response per session 1 1 1.18
Fless Kappa N\A N\A 0.41
Min turn per session 3 3 3
Max ture per session 98 91 45
Average turn per session 6.69 6.75 5.95
Average Word per utterance 18.56 18.50 20.74

The test data contains 1000 dialogue context, and for each context we create 10 responses as candidates. We recruited three labelers to judge if a candidate is a proper response to the session. A proper response means the response can naturally reply to the message given the context. Each pair received three labels and the majority of the labels was taken as the final decision.

As far as we known, this is the first human-labeled test set for retrieval-based chatbots. The entire corpus link https://www.dropbox.com/s/90t0qtji9ow20ca/DoubanConversaionCorpus.zip?dl=0

Data template

label \t conversation utterances (splited by \t) \t response


Paper Code Results Date Stars

Dataset Loaders

No data loaders found. You can submit your data loader here.


Similar Datasets


  • Unknown