A Neural Network Approach to Context-Sensitive Generation of Conversational Responses
- Alessandro Sordoni ,
- Michel Galley ,
- Michael Auli ,
- Chris Brockett ,
- Yangfeng Ji ,
- Meg Mitchell ,
- Jian-Yun Nie ,
- Jianfeng Gao ,
- Bill Dolan
Conference of the North American Chapter of the Association for Computational Linguistics - Human Language Technologies (NAACL-HLT) |
We present a novel response generation system that can be trained end to end on large quantities of unstructured Twitter conversations. A neural network architecture is used to address sparsity issues that arise when integrating contextual information into classic statistical models, allowing the system to take into account previous dialog utterances. Our dynamic-context generative models show consistent gains over both context-sensitive and non-context-sensitive Machine Translation and Information Retrieval baselines.
Publication Downloads
Microsoft Research Social Media Conversation Corpus
June 1, 2015
A collection of 12,696 Tweet Ids representing 4,232 three-step conversational snippets extracted from Twitter logs. Each row in the dataset represents a single context-message-response triple that has been evaluated by crowdsourced annotators as scoring an average of 4 or higher on a 5-point Likert scale measuring quality of the response in the context. The data has been randomly binned into tuning (development) and test sets, comprising 2118 and 2114 triples respectively. It is released to the natural language processing community for academic research purposes only. In order to access the underlying tweets and related metadata, you will need to call the Twitter API.