Fantom: A Crowdsourced Social Chatbot using an Evolving Dialog Graph

Fantom: A Crowdsourced Social Chatbot using an Evolving Dialog Graph. Jonell, P., Bystedt, M., Irmak Do, F., Fallgren, P., Ivarsson, J., Slukova, M., Wennberg, U., Lopes, J., Boye, J., & Skantze, G. 1st Proceedings of Alexa Prize (Alexa Prize 2018)., 2018.

Paper

Website abstract bibtex

In this paper we present Fantom, a social chatbot competing in the Amazon Alexa Prize 2018 1. The system uses a dialog graph for retrieving an approximation of the current dialog context in order to find suitable response candidates in this context. The graph is gradually built using user utterances from actual interactions, and system responses suggested by crowd workers. To this end, we developed an automatic system for finding dialog contexts that were often visited but lacked system responses in order to automatically post tasks on Amazon Mechanical Turk. Workers could see a brief excerpt of past conversation history and were asked to suggest a good response, based on a description of the system's persona and a set of rules that would help foster more engaging conversations. Our main contributions are 1) describing the use of a graph-based approach for context modeling, 2) techniques used in order to make the crowd workers author good content, and 3) discussion of learning outcomes from the Alexa Prize challenge.

@article{
 title = {Fantom: A Crowdsourced Social Chatbot using an Evolving Dialog Graph},
 type = {article},
 year = {2018},
 pages = {1-20},
 websites = {https://wit.ai/},
 id = {7e31239c-24d9-3b86-863d-7bb7b8892bb6},
 created = {2020-01-06T19:15:02.009Z},
 file_attached = {true},
 profile_id = {fb8d345a-1d79-3791-a6c6-00233ea44521},
 last_modified = {2020-01-06T19:27:16.569Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Jonell2018},
 private_publication = {false},
 abstract = {In this paper we present Fantom, a social chatbot competing in the Amazon Alexa Prize 2018 1. The system uses a dialog graph for retrieving an approximation of the current dialog context in order to find suitable response candidates in this context. The graph is gradually built using user utterances from actual interactions, and system responses suggested by crowd workers. To this end, we developed an automatic system for finding dialog contexts that were often visited but lacked system responses in order to automatically post tasks on Amazon Mechanical Turk. Workers could see a brief excerpt of past conversation history and were asked to suggest a good response, based on a description of the system's persona and a set of rules that would help foster more engaging conversations. Our main contributions are 1) describing the use of a graph-based approach for context modeling, 2) techniques used in order to make the crowd workers author good content, and 3) discussion of learning outcomes from the Alexa Prize challenge.},
 bibtype = {article},
 author = {Jonell, Patrik and Bystedt, Mattias and Irmak Do, Fethiye and Fallgren, Per and Ivarsson, Jonas and Slukova, Marketa and Wennberg, Ulme and Lopes, José and Boye, Johan and Skantze, Gabriel},
 journal = {1st Proceedings of Alexa Prize (Alexa Prize 2018).}
}

Downloads: 0