Elamur Research⌕
☰
home / papers / 2018 / Training Millions of Personalized Dialogue Agents

Training Millions of Personalized Dialogue Agents

Summary

End-to-end chit-chat models are often inconsistent and not very engaging. Earlier work showed that giving a model a short textual persona helps. However, the PERSONA-CHAT corpus behind that finding was crowdsourced and contains only about a thousand personas.

This paper scales the idea up using a public REDDIT dump of 1.7 billion comments. For each user, the authors collect sentences from that user's own comments that read like self-descriptions, using simple filters on length, first-person words and part-of-speech content. Consecutive comment pairs in a thread become context–response examples. The resulting corpus links roughly 5 million personas to about 700 million dialogues.

The models are retrieval-based, ranking candidate answers rather than generating them. Context, response and persona each have their own encoder, and the persona is merged in through a single-hop memory network. Conditioning on personas improves accuracy for bag-of-words, LSTM and Transformer encoders alike. Pretraining on the REDDIT data and then fine-tuning gives a large jump on PERSONA-CHAT. This suggests the corpus is a broadly useful resource for personalized dialogue.

Key points

  • Persona sentences must have 4–20 tokens, contain "I" or "my", and include a verb plus a noun, pronoun or adjective. Personas cover 4.6m users with rule-based selection and 7.2m with random selection, out of 13.2m users in total.
  • On REDDIT (hits@1 among 100 candidates), the Transformer reaches 74.4 with personas versus 69.1 without. The LSTM improves from 63.1 to 66.3, and the bag-of-words encoder from 51.7 to 53.9.
  • Persona selection matters:
    • Rule-based selection with up to 100 sentences works best (74.4).
    • Random sentences from the same user score 73.8.
    • Random sentences from other users score only 66.9.
    • Adding a PERSONA-CHAT-trained classifier filter actually hurts.
  • On PERSONA-CHAT (hits@1 among 20 candidates), the previous best was 35.4. A Transformer trained only on PERSONA-CHAT gets 42.1, and REDDIT pretraining plus fine-tuning reaches 60.7. Without fine-tuning, the REDDIT model scores just 25.6.
  • Training used Adamax with batch size 512 and FastText initialization. One epoch took about 3 days on 8 GPUs.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Natural language processing

Cited by

108 papers of our library cite it (of the 140884 whose bibliographies we read); 12 of them have a page here, 96 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/d18-1298

@inproceedings{mazare2018training,
  title = {{Training Millions of Personalized Dialogue Agents}},
  author = {Mazaré, Pierre-Emmanuel and Humeau, Samuel and Raison, Martin and Bordes, Antoine},
  booktitle = {Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing},
  year = {2018},
  doi = {10.18653/v1/d18-1298},
  url = {https://doi.org/10.18653/v1/d18-1298}
}

References

5 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Antoine Bordes, Y-Lan Boureau, and Jason Weston. 2016. Learning end-to-end goal-oriented dialog. arXiv preprint arXiv:1605.07683. find
  2. Jesse Dodge, Andreea Gane, Xiang Zhang, Antoine Bordes, Sumit Chopra, Alexander Miller, Arthur Szlam, and Jason Weston. 2015. Evaluating prerequisite qualities for learning end-to-end dialog systems. arXiv preprint arXiv:1511.06931. find
  3. Marjan Ghazvininejad, Chris Brockett, Ming-Wei Chang, Bill Dolan, Jianfeng Gao, Wen-tau Yih, and Michel Galley. 2017. A knowledge-grounded neural conversation model. CoRR, abs/1702.01932. — [arXiv:1702.01932](https://arxiv.org/abs/1702.01932) find
  4. Chaitanya K Joshi, Fei Mi, and Boi Faltings. 2017. Personalization in goal-oriented dialog. arXiv preprint arXiv:1706.07503. — [arXiv:1706.07503](https://arxiv.org/abs/1706.07503) find
  5. Jiwei Li, Michel Galley, Chris Brockett, Jianfeng Gao, and Bill Dolan. 2016. A persona-based neural conversation model. CoRR, abs/1603.06155. find our card

This page in Markdown → https://md.research.elamur.ai/free/en/papers/mazare-2018-training-millions-personalized-dialogue-agents/