Large Scale Multi-Actor Generative Dialog Modeling

被引:0
|
作者
Boyd, Alex [1 ]
Puri, Raul [2 ]
Shoeybi, Mohammad [2 ]
Patwary, Mostofa [2 ]
Catanzaro, Bryan [2 ]
机构
[1] Univ Calif Irvine, Dept Stat, Irvine, CA 92697 USA
[2] NVIDIA, Santa Clara, CA USA
关键词
D O I
暂无
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Non-goal oriented dialog agents (i.e. chat-bots) aim to produce varying and engaging conversations with a user; however, they typically exhibit either inconsistent personality across conversations or the average personality of all users. This paper addresses these issues by controlling an agent's persona upon generation via conditioning on prior conversations of a target actor. In doing so, we are able to utilize more abstract patterns within a person's speech and better emulate them in generated responses. This work introduces the GENERATIVE CONVERSATION CONTROL model, an augmented and fine-tuned GPT-2 language model that conditions on past reference conversations to probabilistically model multi-turn conversations in the actor's persona. We introduce an accompanying data collection procedure to obtain 10.3M conversations from 6 months worth of Reddit comments. We demonstrate that scaling model sizes from 117M to 8.3B parameters yields an improvement from 23.14 to 13.14 perplexity on 1.7M held out Reddit conversations. Increasing model scale yielded similar improvements in human evaluations that measure preference of model samples to the held out target distribution in terms of realism (31% increased to 37% preference), style matching (37% to 42%), grammar and content quality (29% to 42%), and conversation coherency (32% to 40%). We find that conditionally modeling past conversations improves perplexity by 0.47 in automatic evaluations. Through human trials we identify positive trends between conditional modeling and style matching and outline steps to further improve persona control.
引用
收藏
页码:66 / 84
页数:19
相关论文
共 50 条
  • [1] Multi-Actor Value Modeling for Federated Systems
    Grogan, Paul T.
    Ho, Koki
    Golkar, Alessandro
    de Weck, Olivier L.
    [J]. IEEE SYSTEMS JOURNAL, 2018, 12 (02): : 1193 - 1202
  • [2] Multi-actor systems and ethics
    Pruyt, Erik
    [J]. INTERNATIONAL TRANSACTIONS IN OPERATIONAL RESEARCH, 2010, 17 (04) : 507 - 520
  • [3] The Multi-actor Game of Peacekeeping in Africa
    Brosig, Malte
    [J]. INTERNATIONAL PEACEKEEPING, 2010, 17 (03) : 327 - 342
  • [4] Modelling Multi-actor Security Dilemma
    Drmola, Jakub
    [J]. STRATEGIC ANALYSIS, 2016, 40 (02) : 92 - 100
  • [5] Multi-actor Markov decision processes
    Ahn, HS
    Righter, R
    [J]. JOURNAL OF APPLIED PROBABILITY, 2005, 42 (01) : 15 - 26
  • [6] Multi-actor mechanism for actor-critic reinforcement learning
    Li, Lin
    Li, Yuze
    Wei, Wei
    Zhang, Yujia
    Liang, Jiye
    [J]. INFORMATION SCIENCES, 2023, 647
  • [7] A Relational Approach to Leadership for Multi-Actor Governance
    Craps, Marc
    Vermeesch, Inge
    Dewulf, Art
    Sips, Koen
    Termeer, Katrien
    Bouwen, Rene
    [J]. ADMINISTRATIVE SCIENCES, 2019, 9 (01)
  • [8] Conceptualizing customer experience in multi-actor platforms
    Dhrithi Mahadevan
    G. Shainesh
    [J]. AMS Review, 2024, 14 (1-2) : 83 - 103
  • [9] Management in Networks: On Multi-Actor Decision Making
    Kalu, Kalu N.
    [J]. PUBLIC ADMINISTRATION REVIEW, 2010, 70 (01) : 162 - 167
  • [10] Setting conservation priorities in multi-actor systems
    O'Bryan, Christopher J.
    Rhodes, Jonathan R.
    Osunkoya, Olusegun O.
    Lundie-Jenkins, Geoff
    Mudiyanselage, Nisansala Abeysinghe
    Sydes, Travis
    Calvert, Moya
    McDonald-Madden, Eve
    Bode, Michael
    [J]. BIOSCIENCE, 2023, 73 (07) : 522 - 532