VIP内容

我们研究会话领域探索(CODEX),其中用户的目标是通过与信息机器人交谈来丰富她对给定领域的知识。这样的对话应该以高质量的领域知识为基础,同时也要有吸引力和开放性。食典委机器人应积极主动地介绍相关信息,即使用户没有直接要求。机器人还应该适当地将对话转向域的未发现区域。为了解决这些对话特性,我们引入了一种称为动态组合的新方法,该方法将候选内容生成与机器人响应的灵活组合解耦。这允许机器人控制所提供内容的来源、正确性和质量,同时通过对话管理器以组合方式选择最合适的内容来实现灵活性。我们实现了一个基于动态组合的法典机器人,并将其集成到谷歌助理中。作为一个示例域,该机器人以无缝体验的方式谈论NBA篮球联赛,因此用户不知道他们是在与vanilla系统交谈,还是在与我们的CODEX机器人进行交谈。结果是积极的,并能让你洞悉怎样才能进行一次愉快的谈话。据我们所知,这是作为商业助理系统一部分的开放式对话的第一次真正的用户实验。

成为VIP会员查看完整内容
0
9

最新论文

Recently, Abebe et al. (KDD 2018) and Chan et al. (WWW 2019) have considered an opinion dynamics optimization problem that is based on a popular model for social opinion dynamics, in which each agent has some fixed innate opinion, and a resistance that measures the importance it places on its innate opinion; moreover, the agents influence one another's opinions through an iterative process. Under certain conditions, this iterative process converges to some equilibrium opinion vector. Previous works gave an efficient local search algorithm to solve the unbudgeted variant of the problem, for which the goal is to modify the resistance of any number of agents (within some given range) such that the sum of the equilibrium opinions is minimized. On the other hand, it was proved that the $L_0$-budgeted variant is NP-hard, where the $L_0$-budget is a restriction given upfront on the number of agents whose resistance may be modified. Inspired by practical situations in which the effort to modify an agent's resistance increases with the magnitude of the change, we propose the $L_1$-budgeted variant, in which the $L_1$-budget is a restriction on the sum of the magnitudes of the changes over all agents' resistance parameters. In this work, we show that the $L_1$-budgeted variant is NP-hard via a reduction from vertex cover. However, contrary to the $L_0$-budgeted variant, a very technical argument is needed to show that the optimal solution can be achieved by focusing the given $L_1$-budget on as small a number of agents as possible, as opposed to spreading the budget over a large number of agents.

0
0
下载
预览
Top