With the prosperity of sharing economy, more part-time and freelance suppliers (i.e., drivers) join on-demand mobility services. Because of suppliers’ autonomy and behavioural heterogeneity, the platform cannot ensure that suppliers will accept a dispatch order. One approach to mitigate this supply uncertainty is to provide suppliers with personalized menus of dispatch recommendations. A key issue then is to determine which dispatch orders (that can be passenger or goods services) should be allocated into the assortment menu of each supplier. This paper probabilistically models the suppliers’ order acceptance and choice behaviour, including a decline option. We propose two assortment optimization problems, disjoint and joint menus, to maximize the expected number of matches. We show that the objective function of the disjoint menu assortment problem is monotone non-decreasing submodular. In contrast, the objective function of the joint menu assortment problem is non-monotone and non-submodular. Accordingly, we present a standard greedy (SG) algorithm to solve the disjoint assortment problem, and γ∗-greedy and local search (LS) algorithms for the joint assortment problem. By bundling orders into consolidated routes, this paper extends the proposed menu assortment methods to the context of meal delivery services. A case study is presented based on the real-world demand in the Manhattan road network. The results show that drivers’ autonomy to decline the dispatch orders creates substantial coexistence of idle drivers and unmatched orders in the market. The proposed menu assortment methods curb such matching friction. Moreover, the numerical results demonstrate that the proposed algorithms significantly outperform the traditional dispatching policies applied in practice, e.g., one-to-one matching, in terms of platform efficiency, e.g., achieving more matches, customers’ experiences, e.g., reducing waiting time, and benefits for drivers, e.g., tapering off the income inequality among drivers.

中文翻译：

---

有拒绝选择的自由职业司机：按需出行服务中的调度菜单，以优化分类


随着共享经济的繁荣，越来越多的兼职和自由职业供应商（即司机）加入按需移动服务。由于供应商的自主性和行为异质性，该平台无法确保供应商会接受派单。减轻这种供应不确定性的一种方法是为供应商提供个性化的发货建议菜单。然后，一个关键问题是确定哪些调度订单（可以是客运或货运服务）应该分配到每个供应商的分类菜单中。本文对供应商的订单接受和选择行为进行了概率建模，包括拒绝选项。我们提出了两个分类优化问题，即 disjoint 和 joint menus，以最大化预期的匹配项数。我们表明，不相交菜单分类问题的目标函数是单调非递减子模。相比之下，联合菜单分类问题的目标函数是非单调和非子模的。因此，我们提出了一种标准的贪婪 （SG） 算法来解决不相交的分类问题，并提出了一种用于联合分类问题的γ∗贪婪和局部搜索 （LS） 算法。通过将订单捆绑到合并的路线中，本文将提议的菜单分类方法扩展到送餐服务的上下文中。根据曼哈顿道路网络中的实际需求提供了一个案例研究。结果表明，驾驶员拒绝调度订单的自主性创造了市场上闲置驾驶员和不匹配订单的实质性共存。拟议的菜单分类方法遏制了这种匹配摩擦。此外，数值结果表明，所提出的算法明显优于实际应用的传统调度策略 e。例如，一对一匹配，在平台效率方面，例如实现更多匹配，客户体验，例如减少等待时间，以及为司机带来的好处，例如，逐渐缩小司机之间的收入不平等。