Method and system for determining context-based subscription product suite in ride-hailing platforms

Abstract

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for determining a plurality of context-based subscription products in a ride-hailing platform are described. An exemplary method may include: determining a plurality of transportation contexts to which a plurality of context-based subscription products are applicable; obtaining a plurality of historical trip-requesting sessions between a plurality of riders and the ride-hailing platform, and a plurality of historical subscription products offered to the plurality of riders; constructing at least one key performance indicator (KPI) model that predicts a reward for the ride-hailing platform offering the plurality of context-based subscription products to the plurality of riders; constructing an optimization model to maximize a weighted sum of the at least one KPI model; and determining optimal values of the plurality of parameter vectors corresponding to the plurality of context-based subscription products.

1 . A computer-implemented method, comprising:
 constructing, by a computing device of the ride-hailing platform, a plurality of parameter vectors each representing a context-based subscription product, wherein each of the plurality of context-based subscription products corresponds to a transportation context, and when a transportation trip requested through a ride-hailing platform satisfies one of the plurality of transportation contexts, the corresponding context-based subscription sets a benefit to be applied to the transportation trip;   obtaining, by the computing device, a plurality of historical data comprising a plurality of historical trip-requesting sessions between a plurality of riders and the ride-hailing platform, a plurality of historical context-based subscription products offered to the plurality of riders, and respective labels;   jointly training a feature extractor network, a first neural network, and a second neural network based on the plurality of historical data, wherein the training comprises:
 feeding the plurality of historical data into the feature extractor network, 
 obtaining, from the feature extractor network, a plurality of feature vectors representing each of the plurality of historical trip-requesting sessions, 
 feeding the plurality of feature vectors into the first neural network to predict a plurality of first probabilities for the plurality of historical trip-requesting sessions being converted to trips, 
 feeding the plurality of feature vectors into the second neural network to predict a plurality of second probabilities of the plurality of riders subscribing to one or more of the plurality of subscription products, and 
 adjusting parameters of the feature extractor network, the first neural network, and the second neural network based on the respective labels, the plurality of first probabilities, and the plurality of second probabilities; 
   constructing, by the computing device based on the trained first neural network and the second neural network, at least one key performance indicator (KPI) model that predicts a KPI value for the ride-hailing platform;   constructing, by the computing device, an optimization model to maximize a weighted sum of the at least one KPI value generated by the at least one KPI model, wherein the optimization model comprises the plurality of parameter vectors as decision variables; and   determining, by the computing device solving the optimization model, an optimal configuration for each of the plurality of parameter vectors, wherein the optimal configuration comprises an optimal premium and an optimal benefit of the context-based subscription product corresponding to the each parameter vector.   
     
     
         2 . The method of  claim 1 , further comprising:
 for a given rider, determining, by the computing device of the ride-hailing platform based on the given rider's historical trips, one or more of the plurality of transportation contexts that the given rider's historical trips satisfy; and   sending, by the computing device of the ride-hailing platform, one or more of the plurality of context-based subscription products corresponding to the one or more determined transportation contexts to a computing device of the given rider for purchasing, wherein the one or more of the plurality of context-based subscription products are configured based on the corresponding optimal values.   
     
     
         3 . The method of  claim 1 , wherein the transportation context comprises one or more conditions, and the transportation trip satisfies the transportation context when the transportation trip satisfies the one or more conditions of the transportation context, and the one or more conditions comprises at least one of the following:
 a spatial condition defining an origin region or a destination region;   a route condition defining an origin region and a destination region;   a temporal condition defining a time range;   a spatial-temporal condition comprising any pair of the spatial condition and the temporal condition;   a route-temporal condition comprising any pair of the route condition and the temporal condition;   a weather condition; and   a marketplace condition.   
     
     
         4 . The method of  claim 1 , wherein:
 the KPI value indicates a reward to the ride-hailing platform for offering the plurality of context-based subscription products to the plurality of riders of the plurality of historical trip-requesting sessions, wherein the KPI value is determined by accumulating a plurality of session-level rewards respectively corresponding to the plurality of context-based subscription products; and   each of the plurality of session-level rewards is determined by:   determining, from an output of the first neural network, a first probability of a rider involved in a historical trip-requesting session subscribing to one of the plurality of context-based subscription products;   determining, from an output of a second neural network, a second probability of the rider converting the historical trip-requesting session to a trip;   determining an expected reward when the historical trip-requesting session is converted to the trip; and   determining the session-level reward as a product of the first probability, the second probability, and the expected reward.   
     
     
         5 . The method of  claim 4 , wherein the first probability is determined based at least on a plurality of rider-level features of the rider and the parameter vector of the context-based subscription product. 
     
     
         6 . The method of  claim 4 , wherein the historical trip-requesting session satisfies one of the plurality of transportation contexts, and the second probability is determined based at least on a plurality of session-level features of the historical trip-requesting session, a plurality of rider-level features of the rider, and the parameter vector of the context-based subscription product corresponding to the one transportation context. 
     
     
         7 . The method of  claim 6 , wherein the plurality of session-level features comprise at least one of the following:
 a price displayed by the ride-hailing platform to a computing device of the rider;   a discount displayed by the ride-hailing platform to the computing device of the rider;   estimated time to arrive (ETA) for pick-up;   an estimated route distance to a destination;   time;   an origin location; and   a destination location.   
     
     
         8 . The method of  claim 6 , wherein the rider-level features comprise at least one of the following:
 an average price of historical trips; and   the rider's subscription history.   
     
     
         9 . The method of  claim 1 , wherein the at least one KPI model comprises at least one of the following: a net income model, a gross merchandise value (GMV) model, or a number of trips model. 
     
     
         10 . The method of  claim 1 , wherein each of the plurality of parameter vectors representing a context-based subscription product comprises at least one of the following values:
 a premium of the context-based subscription product;   a benefit amount or percentage of the context-based subscription product;   a period of validity of the context-based subscription product; and   a trip count limit of the context-based subscription product.   
     
     
         11 . The method of  claim 1 , wherein the plurality of transportation contexts are non-overlapping so that each of the transportation contexts has only one applicable subscription product. 
     
     
         12 . The method of  claim 1 , wherein one of the at least one KPI models generates non-summable values, and the weighted sum of the at least one KPI models is determined by:
 dividing the non-summable KPI models into a plurality of summable KPI models;   determining a plurality of KPI values for the plurality of summable KPI models based on the plurality of historical trip-requesting sessions; and   obtaining a value of the non-summable KPI models based on the plurality of KPI values.   
     
     
         13 . The method of  claim 1 , wherein the optimization model is solved by one of the following algorithms:
 grid search;   greedy search; and   Bayesian optimization.   
     
     
         14 . A system comprising one or more processors and one or more non-transitory computer-readable memories coupled to the one or more processors, the one or more non-transitory computer-readable memories storing instructions that, when executed by the one or more processors, cause the system to perform operations comprising:
 constructing a plurality of parameter vectors each representing a context-based subscription product, wherein each of the plurality of context-based subscription products corresponds to a transportation context, and when a transportation trip requested through a ride-hailing platform satisfies one of the plurality of transportation contexts, the corresponding context-based subscription sets a benefit to be applied to the transportation trip;   obtaining a plurality of historical data comprising a plurality of historical trip-requesting sessions between a plurality of riders and the ride-hailing platform, a plurality of historical context-based subscription products offered to the plurality of riders, and respective labels;   jointly training a feature extractor network, a first neural network, and a second neural network based on the plurality of historical data, wherein the training comprises:
 feeding the plurality of historical data into the feature extractor network, 
 obtaining, from the feature extractor network, a plurality of feature vectors representing each of the plurality of historical trip-requesting sessions, 
 feeding the plurality of feature vectors into the first neural network to predict a plurality of first probabilities for the plurality of historical trip-requesting sessions being converted to trips, 
 feeding the plurality of feature vectors into the second neural network to predict a plurality of second probabilities of the plurality of riders subscribing to one or more of the plurality of subscription products, and 
 adjusting parameters of the feature extractor networks, the first neural network, and the second neural network based on the respective labels, the plurality of first probabilities, and the plurality of second probabilities; 
   constructing based on the trained first neural network and the second neural network, at least one key performance indicator (KPI) model that predicts a KPI value for the ride-hailing platform.   constructing an optimization model to maximize a weighted sum of the at least one KPI value generated by the at least one KPI model, wherein the optimization model comprises the plurality of parameter vectors as decision variables; and   determining, by the computing device solving the optimization model, an optimal configuration for each of the plurality of parameter vectors, wherein the optimal configuration comprises an optimal premium and an optimal benefit of the context-based subscription product corresponding to the each parameter vector.   
     
     
         15 . The system of  claim 14 , wherein the operations further comprise:
 for a given rider, determining, based on the given rider's historical trips, one or more of the plurality of transportation contexts that the given rider's historical trips satisfy; and   sending one or more of the plurality of context-based subscription products corresponding to the one or more determined transportation contexts to a computing device of the given rider for purchasing, wherein the one or more of the plurality of context-based subscription products are configured based on the corresponding optimal values.   
     
     
         16 . The system of  claim 14 , wherein each of the plurality of transportation contexts comprises one or more conditions, and the trip satisfies the transportation context when the trip satisfies the one or more conditions of the transportation context, and the one or more conditions comprises at least one of the following:
 a spatial condition defining an origin region or a destination region;   a route condition defining an origin region and a destination region;   a temporal condition defining a time range;   a spatial-temporal condition comprising any pair of the spatial condition and the temporal condition;   a route-temporal condition comprising any pair of the route condition and the temporal condition;   a weather condition; and   a marketplace condition.   
     
     
         17 . A non-transitory computer-readable storage medium storing instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:
 constructing a plurality of parameter vectors each representing a context-based subscription product, wherein each of the plurality of context-based subscription products corresponds to a transportation context, and when a transportation trip requested through a ride-hailing platform satisfies one of the plurality of transportation contexts, the corresponding context-based subscription sets a benefit to be applied to the transportation trip;   obtaining a plurality of historical data comprising a plurality of historical trip-requesting sessions between a plurality of riders and the ride-hailing platform, [[and]], a plurality of historical context-based subscription products offered to the plurality of riders, and respective labels;   jointly training a feature extractor network, a first neural network, and a second neural network based on the plurality of historical data, wherein the training comprises:
 feeding the plurality of historical data into the feature extractor network, 
 obtaining, from the feature extractor network, a plurality of feature vectors representing each of the plurality of historical trip-requesting sessions, 
 feeding the plurality of feature vectors into the first neural network to predict a plurality of first probabilities for the plurality of historical trip-requesting sessions being converted to trips, 
 feeding the plurality of feature vectors into the second neural network to predict a plurality of second probabilities of the plurality of riders subscribing to one or more of the plurality of subscription products, and 
 adjusting parameters of the feature extractor network, the first neural network, and the second neural network based on the respective labels, the plurality of first probabilities, and the plurality of second probabilities; 
   constructing based on the trained first neural network and the second neural network, at least one key performance indicator (KPI) model that predicts a KPI value for the ride-hailing platform;   constructing an optimization model to maximize a weighted sum of the at least one KPI value generated by the at least one KPI model, wherein the optimization model comprises the plurality of parameter vectors as decision variables; and   determining, by the computing device solving the optimization model, an optimal configuration for each of the plurality of parameter vectors, wherein the optimal configuration comprises an optimal premium and an optimal benefit of the context-based subscription product corresponding to the each parameter vector.   
     
     
         18 . The storage medium of  claim 17 , wherein the operations further comprise:
 for a given rider, determining, based on the given rider's historical trips, one or more of the plurality of transportation contexts that the given rider's historical trips satisfy; and   sending one or more of the plurality of context-based subscription products corresponding to the one or more determined transportation contexts to a computing device of the given rider for purchasing, wherein the one or more of the plurality of context-based subscription products are configured based on the corresponding optimal values.   
     
     
         19 . The storage medium of  claim 17 , wherein the at least one KPI models comprise at least one of the following: a net income model, a gross merchandise value (GMV) model, or a number of trips model. 
     
     
         20 . The storage medium of  claim 17 , wherein each of the plurality of transportation contexts comprises one or more conditions, and the trip satisfies the transportation context when the trip satisfies the one or more conditions of the transportation context, and the one or more conditions comprises at least one of the following:
 a spatial condition defining an origin region or a destination region;   a route condition defining an origin region and a destination region;   a temporal condition defining a time range;   a spatial-temporal condition comprising any pair of the spatial condition and the temporal condition;   a route-temporal condition comprising any pair of the route condition and the temporal condition;   a weather condition; and   a marketplace condition.