Point-of-interest recommendation

Patent metadata

Field	Value
Publication number	US-12585984-B2
Application number	US-202117531132-A
Country	US
Kind code	B2
Filing date	Nov 19, 2021
Priority date	Dec 24, 2020
Publication date	Mar 24, 2026
Grant date	Mar 24, 2026

Abstract

A training method for a point-of-interest recommendation model and a method for recommending a point of interest are provided. An implementation solution includes: obtaining training data including a plurality of point-of-interest recommendation requests; determining initialization parameters of the point-of-interest recommendation model; for a first point-of-interest recommendation request among the plurality of point-of-interest recommendation requests, determining a current return for the first point-of-interest recommendation request by utilizing the point-of-interest recommendation model, and determining, based on a second point-of-interest recommendation request initiated after the first point-of-interest recommendation request is completed, a target return for the first point-of-interest recommendation request by utilizing the point-of-interest recommendation model; and adjusting the initialization parameters of the point-of-interest recommendation model based on a difference between the current return and the target return for the first point-of-interest recommendation request, to obtain final parameters of the point-of-interest recommendation model.

First claim

What is claimed is:

1 . A training method for a point-of-interest recommendation model, the method comprising:
obtaining, by a processor, training data that comprises a plurality of point-of-interest recommendation requests; determining, by the processor, initialization parameters of the point-of-interest recommendation model; for a first point-of-interest recommendation request among the plurality of point-of-interest recommendation requests, determining, by the processor, a current return for the first point-of-interest recommendation request by utilizing the point-of-interest recommendation model; determining, by the processor, based on a second point-of-interest recommendation request among the plurality of point-of-interest recommendation requests initiated after the first point-of-interest recommendation request is completed, a target return for the first point-of-interest recommendation request by utilizing the point-of-interest recommendation model, wherein the first point-of-interest recommendation request is completed when service supply starts to be provided from a point-of-interest corresponding to the first point-of-interest recommendation request; and adjusting, by the processor, the initialization parameters of the point-of-interest recommendation model based on a difference between the current return and the target return for the first point-of-interest recommendation request, to obtain updated parameters of the point-of-interest recommendation model; wherein the determining the target return for the first point-of-interest recommendation request by utilizing the point-of-interest recommendation model comprises: determining at least one completed request completed between a first time when the first point-of-interest recommendation request is received and a second time when the second point-of-interest recommendation request is received, wherein the at least one completed request is different from the first point-of-interest recommendation request and the second point-of-interest recommendation request, and the at least one completed request is completed when service supply starts to be provided from at least one point-of-interest corresponding to the at least one completed request; determining an instant return for the first point-of-interest recommendation request based on a waiting time of each completed request of the at least one completed request; processing an identification parameter of a second target point-of-interest and a second environmental state parameter associated with the second point-of-interest recommendation request by utilizing the point-of-interest recommendation model, to obtain a maximum recommended value for a target point-of-interest for the second point-of-interest recommendation request, wherein the maximum recommended value is determined as a long-run return for the first point-of-interest recommendation request; and determining the target return for the first point-of-interest recommendation request based on the instant return for the first point-of-interest recommendation request and the long-run return for the first point-of-interest recommendation request.

2 . The method of claim 1 , wherein the training data further comprises a plurality of environmental state parameters associated with the plurality of point-of-interest recommendation requests, respectively, wherein for each environmental state parameter of the plurality of environmental state parameters, the environmental state parameter comprises resource occupation information and predictive use information of a candidate point of interest at a time when the point-of-interest recommendation request corresponding to the environmental state parameter is initiated.

3 . The method of claim 2 , wherein for each point-of-interest recommendation request of the plurality of point-of-interest recommendation requests, the predictive use information of the environmental state parameter associated with the point-of-interest recommendation request comprises a predicted amount of other point-of-interest recommendation requests initiated within a predetermined distance around the candidate point of interest within a predetermined time range after the point-of-interest recommendation request is received.

4 . The method of claim 3 , wherein the environmental state parameter associated with the point-of-interest recommendation request further comprises:
a time when the point-of-interest recommendation request is received; a traveling time from a geographical location associated with the point-of-interest recommendation request to the candidate point of interest; and service capability information of the candidate point of interest.

5 . The method of claim 4 , wherein the candidate point of interest is an electric vehicle charging station.

6 . The method of claim 5 , wherein the resource occupation information of the candidate point of interest comprises an amount of idle charging spaces of a candidate charging station.

7 . The method of claim 5 , wherein the service capability information of the candidate point of interest is charging power of a candidate charging station.

8 . The method of claim 2 , wherein the determining the current return for the first point-of-interest recommendation request by utilizing the point-of-interest recommendation model comprises:
determining that a target point of interest for the first point-of-interest recommendation request is a first target point of interest; and processing a first environmental state parameter associated with the first point-of-interest recommendation request and an identification parameter of the first target point of interest by utilizing the point-of-interest recommendation model, to obtain the current return for the first point-of-interest recommendation request.

9 . The method of claim 1 , wherein the determining the instant return for the first point-of-interest recommendation request comprises:
determining a completion return for each completed request of the at least one completed request based on the waiting time of each completed request, wherein the waiting time is inverse to the completion return; and determining the instant return for the first point-of-interest recommendation request based on a weighted sum of the completion returns for respective completed requests.

10 . The method of claim 1 , wherein the determining the target return for the first point-of-interest recommendation request based on the instant return for the first point-of-interest recommendation request and the long-run return for the first point-of-interest recommendation request comprises:
determining a weighted sum of the instant return and the long-run return as the target return for the first point-of-interest recommendation request.

11 . A method for recommending a point of interest, the method being applied to a computing device running a point-of-interest recommendation model thereon, wherein the method comprises:
receiving a point-of-interest recommendation request; in response to receiving the point-of-interest recommendation request, determining an environmental state parameter associated with the point-of-interest recommendation request, wherein the environmental state parameter comprises resource occupation information and predictive use information of a candidate point of interest at a time when the point-of-interest recommendation request is initiated; processing the environmental state parameter and an identification parameter of the candidate point of interest by utilizing the point-of-interest recommendation model, to obtain a recommended value of the candidate point of interest; and determining a target point of interes

Assignees

• Beijing Baidu Netcom Sci & Tech Co Ltd

Inventors

• Liu Hao
• Zhang Weijia
• Dou Dejing
• Xiong Hui

Classifications

• G06N3/092

  Reinforcement learning · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/045

  Combinations of networks · CPC title

• B60L2240/72

  Charging station selection relying on external data · CPC title

• B60L2260/50

  by future state prediction · CPC title