CN112182430A

CN112182430A - Method and device for recommending places, electronic equipment and storage medium

Info

Publication number: CN112182430A
Application number: CN202011004220.3A
Authority: CN
Inventors: 李嘉航; 陈弥
Original assignee: Hanhai Information Technology Shanghai Co Ltd
Current assignee: Hanhai Information Technology Shanghai Co Ltd
Priority date: 2020-09-22
Filing date: 2020-09-22
Publication date: 2021-01-05

Abstract

The application provides a method and a device for recommending places, electronic equipment and a storage medium. The method comprises the following steps: acquiring a starting place and recommended place acquisition request carrying the starting time of a target user terminal, which is initiated by the target user terminal; determining click rate characteristics according to a starting place, a starting time and a historical order of a target user terminal; inputting the departure location and the departure time into a trip habit model of the target user terminal, and determining trip habit characteristics, wherein the trip habit characteristics represent the probability that the target user terminal trips by taking the departure time as the historical departure time and the departure location as the historical departure location; splicing the travel habit features and the click rate features and then inputting the spliced travel habit features and click rate features into a place recommendation model to obtain respective corresponding probabilities of a plurality of candidate places; determining a place to be recommended according to the probability corresponding to each of the candidate places; and pushing the place to be recommended to the target user terminal. The method can obviously improve the accuracy of destination prediction.

Description

Method and device for recommending places, electronic equipment and storage medium

Technical Field

The embodiment of the application relates to the technical field of data processing, in particular to a method and a device for recommending places, electronic equipment and a storage medium.

Background

The destination prediction means: and under the condition that the position and the travel time of the user are known, predicting the destination of the user for the travel. The destination prediction has wide application under multiple occasions, for example, in a taxi taking scene, after a user logs in a taxi taking platform, the taxi taking platform predicts and recommends the current trip destination of the user, can help the user to take a taxi quickly, effectively improves taxi taking experience of the user, and brings more benefits for the platform.

In the related art, there are mainly two ways to predict a destination. The first mode is that the click rate of the user on a limited number of candidate destinations is predicted through a click rate prediction model, and then the destination with the highest click rate is determined as the destination to be traveled by the user at present. And secondly, estimating the probability of each destination of the user in the trip through an estimation model of the probability function, and determining the destination with the highest probability as the destination of the user to be currently in the trip.

However, both of the above approaches have certain disadvantages. In the first mode, when the estimation is performed based on the click rate estimation model, the model is a discriminant model, and the conditional probability P (y | x) is directly modeled without considering the data distribution P (x, y), so that the model cannot summarize the behavior habit of the user. For example, the click rate estimation model can predict that the probability of the user A going to the place M at 8 pm is the highest, but the model cannot extract the travel habit of the user A going to the place M at 8 pm. And aiming at the second mode, the pre-estimation model of the probability function is used for modeling the user destination level, and if the frequency of the user for going out of the destination is less and the model modeling cannot be met, the probability of the user for going out of the destination cannot be predicted, so that the mode has higher requirements on the order number of the user.

Disclosure of Invention

The embodiment of the application provides a method and a device for recommending a place, electronic equipment and a storage medium, which can effectively improve the accuracy of destination prediction.

A first aspect of an embodiment of the present application provides a method for place recommendation, where the method includes:

acquiring a recommended place acquiring request initiated by a target user terminal, wherein the recommended place acquiring request carries a starting place and a starting time of the target user terminal;

determining click rate characteristics according to the departure place, the departure time and the historical order of the target user terminal, wherein the click rate characteristics at least comprise the distance between the historical place and the departure place in the historical order of the target user terminal, the historical common trip time and the trip frequency of the historical place;

inputting the departure location and the departure time into a trip habit model of the target user terminal, and determining trip habit characteristics, wherein the trip habit characteristics represent the probability that the target user terminal trips with the departure time as a historical departure time and with the departure location as a historical departure location;

splicing the travel habit features and the click rate features and then inputting the spliced travel habit features and click rate features into a place recommendation model to obtain the probability corresponding to each of a plurality of candidate places;

determining a place to be recommended according to the probability corresponding to each of the candidate places;

and pushing the place to be recommended to the target user terminal.

Optionally, the travel habit model of the target user terminal is obtained according to the following steps:

dividing a plurality of historical orders of the target user terminal into a plurality of types of historical orders by taking the historical places as dimensions, wherein each historical order included in one type of historical orders corresponds to the same historical place;

for each type of historical order, fitting a mixed Gaussian distribution curve of the historical place according to the times that the target user terminal goes to the historical place corresponding to the type of historical order at each historical departure time and the times that the target user terminal goes to the historical place corresponding to the type of historical order from each historical departure place;

and fitting the mixed Gaussian distribution curves of the plurality of historical places to obtain a travel habit model of the target user terminal.

Optionally, the location recommendation model is obtained by training according to the following steps:

determining sample click rate characteristics of a sample user terminal according to a sample starting time and a sample starting place in the sample order and a plurality of historical orders before the sample order is generated by taking each historical order of the sample user terminal as a sample order;

determining a sample travel habit model of the sample user terminal according to a plurality of historical orders of the sample user terminal;

inputting a sample departure time and a sample departure place in a sample order of the sample user terminal into a travel habit model of the sample user terminal, and determining sample travel habit characteristics of the sample user terminal;

splicing respective sample click rate characteristics and sample travel habit characteristics of the sample user terminals to obtain training samples;

and training a click rate estimation model by using a plurality of training samples to obtain the place recommendation model.

Optionally, after obtaining the probability that each of the plurality of candidate locations corresponds to, the method further includes:

when the location input operation of the target user terminal is detected, the candidate locations are sent to the target user terminal as candidate input locations according to the sequence from high probability to low probability, so that the target user terminal displays the candidate locations in sequence.

Optionally, after the place to be recommended is pushed to the target user terminal, the method further includes:

determining a get-off point to be recommended according to the historical order of the place to be recommended;

and pushing the get-off point to be recommended to the target user terminal and/or the driver terminal, and sending prompt information to the target user terminal and/or the driver terminal so as to prompt the user of the target user terminal to get off at the get-off point to be recommended.

A second aspect of the embodiments of the present application provides an apparatus for recommending a location, where the apparatus includes:

the system comprises an obtaining module, a sending module and a receiving module, wherein the obtaining module is used for obtaining a recommended place obtaining request initiated by a target user terminal, and the recommended place obtaining request carries a starting place and a starting time of the target user terminal;

a first determining module, configured to determine click rate characteristics according to the departure location, the departure time, and the historical orders of the target user terminal, where the click rate characteristics at least include a distance between a historical location in the historical orders of the target user terminal and the departure location, a historical frequent trip time, and a trip frequency of the historical location;

a first input module, configured to input the departure location and the departure time into a trip habit model of the target user terminal, and determine a trip habit feature, where the trip habit feature represents a probability that the target user terminal trips with the departure time as a historical departure time and the departure location as a historical departure location;

the second input module is used for splicing the travel habit features and the click rate features and then inputting the spliced travel habit features and click rate features into a place recommendation model to obtain the probability corresponding to each of a plurality of candidate places;

the second determining module is used for determining the place to be recommended according to the probability corresponding to each of the candidate places;

and the first pushing module is used for pushing the place to be recommended to the target user terminal.

Optionally, the apparatus further comprises:

the dividing module is used for dividing a plurality of historical orders of the target user terminal into a plurality of types of historical orders by taking a historical place as a dimension, wherein each historical order included in the one type of historical orders corresponds to the same historical place;

the first fitting module is used for fitting a mixed Gaussian distribution curve of each type of historical order according to the times that the target user terminal goes to the historical place corresponding to the type of historical order at each historical departure time and the times that the target user terminal goes to the historical place corresponding to the type of historical order from each historical departure place;

and the second fitting module is used for fitting the mixed Gaussian distribution curves of the plurality of historical places to obtain the travel habit model of the target user terminal.

Optionally, the apparatus further comprises:

the second determining module is used for determining the sample click rate characteristics of the sample user terminal according to the sample starting time and the sample starting place in the sample orders and a plurality of historical orders before the sample orders are generated by taking each historical order of the sample user terminal as a sample order;

a third determining module, configured to determine a sample travel habit model of the sample user terminal according to a plurality of historical orders of the sample user terminal;

a fourth determining module, configured to input the sample departure time and the sample departure location in the sample order of the sample user terminal into the travel habit model of the sample user terminal, and determine sample travel habit characteristics of the sample user terminal;

the splicing module is used for splicing the respective sample click rate characteristics and the sample travel habit characteristics of the sample user terminals to obtain training samples;

and the training module is used for training the click rate estimation model by using a plurality of training samples to obtain the place recommendation model.

Optionally, the apparatus further comprises:

and the sending module is used for sending the candidate places to the target user terminal as candidate input places according to the sequence from high probability to low probability after the probabilities corresponding to the candidate places are obtained and when the place input operation of the target user terminal is detected, so that the target user terminal sequentially displays the candidate places.

Optionally, the apparatus further comprises:

the fifth determining module is used for determining a get-off point to be recommended according to a historical order of the place to be recommended after the place to be recommended is pushed to the target user terminal;

and the second pushing module is used for pushing the get-off point to be recommended to the target user terminal and/or the driver terminal after pushing the place to be recommended to the target user terminal, and sending prompt information to the target user terminal and/or the driver terminal so as to prompt the user of the target user terminal to get off at the get-off point to be recommended.

A third aspect of the embodiments of the present application provides a computer-readable storage medium, on which a computer program is stored, which when executed by a processor, implements the steps in the method for location recommendation according to the first aspect of the present application.

A fourth aspect of the embodiments of the present application provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor, when executing, implements the steps in the location recommendation described in the first aspect of the present application.

According to the place recommending method, a recommended place obtaining request initiated by a target user terminal is obtained, and the recommended place obtaining request carries a starting place and a starting time of the target user terminal. And then, according to the departure point, the departure time and the historical order of the target user terminal, determining click rate characteristics, wherein the click rate characteristics at least comprise the distance between the historical point and the departure point in the historical order of the target user terminal, the historical frequent trip time and the trip frequency of the historical point. And then, inputting the departure point and the departure time into a trip habit model of the target user terminal, and determining trip habit characteristics, wherein the trip habit characteristics represent the probability of the target user terminal for taking the departure time as the historical departure time and taking the departure point as the historical departure point. And then, splicing the travel habit features and the click rate features, inputting the spliced travel habit features and click rate features into a place recommendation model to obtain respective corresponding probabilities of a plurality of candidate places, and determining a place to be recommended according to the respective corresponding probabilities of the plurality of candidate places. And finally, pushing the place to be recommended to a target user terminal. According to the method, the click rate characteristic and the travel habit characteristic are constructed at the same time, the most possible travel destination of the user is obtained through the two characteristics, and due to the addition of the travel habit which is a probabilistic characteristic, the interpretability of the result can be optimized, and the accuracy of destination prediction can be remarkably improved.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the description of the embodiments of the present application will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art that other drawings can be obtained according to these drawings without inventive exercise.

Fig. 1 is a schematic diagram illustrating an implementation scenario according to an embodiment of the present application;

FIG. 2 is a flow chart illustrating a method for location recommendation in accordance with an embodiment of the present application;

FIG. 3 is a flowchart illustrating a method for obtaining a travel habit model according to an embodiment of the present application;

fig. 4 is a graph illustrating a departure time-trip probability distribution according to an embodiment of the present application;

fig. 5 is a starting point-trip probability distribution surface diagram according to an embodiment of the present application;

fig. 6 is a diagram illustrating a departure time and a departure location-trip probability distribution according to an embodiment of the present application;

FIG. 7 is a flow diagram illustrating a method of obtaining a location recommendation model according to one embodiment of the present application;

FIG. 8 is a schematic diagram illustrating an offline model training process according to an embodiment of the present application;

FIG. 9 is a diagram illustrating an offline Gaussian model training process according to an embodiment of the present application;

fig. 10 is a schematic diagram illustrating a LightGBM model training process according to an embodiment of the present application;

FIG. 11 is a schematic diagram illustrating an online model prediction according to an embodiment of the present application;

FIG. 12 is a block diagram illustrating an apparatus for location recommendation according to an embodiment of the present application;

fig. 13 is a schematic diagram of an electronic device according to an embodiment of the present application.

Detailed Description

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are some, but not all, embodiments of the present application. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.

Fig. 1 is a schematic diagram of an implementation scenario according to an embodiment of the present application. In fig. 1, a destination prediction platform is communicatively coupled to a plurality of user terminals (including user terminal 1-user terminal N). After receiving a recommended place acquisition request initiated by a certain user terminal, a destination prediction platform predicts a destination to be traveled by the user terminal and sends a prediction result to the user terminal. For example, the destination prediction platform may be a taxi-taking platform, and the user terminal may be a terminal used by a user about to take a taxi to log in the taxi-taking platform.

The application provides a method for recommending places, which is applied to a destination prediction platform in FIG. 1. Fig. 2 is a flowchart illustrating a method for location recommendation according to an embodiment of the present application. Referring to fig. 2, the method for place recommendation of the present application includes the steps of:

step S11: and acquiring a recommended place acquiring request initiated by a target user terminal, wherein the recommended place acquiring request carries a starting place and a starting time of the target user terminal.

In this embodiment, when the user terminal detects that the user satisfies the preset location recommendation condition, a recommended location acquisition request is sent to the destination prediction platform. The place recommendation condition may be various, for example, a user logs in the destination prediction platform, inputs a designated character, initiates a destination query, and the like, and the place recommendation condition may be set according to an actual requirement, and the place recommendation condition is not specifically limited in this embodiment. One user corresponds to one user terminal.

The starting point refers to the longitude and latitude of the user terminal when the point recommendation condition is met, and the starting time refers to the time when the point recommendation condition is met.

Step S12: determining click rate characteristics according to the departure place, the departure time and the historical orders of the target user terminal, wherein the click rate characteristics at least comprise the distance between the historical place and the departure place in the historical orders of the target user terminal, the historical common trip time and the trip frequency of the historical place.

In this embodiment, analyzing the historical orders of the user can obtain the historical travel record of the user, including: historical departure place-historical destination, travel time, arrival time, etc. And after receiving the recommended place acquisition request, the destination prediction platform acquires the current departure place and departure time of the user from the recommended place acquisition request. Then, the starting place, the starting time and the historical travel record are analyzed, so that the click rate characteristics of the user can be obtained, and the method comprises the following steps: the distance between the current departure place and each historical destination in the historical order, the commonly used historical travel time, the travel frequency of the historical place and the like. Of course, the click rate characteristics may also include other various types, which may be specifically set according to actual requirements, and the present application is not limited thereto. Wherein, the historical place in the historical order is the historical destination in the historical order.

Step S13: and inputting the departure place and the departure time into a trip habit model of the target user terminal, and determining trip habit characteristics, wherein the trip habit characteristics represent the probability that the target user terminal trips by taking the departure time as a historical departure time and the departure place as a historical departure place.

In this embodiment, the travel habit model is obtained by training in advance, and the travel habit characteristics can be obtained by inputting the departure location and the departure time to the travel habit model. The travel habit feature represents the probability that the user travels from the departure place at the departure time.

Step S14: and splicing the travel habit features and the click rate features and inputting the spliced travel habit features and click rate features into a place recommendation model to obtain the probability corresponding to each candidate place.

In this embodiment, the travel habit feature and the click rate feature are input to a location recommendation model trained in advance, that is, the probability of traveling from the starting location to each candidate destination at the starting time can be output. When the travel habit features and the click rate features are input into the place recommendation model, the travel habit features and the click rate features can be spliced according to a preset splicing rule, and a splicing result is used as the input of the model. The preset splicing rule can be set according to actual requirements, and this embodiment does not specifically limit this.

Wherein the candidate destinations are all of the historical destinations in the user's historical orders.

Step S15: and determining the place to be recommended according to the probability corresponding to each of the candidate places.

In this embodiment, after obtaining the probabilities of the user going out from the departure point to the respective candidate destinations at the departure time, the candidate destination with the highest probability may be taken as the point to be recommended, and represents the destination that the user is most likely to go out.

Step S16: and pushing the place to be recommended to the target user terminal.

In this embodiment, pushing the place to be recommended to the target user terminal is to facilitate the operation of the user, for example, when the user wants to go to city X, the city X is recommended to the user as the place to be recommended, and the user can directly select the option to perform the relevant operation without inputting information such as "city X" again.

In this embodiment, the travel habit model may be obtained by training a probability model. The probabilistic model may be, for example, a markov model, a gaussian mixture model, a bayesian model, etc., and this embodiment does not specifically limit this. The travel habit characteristics of the user, namely the probability characteristics of the user traveling to the destination, can be obtained through the travel habit model.

Illustratively, when a user A wants to go to a place Y at a certain time X, a taxi taking platform Z is logged in through a mobile phone carried by the user A, the taxi taking platform Z detects that the taxi taking requirement exists at the time of the user A, a recommended place acquisition request is sent to a background server, the background server acquires the time X and the place Y from the recommended place acquisition request, then the time X, the place Y and historical orders of the user A are analyzed, and the probability that the user A goes to each historical destination in the historical orders is obtained, wherein the probability comprises the place L-10%, the place M-30%, the place N-20% and the place Y-40%, for example, the recommended probability of the place Y is 40% highest, so that the taxi taking platform Z recommends the place Y to the user A, and the user operation is facilitated.

By the method for recommending places according to the embodiment, a recommended place obtaining request initiated by a target user terminal is obtained, and the recommended place obtaining request carries a starting place and a starting time of the target user terminal. And then, according to the departure point, the departure time and the historical order of the target user terminal, determining click rate characteristics, wherein the click rate characteristics at least comprise the distance between the historical point and the departure point in the historical order of the target user terminal, the historical frequent trip time and the trip frequency of the historical point. And then, inputting the departure point and the departure time into a trip habit model of the target user terminal, and determining trip habit characteristics, wherein the trip habit characteristics represent the probability of the target user terminal for taking the departure time as the historical departure time and taking the departure point as the historical departure point. And then, splicing the travel habit features and the click rate features, inputting the spliced travel habit features and click rate features into a place recommendation model to obtain respective corresponding probabilities of a plurality of candidate places, and determining a place to be recommended according to the respective corresponding probabilities of the plurality of candidate places. And finally, pushing the place to be recommended to a target user terminal. According to the method, the click rate characteristic and the travel habit characteristic are constructed at the same time, the most possible travel destination of the user is obtained through the two characteristics, and due to the addition of the probability characteristic of the travel habit, the interpretability of the result can be optimized, and the accuracy of destination prediction can be improved.

With reference to the foregoing embodiments, in one implementation manner, the present application further provides a method for obtaining a travel habit model. Fig. 3 is a flowchart illustrating a method for obtaining a travel habit model according to an embodiment of the present application. Referring to fig. 3, the travel habit model may be obtained according to the following steps:

step S21: and dividing a plurality of historical orders of the target user terminal into a plurality of types of historical orders by taking the historical places as dimensions, wherein each historical order included in the one type of historical orders corresponds to the same historical place.

In this embodiment, each user may generate a plurality of historical orders, and by analyzing the historical orders, all historical travel destinations of each user in a certain period of time may be counted. The historical orders of each user can be divided into a plurality of categories of historical orders by taking the historical destination as a dividing basis.

In the present embodiment, since the number of history orders per user is accumulated over time, when dividing the history orders, the history orders within a preset time period may be obtained first. The preset time period may be set according to actual requirements, and this embodiment does not specifically limit this.

Illustratively, the historical orders of the user a in the preset time period are 30, wherein the historical destinations of 10 orders are the place 1, the historical destinations of 5 orders are the place 2, the historical destinations of 3 orders are the place 3, the historical destinations of 2 orders are the place 4, and the historical destinations of 10 orders are the place 5. Then, the historical orders of the user a in the preset time period may be divided into 5 categories, and the dimensions are: location 1, location 2, location 3, location 4, and location 5. The historical order number corresponding to the location 1 is 10, the historical order number corresponding to the location 2 is 5, the historical order number corresponding to the location 3 is 3, the historical order number corresponding to the location 4 is 2, and the historical order number corresponding to the location 4 is 10.

Step S22: and for each type of historical orders, fitting a mixed Gaussian distribution curve of the historical place according to the times that the target user terminal goes to the historical place corresponding to the type of historical orders at each historical departure time and the times that the target user terminal goes to the historical place corresponding to the type of historical orders from each historical departure place.

In the present embodiment, the probability model may adopt a gaussian model. Each user terminal may have a plurality of historical destinations, each historical destination may have a plurality of types of gaussian distribution curves, the gaussian mixture model is a model composed of a plurality of sets of gaussian distributions, and the probability density function of the gaussian mixture model is a linear superposition of the probability densities of the plurality of gaussian distributions.

In the present embodiment, each history destination has at least three types of gaussian distribution curves:

first, departure time-trip probability distribution curve

Fig. 4 is a diagram illustrating a departure time-trip probability distribution according to an embodiment of the present application. In fig. 4, the abscissa represents each travel time in a day, and the ordinate represents a travel probability corresponding to each travel time. Two curves are shown in fig. 4, the higher curve being a gaussian distribution curve and the lower curve representing a mixed gaussian distribution curve at the peak.

For each history destination, counting the departure time of a plurality of history orders corresponding to the history destination can obtain the number of times that the user goes to the history destination at each history departure time. And finally, taking the ratio of the trip times corresponding to each historical departure time to the sum of the trip times corresponding to all historical departure times as the probability of the user going to the historical destination at the departure time.

Second, departure place-trip probability distribution curved surface

Fig. 5 is a starting point-trip probability distribution surface diagram according to an embodiment of the present application. In fig. 5, longitude and latitude information of each departure point is distributed on a horizontal plane, and each coordinate value on a mesh plane perpendicular to the horizontal plane represents a trip probability corresponding to each departure point. For example, from the departure location-trip probability distribution surface map, it can be known that when the user goes to the a location, the probability of departure from latitude and longitude (116.470932, 40.021655) is high, and the probability of departure from latitude and longitude (116.471055, 40.020734) is low.

For each historical destination, the starting points of a plurality of corresponding historical orders are counted, so that the number of times that the user goes to the historical destination at each historical starting point can be obtained. And finally, taking the ratio of the trip times corresponding to each historical departure place to the sum of the trip times corresponding to all the historical departure places as the probability of the user going to the historical destination at the departure place.

Thirdly, a departure time and a departure place-trip probability distribution diagram

Fig. 6 is a departure time and departure location-trip probability distribution diagram according to an embodiment of the present application. In fig. 6, since the probability distribution diagram is a 3-dimensional ellipsoid, the present application uses a cross-sectional view to represent the probability distribution for visual observation. The closer to the longitude and latitude (116.470932, 40.021655, 9) the higher the probability value of the ball center, the farther away the ball is. For example, from the departure time and departure point-trip probability distribution map, it can be seen that the probability of departure from point (116.470932, 40.021655) is high at 9 am when the user goes to point a.

The historical departure time and the historical departure place are considered as a whole for each historical destination, and the departure places of the corresponding historical orders are counted, so that the frequency of the user going to the historical destination at each historical departure time and each historical departure place can be obtained. And finally, taking the ratio of the trip times corresponding to each historical departure time and each historical departure place to the sum of the trip times corresponding to all the historical departure times and the historical departure places as the probability of the user going to the historical destination at the departure time and the departure place.

Step S23: and fitting the mixed Gaussian distribution curves of the plurality of historical places to obtain a travel habit model of the target user terminal.

In this embodiment, a plurality of mixed gaussian distribution curves of each user are fitted, so as to obtain a travel habit model for the user. When a destination which is possible to travel at present of the user needs to be predicted, the longitude and latitude and the current time of the user are only required to be input into a travel habit model corresponding to the user.

In this embodiment, the probabilistic model may also adopt a markov model, a gaussian mixture model, a bayesian model, etc., which is not described herein again.

In this embodiment, the following effects are obtained by using a gaussian model as a probability model and training the probability model based on the gaussian model:

firstly, the adopted Gaussian model belongs to a generative model and is a probability model related to the data distribution P (x, y), so that the output result has better interpretability.

Secondly, the probability model can summarize the behavior habits of the user based on the historical travel orders of the user, so that the travel habits of the user on various destinations can be observed more conveniently through the travel habit model obtained based on the probability model training.

And thirdly, the probability value output by the travel habit model has higher accuracy, so that the obtained travel habit features have higher accuracy, and after the travel habit features and the click rate features are input into the place recommendation model, the probability with higher accuracy can be output, and the accuracy of predicting and determining the place to be recommended is further improved.

In an implementation manner, with reference to the above embodiments, the present application further provides a method for obtaining a location recommendation model. FIG. 7 is a flow chart illustrating a method of obtaining a location recommendation model according to an embodiment of the present application. Referring to fig. 7, the present application may take the following steps to obtain a location recommendation model:

step S31: and determining the sample click rate characteristic of the sample user terminal according to the sample departure time and the sample departure place in the sample orders and a plurality of historical orders before the sample orders are generated by taking each historical order of the sample user terminal as a sample order.

In this embodiment, each user may generate a plurality of history orders, where a user generating a history order is referred to as a sample user, a terminal used by a sample user is referred to as a sample user terminal, and a history order generated by a sample user is referred to as a sample order. For each sample order of each user, the sample departure time, the sample departure location, and a plurality of historical orders before the sample order is generated in the sample order may be analyzed to obtain click rate characteristics for the sample order.

For example, user a generates 10 orders within a preset time period (orders X1-X10 in order of generation time), and if user a is taken as a sample user and the latest generated order X10 is taken as a sample order X10, the sample departure time, the sample departure location, and the historical orders before sample order X10 (i.e., orders X1-X9) in sample order X10 may be analyzed to obtain click rate characteristics for sample order X10. Similarly, if order X9 is to be taken as a sample order X9, then the sample departure time, sample departure location, and historical orders prior to sample order X9 (i.e., orders X1-X9) in sample order X9 may be analyzed to obtain a click rate characteristic for sample order X9.

Step S32: and determining a sample travel habit model of the sample user terminal according to the plurality of historical orders of the sample user terminal.

In this embodiment, each user (or sample user terminal) has a corresponding sample travel habit model, and therefore, the destination prediction platform can determine which sample travel habit model should be used to obtain the sample travel habit characteristics of the sample user according to the user corresponding to the historical order.

Step S33: inputting the sample departure time and the sample departure place in the sample order of the sample user terminal into the travel habit model of the sample user terminal, and determining the sample travel habit characteristics of the sample user terminal.

In this embodiment, when a place recommendation model is trained, not only a click rate feature but also a travel habit feature is required. Therefore, the sample departure time and the sample departure location in the sample order are also required to be input into the corresponding travel habit model of the sample user to determine the sample travel habit characteristics of the sample user.

For example, following the embodiment in step S31, when the order X10 is used as the sample order of the sample user a, the travel habit model corresponding to the sample user a may be input with the sample departure time and the sample departure location of the sample order X10, and the travel habit characteristics of the sample user a for the sample order X10 may be obtained. Similarly, when the order X9 is used as the sample order of the sample user a, the travel habit model corresponding to the sample user a may be input by the sample departure time and the sample departure location of the sample order X9, so as to obtain the travel habit characteristics of the sample user a for the sample order X9.

Step S34: and splicing the respective sample click rate characteristics and the sample travel habit characteristics of the sample user terminal to obtain a training sample.

In this embodiment, the sample travel habit characteristics and the sample click rate characteristics may be spliced according to a preset splicing rule, and a splicing result is used as an input of the model. The splicing rule may be direct splicing of vectors, for example, the travel habit feature is an M-dimensional vector, the sample click rate feature is an N-dimensional vector, and the spliced vector is an M + N-dimensional vector. The preset splicing rule can be set according to actual requirements, and this embodiment does not specifically limit this.

For example, following the above embodiments in step S31 and step S33, the click rate characteristic of the sample order X10 and the travel habit characteristic of the sample order X10 may be spliced to obtain a training sample, and the click rate characteristic of the sample order X9 and the travel habit characteristic of the sample order X9 may be spliced to obtain another training sample. In a similar principle, a plurality of training samples may be obtained.

Step S35: and training a click rate estimation model by using a plurality of training samples to obtain the place recommendation model.

In this embodiment, after obtaining a plurality of training samples, the click rate estimation model may be trained by using the plurality of training samples to obtain a location recommendation model. The click rate estimation model may adopt a logic Regression model, an XGBoost model, a light Gradient Boosting model, or other conventional machine learning models, or Deep learning models such as DNN (Deep Neural Networks), Wide & Deep, DIN (Deep Interest Networks), which is not specifically limited in this embodiment.

In the embodiment, the destination prediction task is modeled into a limited number of candidate destinations, click rate characteristics and travel habit characteristics of each destination are counted based on historical travel orders, and then the click rate characteristics and the travel habit characteristics are used for predicting the probability of clicking on each candidate destination by a user through a click rate prediction model.

In the embodiment, not only the commonly used click rate estimation characteristics are constructed during destination prediction, but also the probability function characteristics output by the probability model are spliced, and then the spliced characteristics are used as the input of the click rate estimation model, so that the probability that the user goes out to each destination is finally predicted. If a candidate destination does not have a probability function because orders are rare, the probability function feature is filled with a zero value. By the method, the characteristic that the click rate estimation model has low requirement on the order is fully utilized, and additional probability function characteristics are added for the destination prediction with enough order frequency, so that the interpretability of the result is optimized, and the accuracy of the destination prediction is improved.

With reference to the foregoing embodiments, in an implementation manner, the present application further provides an application scenario of predicted probabilities of each candidate location. Specifically, after obtaining the probability corresponding to each of the plurality of candidate locations, the method of the present application further includes:

In this embodiment, after predicting the probability of the target user going out to each candidate location, if the target user triggers a location input operation on the target user terminal, the destination prediction platform sends a plurality of candidate locations as candidate input locations to the target user terminal in the order from high to low of the probability for the target user to select.

In the embodiment, the candidate places are recommended to the user according to the predicted probability, so that the user can select places needing to be input from the recommended multiple candidate places according to actual requirements, and the use experience of the user is effectively enhanced.

With reference to the foregoing embodiments, in an implementation manner, the present application further provides another application scenario of predicted probabilities of respective candidate locations. Specifically, after the place to be recommended is pushed to the target user terminal, the method of the present application further includes:

The embodiment provides a use method of a predicted place to be recommended in a taxi taking scene. After the place to be recommended is pushed to the target user terminal, if the user selects the place to be recommended as the destination, the destination prediction platform can also send the get-off point to be recommended to the target user terminal and/or the driver terminal, and send prompt information to the target user terminal and/or the driver terminal to prompt the user of the target user terminal to get off at the get-off point to be recommended, so that the target user is prevented from missing the get-off point, and the use experience of the user can be effectively improved.

The method of place recommendation of the present application will be described in detail below with a specific embodiment.

In this embodiment, a gaussian model is used as the probability model, and a LightGBM model is used as the click rate estimation model. The whole process comprises the following steps: off-line model training and on-line model prediction.

One, off-line model training

The offline model training includes offline gaussian model training and LightGBM model training, as shown in fig. 8. Fig. 8 is a schematic diagram illustrating an offline model training process according to an embodiment of the present application. In fig. 8, a mixture gaussian model is first trained by using a historical order, then the mixture gaussian model obtained by training is stored in a mixture gaussian model library, and then the LightGBM model is trained based on the historical order and the output value of the mixture gaussian model, so as to obtain the LightGBM model.

1. Offline Gaussian model training

Fig. 9 is a schematic diagram illustrating an offline gaussian model training process according to an embodiment of the present application. Referring to fig. 9, the off-line gaussian model training may be performed using the following steps:

step 1: according to the user dimension and the destination dimension, the historical orders of the users are clustered, all the historical orders of the same destination of the same user are divided into a category, for example, for the user 1, the order reaching the destination A can be divided into a category, namely 'the user 1, the destination A', the order reaching the destination B can be divided into a category, namely 'the user 1, the destination B', and the order reaching the destination C can be divided into a category, namely 'the user 1, the destination C'.

Step 2: the Gaussian mixture model of the destination is fitted by using the information of the departure time and the departure place of all the historical orders in one category, for example, the Gaussian mixture model of user 1 and destination A is fitted according to the historical orders in user 1 and destination A, the Gaussian mixture model of user 1 and destination B is fitted according to the historical orders in user 1 and destination B, and the Gaussian mixture model of user 1 and destination C is fitted according to the historical orders in user 1 and destination C.

And step 3: and persistently storing the trained mixed Gaussian models of all the users in a model database.

2. LightGBM model training

Fig. 10 is a schematic diagram of a LightGBM model training process according to an embodiment of the present application. Referring to fig. 10, LightGBM model training may be performed using the following steps:

step 1: and (4) producing click rate characteristics by using the historical orders of the user, namely generating CTR characteristic training data by using the historical orders.

Step 2: and querying a Gaussian mixture model of the user by using the historical orders of the user, adding an output value of a probability function for each historical order as a probability characteristic, namely querying the Gaussian mixture model by using the historical orders to generate probability characteristic training data.

And step 3: and combining the click rate characteristic and the probability characteristic to obtain training data of the LightGBM model.

And 4, step 4: and training the LightGBM model by using the training data obtained by merging to obtain the LightGBM model.

Second, on-line model prediction

FIG. 11 is a schematic diagram illustrating online model prediction according to an embodiment of the present application. Referring to FIG. 11, online model prediction may be performed using the following steps:

step 1: when a destination prediction request of a target user is received, the time t and longitude and latitude information of the request are obtained.

Step 2: and acquiring all mixed Gaussian models of the target user.

And step 3: the probability of the destination is estimated through the Gaussian mixture model to obtain the probability characteristics, for example, the destination A and the probability characteristics are obtained through the Gaussian mixture model of the destination A, the destination B and the probability characteristics are obtained through the Gaussian mixture model of the destination B, and the destination C and the probability characteristics are obtained through the Gaussian mixture model of the destination C and the probability characteristics.

And 4, step 4: the click rate characteristics are constructed according to the time t and the latitude and longitude information, and for example, "destination a, CTR characteristics", "destination B, CTR characteristics" and "destination C, CTR characteristics" are constructed.

And 5: and combining the probability characteristic and the click rate characteristic, and predicting the destination trip probability through the LightGBM model, wherein the predicted destination trip probability is 'destination A, trip probability', 'destination B, trip probability' and 'destination C, trip probability'.

Step 6: and taking the destination with the highest probability as a destination prediction result.

In this embodiment, the probability function selection uses a mixed Gaussian model (GMM) output model, where the mixed Gaussian model is a model composed of multiple sets of Gaussian distributions, each Gaussian distribution is called a component, and the peak of each user trip can be described by one or more components. The probability density function of the Gaussian mixture model is the linear superposition of the probability densities of a plurality of Gaussian distributions, so that the multimodal situation can be well fitted. The Gaussian mixture model is a smooth continuous curve, so the recommendation is not critical for transitions. And by setting the number of the super-parameter components, the method can be degenerated into a single-peak Gaussian distribution when the historical orders of the user are fewer.

In the scene of solving the prediction of the destination click rate through the LightGBM model, the method adds the characteristic of a historical order fitting Gaussian mixture model. The mixed Gaussian model has the advantage of being capable of fitting multimodal travel habits of the user, and can depict various travel habits of the user on a certain destination, so that model prediction result interpretability is increased, accuracy of a destination prediction scene is improved, the requirement for order quantity by using probability function prediction alone is relieved, and the LightGBM can make reasonable travel probability prediction for the destination even under the condition that the destination has no probability function, and the destination prediction scene recall rate is improved.

Based on the same inventive concept, an embodiment of the present application provides an apparatus 1200 for location recommendation. Referring to fig. 12, fig. 12 is a block diagram illustrating a structure of a device for recommending a location according to an embodiment of the present application. As shown in fig. 12, the location recommendation apparatus 1200 includes:

an obtaining module 1201, configured to obtain a recommended location obtaining request initiated by a target user terminal, where the recommended location obtaining request carries a departure location and a departure time of the target user terminal;

a first determining module 1202, configured to determine click-through rate characteristics according to the departure location, the departure time, and the historical orders of the target user terminal, where the click-through rate characteristics at least include a distance between a historical location in the historical orders of the target user terminal and the departure location, a historical frequent trip time, and a trip frequency of the historical location;

a first input module 1203, configured to input the departure point and the departure time into a trip habit model of the target user terminal, and determine a trip habit feature, where the trip habit feature represents a probability that the target user terminal trips with the departure time as a historical departure time and the departure point as a historical departure point;

a second input module 1204, configured to input the trip habit features and the click rate features into a location recommendation model after being spliced, so as to obtain respective corresponding probabilities of multiple candidate locations;

a second determining module 1205, configured to determine a location to be recommended according to respective corresponding probabilities of the multiple candidate locations;

a first pushing module 1206, configured to push the place to be recommended to the target user terminal.

Optionally, the apparatus 1200 further comprises:

Based on the same inventive concept, another embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored, which when executed by a processor implements the steps in the method for location recommendation according to any of the above-mentioned embodiments of the present application.

Based on the same inventive concept, another embodiment of the present application provides an electronic device 1300, as shown in fig. 13. Fig. 13 is a schematic diagram of an electronic device according to an embodiment of the present application. The electronic device comprises a memory 1302, a processor 1301 and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the method for location recommendation according to any of the embodiments described above.

For the device embodiment, since it is basically similar to the method embodiment, the description is simple, and for the relevant points, refer to the partial description of the method embodiment.

The embodiments in the present specification are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other.

As will be appreciated by one of skill in the art, embodiments of the present application may be provided as a method, apparatus, or computer program product. Accordingly, embodiments of the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, embodiments of the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

Embodiments of the present application are described with reference to flowchart illustrations and/or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing terminal to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing terminal to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing terminal to cause a series of operational steps to be performed on the computer or other programmable terminal to produce a computer implemented process such that the instructions which execute on the computer or other programmable terminal provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

While preferred embodiments of the present application have been described, additional variations and modifications of these embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including the preferred embodiment and all such alterations and modifications as fall within the true scope of the embodiments of the application.

Finally, it should also be noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or terminal. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or terminal that comprises the element.

The method, the apparatus, the electronic device, and the storage medium for location recommendation provided by the present application are introduced in detail above, and a specific example is applied in the present application to explain the principle and the implementation of the present application, and the description of the above embodiment is only used to help understand the method and the core idea of the present application; meanwhile, for a person skilled in the art, according to the idea of the present application, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present application.

Claims

1. A method for location recommendation, the method comprising:

and pushing the place to be recommended to the target user terminal.

2. The method according to claim 1, wherein the travel habit model of the target user terminal is obtained according to the following steps:

3. The method of claim 2, wherein the location recommendation model is trained according to the following steps:

4. The method according to any one of claims 1-3, wherein after obtaining the probability that each of the plurality of candidate locations corresponds to, the method further comprises:

5. The method according to any one of claims 1 to 3, wherein after the place to be recommended is pushed to the target user terminal, the method further comprises:

6. An apparatus for location recommendation, the apparatus comprising:

7. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the method of location recommendation according to any one of claims 1-5.

8. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor when executing performs the steps of the method of location recommendation according to any one of claims 1-5.