Intermodal scheme recommendation method and device, training method and device and electronic equipment

Abstract

The embodiment of the invention provides a combined transport scheme recommendation method and device, a training method and device and electronic equipment, and relates to the technical field of traffic transportation, and the method can recommend richer and more reasonable combined transport schemes. The air-railway combined transportation scheme recommendation method comprises the steps of obtaining starting point information and arrival point information in response to user operation; calling a trained recommendation model, and obtaining M recommendation schemes corresponding to the starting point information and the arrival point information through the recommendation model; each scheme in the M recommendation schemes comprises transit point information and a combined transport mode; wherein M is a positive integer, and M is greater than or equal to 3; the trained recommendation model is obtained by adopting system combined transport data and autonomous combined transport data to form multi-source heterogeneous data and performing enhanced training on the recommendation model.

Description

Technical Field

[0001] The present application relates to the field of transportation technology, and in particular to a method for recommending an intermodal transport solution, a training method, a device and an electronic device. Background Art

[0002] The intermodal transport of passengers by rail and other modes of transportation can bring more convenience to passengers and improve the user experience. For example, air-rail intermodal transport is the combination of air transport and rail transport, providing passengers with more efficient and diversified travel options.

[0003] Currently, artificial intelligence can be used to recommend a variety of candidate intermodal transport options to passengers. However, due to the insufficient predictive performance of the model, the average number of recommended options is about 3, and users have little choice. Some more efficient and reasonable intermodal transport options may not be recommended, and user experience needs to be improved. Summary of the invention

[0004] In view of this, embodiments of the present application provide a method for recommending an intermodal transport solution, a training method, a device, and an electronic device to solve the above-mentioned problems.

[0005] In a first aspect, an embodiment of the present application provides a method for recommending an air-rail intermodal transport scheme, which is applied to an electronic device and may include: in response to a user operation, obtaining departure point information and arrival point information; calling a trained recommendation model, and obtaining M recommended schemes corresponding to the departure point information and the arrival point information through the recommendation model; each of the M recommended schemes includes transfer point information and an intermodal transport mode; wherein M is a positive integer, and M≥3; the intermodal transport mode includes that a passenger first takes a first means of transportation and then transfers to a second means of transportation; or, a passenger first takes a second means of transportation and then transfers to the first means of transportation; the first means of transportation and the second means of transportation are means of transportation based on different modes of transportation; wherein the trained recommendation model is obtained by performing enhanced training on the recommendation model using multi-source heterogeneous data composed of system intermodal transport data and autonomous intermodal transport data; the system intermodal transport data includes multiple intermodal transport orders; the autonomous intermodal transport data includes multiple groups of orders, one group of orders includes two non-intermodal transport orders corresponding to the same user; the time difference between the two non-intermodal transport orders is less than a predetermined duration, and the arrival point of one of the two non-intermodal transport orders and the departure point of the other order are in the same city.

[0006] In one possible implementation manner, after obtaining M recommended schemes corresponding to the departure point information and the arrival point information, the method further includes: calculating evaluation scores corresponding to the M recommended schemes respectively according to the intermodal mode coefficient, the proportion of the first distance to the sum of the second distance and the third distance, and the predicted values ​​corresponding to the M recommended schemes output by the recommendation model; wherein the first distance is the distance between the departure point and the arrival point, the second distance is the distance between the departure point and the transfer point; the third distance is the distance between the transfer point and the arrival point; the intermodal mode coefficient is determined according to whether the distance corresponding to the fastest means of transportation between the first means of transportation and the second means of transportation is greater than the distance corresponding to the other means of transportation; according to the evaluation scores, the M recommended schemes are sorted, and the recommended schemes with evaluation scores lower than the threshold are filtered out to obtain Q recommended schemes; wherein Q is a positive integer, M≥Q.

[0007] In one possible implementation, the method further includes: obtaining information of a priority recommended transit point; determining a constant corresponding to the priority recommended transit point; after obtaining M recommended schemes corresponding to the departure point information and the arrival point information, the method further includes: when determining that a transit point among the M recommended schemes is a transit point that needs to be prioritized, calculating an evaluation score corresponding to the recommended scheme including the priority recommended transit point based on the intermodal mode coefficient, the proportion of the first distance to the sum of the second distance and the third distance, the predicted value output by the recommendation model, and the constant corresponding to the priority recommended transit point.

[0008] In one possible implementation, after obtaining Q recommended plans, the method further includes: querying the schedule information corresponding to the first means of transportation and the second means of transportation respectively according to the intermodal transport mode of each plan in the Q recommended plans; and / or, respectively calculating the transfer information of transferring from the first means of transportation to the second means of transportation or from the second means of transportation to the first means of transportation at the transfer point in the Q recommended plans; the transfer information at least includes the transfer time.

[0009] In a second aspect, an embodiment of the present application also provides a model training method, which is applied to an electronic device; the method includes: acquiring system intermodal data and autonomous intermodal data to form multi-source heterogeneous data; wherein the system intermodal data includes multiple intermodal orders; the autonomous intermodal data includes multiple groups of orders, and a group of orders includes two non-intermodal orders corresponding to the same user; the time difference between the two non-intermodal orders is less than a predetermined duration, and the arrival point of one of the two non-intermodal orders is the same as the departure point of the other order; based on the multi-source heterogeneous data, multiple training samples and sample labels are obtained, and the recommendation model is enhanced trained using the multiple training samples and sample labels to obtain a trained recommendation model; the trained recommendation model is used to: predict M recommended schemes corresponding to the departure point information and the arrival point information; each of the M recommended schemes includes transfer point information and an intermodal mode; wherein M is a positive integer, and M≥3; the intermodal mode includes that the passenger first takes a first means of transportation and then transfers to a second means of transportation; or, the passenger first takes the second means of transportation and then transfers to the first means of transportation; the first means of transportation and the second means of transportation are means of transportation based on different modes of transportation.

[0010] In one possible implementation, the method further includes: combining X transfer point labels and Y intermodal transport mode labels corresponding to multiple training samples to obtain X*Y combined labels, and numbering the combined labels to obtain X*Y label IDs; the X*Y label IDs and the recommendation model output data include X*Y dimensions, representing X*Y categories; and the X*Y label IDs correspond one-to-one to the X*Y categories.

[0011] In one possible implementation, a training sample among multiple training samples includes multiple of the following information: departure point location information, arrival point location information, age, gender; or, also includes the departure time of a first means of transportation; the first means of transportation is the fastest means of transportation among at least two means of transportation that passengers take from the departure point to the arrival point.

[0012] On the third aspect, the embodiment of the present application also proposes an air-rail transport scheme recommendation device, which includes: a first recommendation module for obtaining departure point information and arrival point information in response to user operations; calling a trained recommendation model, and obtaining M recommended schemes corresponding to the departure point information and the arrival point information through the recommendation model; each of the M recommended schemes includes transfer point information and an intermodal transport mode; wherein M is a positive integer, and M≥3; the intermodal transport mode includes the passenger taking the first means of transportation first and then transferring to the second means of transportation; or the passenger taking the second means of transportation first and then transferring to the second means of transportation. Transfer to a first means of transportation; the first means of transportation and the second means of transportation are means of transportation based on different modes of transportation; wherein the trained recommendation model is obtained by performing enhanced training on the recommendation model using multi-source heterogeneous data composed of system intermodal transport data and autonomous intermodal transport data; the system intermodal transport data includes multiple intermodal transport orders; the autonomous intermodal transport data includes multiple groups of orders, one group of orders includes two non-intermodal transport orders corresponding to the same user; the time difference between the two non-intermodal transport orders is less than the predetermined duration, and the arrival point of one of the two non-intermodal transport orders and the departure point of the other order are in the same city.

[0013] In one possible implementation, the device also includes an optimization module; after obtaining M recommended schemes corresponding to the departure point information and the arrival point information, the optimization module is used to: calculate the evaluation scores corresponding to the M recommended schemes according to the intermodal mode coefficient, the proportion of the first distance to the sum of the second distance and the third distance, and the predicted values ​​corresponding to the M recommended schemes output by the recommendation model; wherein the first distance is the distance between the departure point and the arrival point, the second distance is the distance between the departure point and the transfer point; the third distance is the distance between the transfer point and the arrival point; the intermodal mode coefficient is determined according to whether the distance corresponding to the fastest means of transportation between the first means of transportation and the second means of transportation is greater than the distances corresponding to other means of transportation; according to the evaluation scores, the M recommended schemes are sorted, and the recommended schemes with evaluation scores lower than the threshold are filtered out to obtain Q recommended schemes; wherein Q is a positive integer, M≥Q.

[0014] In one possible implementation, the device also includes: a query module, which is used to query the schedule information corresponding to the first means of transportation and the second means of transportation respectively according to the intermodal transport mode of each scheme in the Q recommended schemes; and / or, a transfer calculation module, which is used to calculate the transfer information of transferring from the first means of transportation to the second means of transportation or from the second means of transportation to the first means of transportation at the transfer point in the Q recommended schemes; the transfer information at least includes the transfer time.

[0015] In one possible implementation, the device may further include a second recommendation module, which is used to add the schedule information output by the query module and the transfer information output by the transfer calculation module to the Q recommended solutions output by the optimization module to obtain Q complete recommended solutions. The second recommendation module may combine or sort the Q recommended solutions using strategies such as earliest arrival, shortest time, lowest cost, and high-speed rail priority.

[0016] In a fourth aspect, an embodiment of the present application further provides an electronic device, the electronic device comprising: a processor, the processor being used to execute a computer program or instruction in a memory to implement a method as described in any one of the first or second aspects above.

[0017] In a fifth aspect, an embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium includes a stored program, wherein when the program is executed by a processor, the method described in any one of the above items is implemented.

[0018] In a sixth aspect, an embodiment of the present application further provides a software program product, which includes a program. When the program is executed by an electronic device, the electronic device implements any of the methods described above.

[0019] In the seventh aspect, an embodiment of the present application also provides a chip system, including: a communication interface for inputting and / or outputting data; a processor for executing a computer executable program so that a device equipped with the chip system executes any of the methods described above. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0021] Figure 1 A schematic diagram of an intermodal transport solution recommended by a model in the related art;

[0022] Figure 2 A diagram of the software module architecture in some embodiments of the air-rail transport solution recommendation method provided in the embodiments of the present application;

[0023] Figure 3 A schematic diagram of using multi-source heterogeneous data to enhance the training of the recommendation model in some embodiments of the air-rail transport solution recommendation method provided in the embodiments of the present application;

[0024] Figure 4 A schematic diagram of label combinations in some embodiments of the air-rail transport solution recommendation method provided in the embodiments of the present application;

[0025] Figure 5 A schematic diagram of more intermodal transport solutions output by the enhanced trained model in some embodiments of the air-rail intermodal transport solution recommendation method provided in the embodiments of the present application;

[0026] Figure 6 A diagram of the software module architecture in other embodiments of the air-rail transport solution recommendation method provided by the embodiment of the present application;

[0027] Figure 7 A diagram of the software module architecture in some other embodiments of the air-rail transport solution recommendation method provided in the embodiment of the present application;

[0028] Figure 8 Flow charts of some embodiments of the air-rail transport solution recommendation method provided in the embodiments of the present application;

[0029] Fig. 9 Flowcharts of some embodiments of the model training method provided in the embodiments of the present application;

[0030] Fig.10 A schematic diagram of the software module architecture of the air-rail transport solution recommendation device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0031] In order to better understand the technical solution of the present application, the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0032] It should be clear that the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0033] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present application. The singular forms "a", "said" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings.

[0034] It should be understood that the term "and / or" used in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0035] Before introducing the embodiments of the present application, the related technologies and their technical problems are first explained.

[0036] Intermodal transport refers to the combination of two or more modes of transport. Among them, intermodal transport of passengers by rail and other modes of transport is a business type with high demand. In particular, air-rail intermodal transport, as an innovative travel mode, provides passengers with more convenient and efficient travel solutions.

[0037] At present, there are more and more studies and practices on air-rail transport, but there are many deficiencies in these studies and practices. For example, when machine learning models predict multiple plans, unreasonable plans may appear in the plans due to the influence of data imbalance and the limitations of evaluation functions.

[0038] At present, when using artificial intelligence to predict intermodal solutions suitable for different passengers and different itineraries, since air-rail intermodal is currently a new mode of transportation, the amount of historical order data is insufficient. Based on machine learning technology, when using historical actual intermodal orders as training data to train the recommendation model, due to the single data source and insufficient data types, it is difficult for the model to fully learn, and the trained model has certain limitations. The most important aspect of the limitation is the insufficient number of recommended solutions. For example, some models can only recommend 3 or fewer candidate intermodal solutions on average. For example, Figure 1 As shown, with city S as the starting point and city e as the arrival point, there are only two alternative intermodal transport solutions intelligently recommended by the relevant technology, namely, the path s→t1→e with the transit city t1 as the transit point, and the path s→t2→e with the transit city t2 as the transit point. The dotted line indicates that the mode of transportation is aviation and the means of transportation is airplane, and the solid line indicates that the mode of transportation is railway and the means of transportation is train (including high-speed rail). Users have fewer options, and other practically feasible solutions are not recommended.

[0039] In addition, if you want to obtain a model that meets performance standards, the evaluation method of the model is also one of the key factors. Some model evaluation methods can only evaluate the first recommended result (i.e. the first recommended intermodal solution). The evaluation mechanism is not comprehensive enough, and the role of the evaluation needs to be strengthened. In addition, due to the impact of data imbalance, the recommended solutions are not reasonable and have low usability.

[0040] In addition, when combining specific plans, it is necessary to refer to the transfer information, which requires the design and implementation of a complete framework for recommending air-rail transport plans to better ensure that the recommended plans meet the needs.

[0041] In view of this, an embodiment of the present application designs a method for recommending air-rail intermodal transport schemes, which is a time-space composite modular air-rail intermodal transport scheme real-time recommendation algorithm, which uses system intermodal transport data and autonomous intermodal transport data to form multi-source heterogeneous training data, and uses multi-source heterogeneous training data to train the recommendation model (a model with classification function), which can enhance the model performance, solve the recommendation limitations of the model, and make up for the problem of a single data source and insufficient data when only using system intermodal transport data. The trained model can recommend more than 3 intermodal transport schemes, and generally can recommend 5-10 or more intermodal transport schemes for users to choose from, or can be set to recommend 5-8 candidate intermodal transport schemes, providing users with more diverse and reasonable candidate intermodal transport schemes.

[0042] In addition, in the method proposed in the embodiment of the present application, modules such as an optimization module and a transfer calculation module can be added to make the real-time recommendation framework of the air-rail transport plan more perfect and the final recommendation result more accurate and reasonable.

[0043] It should be noted that the method proposed in the embodiment of the present application can be applied to intermodal transport scenarios of various combinations of different transport modes, for example, it can be applied to intermodal transport of railways and aviation, or intermodal transport of railways and highways, or intermodal transport of railways and ocean freight, or intermodal transport of aviation and ocean freight, etc. For ease of description, the following description will be mainly based on the intermodal transport scenario of railways as an example.

[0044] like Figure 2 As shown, the air-rail transport solution recommendation method proposed in the embodiment of the present application can be based on the following Figure 2 The software module architecture diagram shown is implemented. The software module architecture diagram includes multiple modules, namely a first recommendation module, an optimization module, a query module, a transfer calculation module and a second recommendation module.

[0045] The first recommendation module, or the overall plan recommendation module, is used to generate the path information and intermodal transport mode of the air-rail transport plan according to the determined departure point and arrival point, and at least to recommend the transfer points, intermodal transport modes and other contents of the air-rail transport.

[0046] Among them, the intermodal mode refers to the combination of multiple modes of transportation corresponding to multiple trips of the same user. For example, the intermodal mode includes the passenger taking the first means of transportation from the departure point to the transfer point, and then transferring to the second means of transportation from the transfer point to the arrival point; or the passenger first takes the second means of transportation from the departure point to the transfer point, and then transfers to the first means of transportation from the transfer point to the arrival point. Among them, the first means of transportation and the second means of transportation are means of transportation based on different modes of transportation, for example, the first means of transportation is a train based on the railway transportation mode, and the second means of transportation is an airplane based on the air transportation mode.

[0047] In the scenario of air and rail intermodal transport, when there is only one transfer point between the departure point and the arrival point, there are two intermodal transport modes: rail first, then air, or air first, then rail. That is, taking the train or high-speed rail for the first leg of the journey and the plane for the second leg is one intermodal transport mode; taking the plane for the first leg of the journey and the train or high-speed rail for the second leg is another intermodal transport mode.

[0048] The departure point and the arrival point are generally determined according to user operations. For example, the departure point and the arrival point may be received through an interactive interface, which is input or selected by the user.

[0049] It should be noted that in some embodiments of the present application, the departure point, arrival point and transfer point are generally cities, for example, cities at the county level or above. In other embodiments, the arrival point, departure point or transfer point may also be a railway station, airport, transportation hub, high-speed rail station or other site.

[0050] The first recommendation module can be implemented by a machine learning model with classification function. For example, a decision tree model suitable for multi-classification problems can be used to divide the data into different categories by building a tree structure.

[0051] Figure 2 The optimization module shown in , which can also be called the overall solution optimization module, further optimizes the transfer points and intermodal modes recommended by the first recommendation module, and further filters and sorts the multiple candidate solutions initially recommended by the first recommendation module, making the recommended candidate solutions more reasonable. At the same time, the priority recommendation strategy of the service provider can also be connected, that is, some designated cities can be recommended first to meet the needs of operations or to meet the diverse needs of users.

[0052] Among them, the query module can query the corresponding accessibility in the pre-established air-rail transport network based on the departure point information, transfer point information and arrival point information in the currently recommended multiple solutions. Accessibility is determined by judging whether there are actual railway schedules and flights. The air-rail transport network can be constructed by pre-obtaining the flight schedule information and railway schedule information of each transfer city. In this way, the specific schedule information in the intermodal transport solution can be directly obtained.

[0053] The transfer calculation module is used to provide transfer-related information, at least calculating the transfer time of switching different modes of transportation at the transfer point, for example, at least calculating the transfer time from high-speed rail to airplane or airplane to high-speed rail. Specifically, for example, the transfer information may include information such as transfer time, ticket price, and public transportation mode. The provided transfer information is used to provide support for the specific solution combination of the latter module (second recommendation module) according to the former module (optimization module).

[0054] The second recommendation module, also known as the specific plan recommendation module, combines the flight information obtained by the query module, the transfer information (such as transfer time) calculated by the transfer calculation module, and the transfer point information and intermodal transport methods obtained after optimization by the optimization module to obtain specific plans after multiple combinations, and can further obtain the optimal recommendation result (i.e., intermodal transport plan result) set according to the recommendation strategy.

[0055] Combination Figure 2 The software module architecture is described below with a specific implementation example.

[0056] In this embodiment, the first recommendation module is implemented using an XGBoost (eXtreme Gradient Boosting) model, an ensemble learning algorithm based on a gradient boosting decision tree.

[0057] First, the XGBoost model needs to be trained to obtain a model that can predict transfer points and intermodal transport methods based on the departure and arrival points.

[0058] like Figure 3 As shown, in this embodiment, multi-source heterogeneous data is first obtained, and the processed training data is obtained based on the multi-source heterogeneous data to train the XGBoost model. The multi-source heterogeneous data includes system intermodal data and autonomous intermodal data. The system intermodal data includes multiple actual intermodal orders. The autonomous intermodal data includes multiple groups of orders, one group of orders includes two non-intermodal orders corresponding to the same user, the time difference between the two non-intermodal orders is less than the predetermined time, and the arrival point of one of the two non-intermodal orders is in the same city as the departure point of the other order.

[0059] At present, there are some systems, such as the air-rail intermodal system, which can support users to directly select intermodal orders. An intermodal order includes two trips, and the two trips are respectively carried out by different means of transportation. For example, the first trip is by train (high-speed rail, intercity, train, etc.), and the second trip is by plane; or the first trip is by plane, and the second trip is by train. System intermodal data can be multiple intermodal orders generated by the air-rail intermodal system.

[0060] A non-intermodal order can be a one-way order, that is, the order only includes one section of the journey and only takes one mode of transportation. For example, an order corresponding to only one section of the journey by high-speed rail or only one section of the journey by plane can be regarded as a non-intermodal order.

[0061] S101: Acquire multi-source heterogeneous data.

[0062] In this embodiment, on the one hand, the actual system intermodal transport data within a period of time is directly obtained from the system providing intermodal transport services, for example, the system intermodal transport data (or product intermodal transport data) from October 2023 to December 2023 is obtained from the air-rail intermodal transport service system.

[0063] On the other hand, we obtained order data from October 2023 to December 2023 from the train ticket sales system and the air ticket sales system, and combined the data of the same passenger's two trips with a time difference less than the scheduled duration and the arrival point of the previous trip and the departure point of the next trip to form autonomous intermodal data according to the passenger travel dimension. A total of 444,562 system intermodal data and autonomous intermodal data were obtained.

[0064] For example, using the user's account name or identity ID as the index, match any two trips in multiple trips with the same index to see if they meet the following conditions. If they do, merge them into one set of data:

[0065] The arrival point of the first trip of a user (or passenger) and the departure point of the second trip are in the same city; the arrival point and departure point information can be obtained from the orders corresponding to the two trips;

[0066] The time difference between the actual arrival time of the first segment of the journey and the actual departure time of the second segment of the journey is less than the predetermined duration, which may be 2.5-5.5 hours, for example, 4 hours.

[0067] If the above two conditions are met, the two orders are combined into a set of data. This part of the data is based on the actual situation that the user has a transfer during the trip, but the user has not placed an order through the intermodal system. This part of the data can supplement the data source and make up for the defects of insufficient or single data source.

[0068] Based on the structure of product intermodal transport data, the autonomous intermodal transport data is processed into the same data format as the product intermodal transport data to obtain data in a unified format, in which each data item consists of train ticket order information and flight ticket order information.

[0069] For example, a train ticket order information may include one or more of the following items: travel date, departure train number, travel city, travel station, travel station telegraph code, departure date, arrival city, arrival station, arrival station telegraph code, arrival station bureau, arrival date, ticket type, fare, payment service serial number, mileage, user name, final status, statistical flag, transfer ticket type, creation time, etc.;

[0070] An airline ticket order information may include the following: order number, whether refunded, departure airport code, departure airport name in Chinese, departure terminal, departure city name, departure time, arrival airport code, arrival airport name in Chinese, arrival terminal, arrival city name, arrival time, duration, market carrier, market carrier flight number, date of birth, gender, transaction generation date, etc.

[0071] S102: Process and select multi-source heterogeneous data to obtain training sample data.

[0072] First, according to the departure and arrival times of trains and planes in the orders corresponding to each user in the autonomous intermodal transport data, the intermodal transport mode is calculated. There are two intermodal transport modes. If the arrival time of the train is earlier than the departure time of the plane, the intermodal transport mode is to take the train first and then the plane (rail first and then air); if the arrival time of the plane is earlier than the departure time of the train, the intermodal transport mode is to take the plane first and then the train (air first and then rail).

[0073] Next, we counted the transit cities in the autonomous intermodal transport data and product intermodal transport data, and there were a total of 161 cities.

[0074] The data is further processed to filter out data with the same departure point and arrival point, that is, the order data with the departure point and arrival point in the same city or station is eliminated. The same departure point and arrival point means a return or round trip along the same route.

[0075] For the departure and arrival point information, such as the departure or arrival city information, the longitude and latitude information is used to form coordinate data to facilitate the use of machine learning models; the age data is converted into a numeric type; for gender, OneHot encoding is used to form two independent columns, for example, 10 for male, 01 for female; or 10 for female, 01 for male. The departure time of the flight is fuzzy processed, for example, only the hour part is retained.

[0076] Then, the key fields are selected, and the selected key fields may include: departure point location information, arrival point location information, passenger age, passenger gender and other fields. Specifically, the location information may be longitude and latitude information, and the selected key fields may include departure city latitude, departure city longitude, arrival city latitude, arrival city longitude, age, gender, forming a vector or array or matrix as input data in a training sample. Alternatively, in other embodiments, the field of flight departure time may also be selected, that is, the selected key fields may include: departure city latitude, departure city longitude, arrival city latitude, arrival city longitude, age, gender, flight departure time and other fields.

[0077] Among them, airplanes are currently the fastest means of transportation. Due to the relatively high cost of flying (factors such as ticket prices and rebooking), as well as the longer waiting and departure time for flying than for other means of transportation, it has a great impact on passengers' choice of options. Therefore, the departure time of the airplane can be used as a key field as part of the input data in the training sample.

[0078] The label corresponding to a training sample may include a transit city label and an intermodal transport mode label.

[0079] S103: Perform enhanced training on the model.

[0080] In this embodiment, the two labels of transit city and intermodal transport mode are combined, that is, 161 transit cities are combined with 2 intermodal transport modes, and 322 categories are obtained. The 322 categories are numbered respectively, for example, numbers are numbered in ascending order and correspond to the 322 categories one by one, and 322 new digital type labels are generated to replace the original two labels. A total of 322 labels are generated, corresponding to 322 classification categories.

[0081] For example, Figure 4 As shown, 161 labels such as transit city labels "City A", "City B", and "City C" are combined with intermodal transport labels "train first, then plane" and "plane first, then train" respectively, and a total of 322 labels can be obtained. The 322 labels are numbered in sequence to obtain 322 digital labels. For example, the combined digital labels are represented by 9-bit binary codes. "City A" is combined with the intermodal transport label "train first, then plane" to obtain a combination label of 000000000 corresponding to the scheme of selecting City A as a transit city and the intermodal transport mode of first train and then plane. "City A" is combined with the intermodal transport label "plane first, then train" to obtain a combination label of 000000001 corresponding to the scheme of selecting City A as a transit city and the intermodal transport mode of first plane and then train. Using 9-bit binary codes to number the combined labels is only an example. In other embodiments, decimal or other bases may be used for numbering, which are not listed one by one in this specification.

[0082] The processed and selected data are divided into training set and test set in a 9:1 ratio, and the XGBoost model is trained and tested respectively.

[0083] During the training phase, when using XGBoost for regression, some parameters need to be set. These parameters include learning rate, number of trees (nesimators), maximum depth of trees (maxdepth), etc. The specific parameters that need to be set can include one or more of the following parameters:

[0084] Learning rate (eta): used to control the reduction of each tree's weight, and to reduce the impact of each tree on the final result. A smaller learning rate can make the model more stable, but the number of trees needs to be increased to maintain the model's expressiveness.

[0085] Number of trees (nesimators): Specifies the number of trees to build during training. Larger values ​​can improve the expressiveness of the model, but will also increase training time.

[0086] Maximum depth of tree (max_depth): specifies the maximum depth of each tree. Deeper trees can learn more complex features, but are also prone to overfitting. In this embodiment, the optimal maximum depth can be selected through cross-validation.

[0087] Exemplarily, in this embodiment, the tree maximum depth parameter (max_depth) is set to 4, the learning rate parameter (eta) is set to 0.1, and the number of trees parameter (nestimators) is set to 200, and XGBoost model training is performed.

[0088] After training, the performance of the model is evaluated using the accuracy function, where the accuracy is calculated as follows:

[0089]

[0090] Where TP represents the number of samples whose training samples are positive samples and whose prediction results are positive samples. TN represents the number of samples whose training samples are negative samples and whose prediction results are negative samples. FP represents the number of samples whose training samples are negative samples and whose prediction results are positive samples. FN represents the number of samples whose training samples are positive samples and whose prediction results are negative samples.

[0091] The final model achieved an accuracy of 87.23% and 85.62% on the training set and test set respectively, indicating that the model has learned the passenger selection patterns and has good generalization ability.

[0092] Experiments have shown that using multi-source heterogeneous data to train an enhanced model can recommend more options for passengers' air-rail intermodal travel, and the recommended intermodal travel options are more flexible and rich.

[0093] Specifically, compared with only product intermodal transport data, the total number of transit cities increased from 76 to more than 160, the number of recommended options each time increased from an average of 3 to more than 6, and the total number of accessible railway stations expanded from 2,000 to more than 8,000, significantly improving the range of options and accessibility for passengers.

[0094] The trained model is used for prediction to determine the departure city latitude, departure city longitude, arrival city latitude, arrival city longitude, user age, user gender and other information. After determination, it is used as input data and input into the trained XGBoost model. The XGBoost model will output a 322-dimensional vector or array, from which multiple elements with predicted score values ​​greater than the set threshold are selected. According to the position of multiple elements in the 322-dimensional vector, the city and intermodal mode corresponding to the element are determined. For example, Figure 4 In the label combination shown, the digital label corresponds to each element in the output vector one by one. According to the position of each element in the output 322-dimensional vector, the corresponding transit city and intermodal mode can be determined. For example, the digital label 0 corresponds to the first element in the output vector, so the transit city corresponding to the first element in the output 322-dimensional vector is Wuhan, and the intermodal mode is first train and then plane. The digital label corresponds to the fourth element in the output vector, so the transit city corresponding to the fourth element in the output vector is Chengdu, and the intermodal mode is first plane and then train.

[0095] The specific transit city and intermodal transport mode corresponding to each position in the 322-dimensional vector is determined based on the mapping relationship between the digital label obtained by combining the transit city label and the intermodal transport mode label and each element of the output vector. The numbering sequence is not limited to Figure 4 As shown, other orders may also be used, as long as the numbers are in one-to-one correspondence with the positions of the elements of the output vector, and the positions can be inversely mapped into transit cities and intermodal transport mode information through inverse mapping.

[0096] Assume that in the 322-dimensional vector predicted by the model, the value of each element is the predicted value corresponding to the 322 categories output by the model, and the predicted value can be used as the predicted score of the model output. Assume that the values ​​of 8 elements are not 0, and the values ​​of the other elements are all 0. The 8 non-0 elements are 0.42, 0.25, 0.15, 0.07, 0.05, 0.04, 0.01, and 0.01, respectively. The threshold is set to 0.02, then the 6 elements selected are 0.42, 0.25, 0.15, 0.07, 0.05, and 0.04. The transit cities and intermodal modes corresponding to these 6 elements are the 6 recommended options.

[0097] like Figure 5 As shown in the figure, similarly, assuming that s is the starting point and e is the arrival point, the enhanced trained model can predict at least 6 intermodal transport schemes. The predicted transit cities are t1, t2, ... t6, and the recommended 6 intermodal transport methods are as follows:

[0098] Plan 1: Transit city t1, path s→t1→e, intermodal mode: air-rail;

[0099] Plan 2: Transit city t2, path s→t2→e, intermodal mode: air-rail;

[0100] Plan 3: Transit city t3, path s→t3→e, intermodal mode: air-rail;

[0101] Plan 4: Transit city t4, path s→t4→e, intermodal mode: air-rail;

[0102] Plan 5: Transit city t5, path s→t5→e, intermodal transport mode: rail-air;

[0103] Plan 6: Transit city t6, path s→t6→e, intermodal transport mode: air-rail.

[0104] Compared to Figure 1 , significantly increasing the number of recommended options, diversifying user choices and making it easier for users to choose travel routes that better meet their actual needs.

[0105] Recommending solutions with good effects can not only meet the needs of passengers, but also speed up the calculation speed of the overall algorithm.

[0106] In this embodiment, for the multiple solutions output by the first recommendation module, the scores of the recommended solutions can be further optimized, and the calculation formula of the new score is as follows.

[0107]

[0108] Wherein is the predicted score of scheme t output in the first recommendation module (that is, the value of each element in the output 322-dimensional vector); k is the transportation mode coefficient. In this embodiment, when the predicted distance by plane is greater than the distance by train, k takes the value of 1, otherwise it takes the value of 0.5, that is, when the distance by plane is less than or equal to the distance by train, k=0.5; , and respectively represent the straight-line distance between the departure point s and the arrival point e, the straight-line distance between the departure point s and the transfer point t, and the distance between the transfer point t and the arrival point e.

[0109] According to the above formula, the new scores of all the solutions recommended by the first recommendation module are calculated, and then they are re-sorted from large to small to obtain a new queue.

[0110] Set the minimum recommendation score to filter out solutions with scores below the minimum recommendation score in the new queue. For example, set the minimum recommendation score to 0.01. When the score is less than this value, the solution will be filtered out.

[0111] It should be noted that this embodiment can support the configuration of priority recommended cities. In actual applications, in order to meet the diverse needs of users and the promotion needs of some cities, it is necessary to give priority to the recommendation of air-rail transport products in some cities, and then it is necessary to configure the priority level of the city. When the recommended solution includes the city, the city is extracted to the front of the queue, so as to achieve priority recommendation of the city.

[0112] For example, for cities that need to be recommended first, a recommendation constant can be set. Specifically, the score can be calculated using the following formula:

[0113]

[0114] Among them, C is the recommendation constant corresponding to the city that needs to be recommended first, for example, the recommended constant takes a value of 0.1-0.3, for example, a value of 0.2.

[0115] For example, assuming that the above six recommended solutions are optimized, the recommended queue is obtained as follows:

[0116] Plan 1: Transit city t1, path s→t1→e, intermodal mode: air-rail;

[0117] Plan 2: Transit city t2, path s→t2→e, intermodal mode: air-rail;

[0118] Plan 3: Transit city t3, path s→t3→e, intermodal mode: air-rail;

[0119] Plan 5: Transit city t5, path s→t5→e, intermodal transport mode: rail-air;

[0120] Plan 6: Transit city t6, path s→t6→e, intermodal transport mode: rail and air.

[0121] The original solution 4 was filtered out because the recalculated score was lower than the minimum score.

[0122] After the optimization module, the multiple initial solutions recommended by the first recommendation module are sorted and filtered. Next, the corresponding railway schedule information and civil aviation flight information need to be queried to calculate in advance whether the cities are reachable and store them in the cache. The reachability standard is determined by whether there is a connectable railway-to-air or air-to-rail solution.

[0123] Specifically, after determining the departure point, transfer point, arrival point and intermodal transport mode, you can obtain the railway schedule information or flight schedule information from the departure point to the transfer point, as well as the schedule information from the transfer point to the arrival point, and check whether there is corresponding schedule information that makes it reachable from the departure point to the arrival point according to the recommended transfer city and intermodal transport mode.

[0124] After the optimization module, you can also connect the transit calculation module.

[0125] The transfer calculation module sets the transfer information between all stations (train stations and high-speed rail stations) and airports based on the actual transfer experience within the transfer point. The transfer information includes at least the transfer time, and may also include information such as transfer transportation and fare cost. Among them, the transfer time includes the time to leave the station / airport, the time to take public transportation, the time to enter the station / airport, the time to wait for the train / plane, etc. The transfer time is set to ensure that passengers have enough time to realize the transfer of air-rail transport as much as possible.

[0126] The second recommendation module is used to combine the information obtained from the above multiple modules to obtain a specific recommendation scheme after combination. The first recommendation module only recommends transit cities and intermodal transport modes. The optimization module further sorts and optimizes the multiple schemes recommended by the first recommendation module to obtain multiple candidate schemes. The query module obtains the flight information corresponding to each candidate scheme through query, and the transfer calculation module calculates the transfer time and other information. The second recommendation module, based on the transfer time, selects the schemes that meet the time difference between the flight information corresponding to the two segments of the journey that is greater than the transfer time from the multiple candidate schemes output by the optimization module, and combines the information such as transit cities, intermodal transport modes, flight information, and transfer information to obtain a specific recommendation scheme after combination.

[0127] In other words, the second recommendation module (or specific plan recommendation module) searches and splices the specific plans of trains / planes based on the optimized and achievable recommended plans, and uses the transfer time of the transfer calculation module as the minimum time control of the combined plan to form all the specific plans for air-rail transport.

[0128] For example, the specific solution recommended by the second recommendation module is:

[0129] Plan 1: transfer city t1, path s→t1→e, intermodal mode: air (C*1133) rail (T28*), transfer time 2 hours;

[0130] Plan 2: transfer city t2, path s→t2→e, intermodal mode: air (**1001) rail (G50*), transfer time 3 hours;

[0131] Plan 3: Transit city t3, path s→t3→e, intermodal mode: air (C*1003) rail (G20*), transit time 3 hours;

[0132] Plan 5: Transit city t5, route s→t5→e, intermodal transport mode: rail (G67*) air (**917), transit time 1.5 hours;

[0133] Plan 6: Transit city t6, path s→t6→e, intermodal transport mode: rail (G28*) air (**606), transit time 2 hours.

[0134] Optionally, in this embodiment, the use of recommendation strategies can also be supported, and the recommendation strategies include one or more of a variety of strategies such as earliest arrival, shortest time, lowest cost or high-speed rail priority, so as to recommend the best specific plan to passengers.

[0135] The earliest arrival time is to calculate the arrival time of various plans respectively, sort them according to the arrival time, and give priority to the plan with the earliest arrival time.

[0136] The lowest cost means calculating the ticket prices of various options separately, and recommending the most cost-saving option in order of ticket prices from low to high.

[0137] The shortest time means calculating the total duration of multiple journeys of various options respectively, and recommending the option with the shortest journey time in order from least to most time.

[0138] It should be noted that an early arrival time does not necessarily mean a short travel time. The arrival time is the moment when the passenger arrives at the destination. The travel time in the shortest time strategy is the time difference between the moment when the passenger arrives at the destination and the departure time.

[0139] It should be noted that Figure 2 The software module architecture shown is only an example. In other embodiments, the optimization module is an optional module, such as Figure 6 In the software module architecture shown, the first recommendation module may be connected to the query module and the transfer calculation module respectively. The multiple schemes output by the first recommendation module may not be optimized, but directly calculate the transfer time and query the corresponding flight information to obtain a specific intermodal transport scheme.

[0140] Alternatively, in another embodiment, the transfer calculation module, the query module and the second recommendation module are all optional modules, for example Figure 7 As shown, after optimization, the scheme including the transfer point information and the intermodal transport mode information can be directly recommended to the user, without displaying the transfer time and flight information in the recommended scheme.

[0141] According to the above exemplary description, the air-rail transport solution recommendation method provided in the embodiment of the present application may include the following process:

[0142] S10: In response to a user operation, obtaining departure point information and arrival point information.

[0143] For example, as mentioned in the above exemplary description, the departure point information and the arrival point information can be obtained according to user operation, such as the user inputting the departure point and the arrival point information in the interactive interface, such as text input or voice input. Alternatively, the user can select in the interactive interface and select the departure point and the arrival point from multiple candidate cities.

[0144] S11: calling a trained recommendation model, and obtaining M recommendation solutions corresponding to the departure point information and the arrival point information through the recommendation model.

[0145] Each of the M recommended solutions includes transfer point information and a combined transport mode, wherein M is a positive integer, and M≥3. For example, in the above embodiment, M=6.

[0146] The intermodal transport mode includes that the passenger first takes the first means of transportation from the departure point to the transfer point, and then transfers to the second means of transportation from the transfer point to the arrival point; or, the passenger first takes the second means of transportation from the departure point to the transfer point, and then transfers to the first means of transportation from the transfer point to the arrival point; the first means of transportation and the second means of transportation are of different categories.

[0147] In other words, the first means of transportation and the second means of transportation are means of transportation based on different types of transportation modes. For example, the first means of transportation is various means of transportation under the railway transportation mode, specifically, it can be a high-speed rail, an intercity train, a direct train, an express train, etc., and the second means of transportation is a means of transportation under the air transportation mode, such as an airplane, and the intermodal transportation mode is rail-air (train first, then airplane), or air-rail (plane first, then train).

[0148] Alternatively, in other embodiments, the first means of transportation is a train and the second means of transportation is a ship; or the first means of transportation is a train and the second means of transportation is a public transportation such as a bus or a subway. The combined transportation mode may be a train first and then a ship, or a ship first and then a train; or a bus first and then a train, or a train first and then a bus, etc.

[0149] The recommendation model is trained using multi-source heterogeneous data (as training data) consisting of system intermodal data and autonomous intermodal data. As mentioned in the above exemplary description, the system intermodal data includes multiple actual intermodal orders; the autonomous intermodal data includes multiple groups of order combinations, one group of order combinations includes two orders corresponding to the same user; the time difference between the two orders is less than the predetermined time (for example, 4 hours), and the arrival point of one of the two orders is the same as the departure point of the other order.

[0150] The departure point information and the arrival point information obtained in S10 may be city names. After obtaining the departure point information and the arrival point information in S10 and before S11, the following steps may be performed:

[0151] According to the city name, the geographic coordinate information or longitude and latitude information of the departure point and the arrival point are obtained, and the passenger information is obtained. The passenger information may be information such as gender and age. In some embodiments, the departure time of the first means of transportation may also be obtained, and the departure time may be fuzzy processed to retain only the hour information. The first means of transportation may be the fastest means of transportation among multiple means of transportation, for example, the departure time of an airplane may be obtained.

[0152] Then, input data is composed based on the location information of the departure and arrival points, the passenger's gender, age and other information, and the flight departure time. The input data is input into the trained recommendation model, and the recommendation model outputs a multi-dimensional vector. For example, a 322-dimensional vector is output. The 322 dimensions correspond to 322 categories and also correspond to 322 combination labels.

[0153] Next, N recommended solutions are further selected from the M recommended solutions (preliminary solutions) and displayed to the user as the final recommendation results.

[0154] Each of the M recommendations may only include transfer point information and intermodal transport information. Each of the N recommendations may include: transfer point information and intermodal transport information, the schedule information corresponding to the first means of transport and the second means of transport, and transfer information (e.g., transfer time) from the first means of transport to the second means of transport at the transfer point, where N is a positive integer, N≤M, and N is greater than or equal to 3.

[0155] In some embodiments, after obtaining the M recommended solutions corresponding to the departure point information and the arrival point information, the M recommended solutions may also be evaluated in the following manner:

[0156] According to the intermodal coefficient, the proportion of the first distance to the sum of the second distance and the third distance, and the predicted values ​​corresponding to the M recommended solutions output by the recommendation model, the evaluation scores corresponding to the M recommended solutions are calculated.

[0157] Among them, the first distance is the distance between the starting point and the arrival point, the second distance is the distance between the starting point and the transfer point; the third distance is the distance between the transfer point and the arrival point.

[0158] The intermodal mode coefficient is determined based on whether the distance corresponding to the fastest means of transport between the first means of transport and the second means of transport is greater than the distances corresponding to the other means of transport; for example, in the calculation formula (2) in the above embodiment, k is 1 when the travel distance corresponding to the airplane is greater than the travel distance corresponding to the train, otherwise it takes the value of 0.5.

[0159] Next, the M recommended solutions are sorted according to the evaluation scores, and the recommended solutions with evaluation scores lower than the threshold are filtered out to obtain Q recommended solutions, where Q is a positive integer and M≥Q. For example, in the above embodiment, Q=5, and solution 4 is filtered out.

[0160] In some embodiments, priority recommendation is supported for some cities, and the priority of recommendation for different cities can be set, and different priorities correspond to different recommendation constants. In this part of the embodiments, the information of the transit point for priority recommendation can be obtained; the constant corresponding to the transit point for priority recommendation (i.e., the recommendation constant) can be determined; after obtaining the M kinds of recommended schemes corresponding to the departure point information and the arrival point information, it can also be determined that the transit point in the M kinds of recommended schemes is the transit point that needs to be recommended first, and the evaluation score corresponding to the recommended scheme containing the transit point for priority recommendation is calculated according to the intermodal mode coefficient, the proportion of the first distance to the sum of the second distance and the third distance, the predicted value output by the recommendation model, and the constant corresponding to the transit point for priority recommendation. For example, as shown in formula (3), the constant C in formula (3) is the recommendation constant. Different priorities can set different constant values, or a unified constant value can be used.

[0161] In some embodiments, after obtaining Q recommended solutions, the schedule information corresponding to the first means of transportation and the second means of transportation may be queried according to the intermodal transport mode of each of the Q recommended solutions.

[0162] In some embodiments, transfer information of transferring from a first means of transportation to a second means of transportation at a transfer point in the Q recommended solutions may also be calculated respectively; the transfer information at least includes the transfer time.

[0163] Next, N solutions can be selected from the Q solutions. Specifically, two flights and corresponding transfer points with a time difference greater than or equal to the transfer time can be selected from the Q transfer points according to the flight information and transfer information to obtain N solutions. N≤Q.

[0164] The present application also provides a model training method, which is applied to an electronic device. The method may include the following steps:

[0165] S20: Acquire system intermodal data and autonomous intermodal data to form multi-source heterogeneous data.

[0166] Among them, the system intermodal transport data includes multiple actual intermodal transport orders; the autonomous intermodal transport data includes multiple groups of order combinations, one group of order combinations includes two orders corresponding to the same user; the time difference between the two orders is less than the predetermined time, and the arrival point of one of the two orders is the same as the departure point of the other order.

[0167] S21: Based on multi-source heterogeneous data, multiple training samples and sample labels are obtained, and the recommendation model is enhanced by using the multiple training samples and sample labels to obtain a trained recommendation model.

[0168] The trained recommendation model has the following functions: predicting M recommendation solutions corresponding to the departure point information and the arrival point information.

[0169] In some embodiments, the present application also proposes a label combination method, combining X transit point labels and Y intermodal transport mode labels corresponding to multiple training samples to obtain X*Y combined labels, and numbering the combined labels to obtain X*Y label IDs; the X*Y label IDs and the recommendation model output data include X*Y dimensions, representing X*Y categories; the X*Y label IDs correspond one to one with the X*Y categories. For example, in the above embodiment, 161 transit city labels are combined with 2 intermodal transport mode labels to obtain 161*2=322 combined labels, and the dimension of the vector output by the recommendation model is also 322 dimensions, representing 322 categories. The combined label can be a digital label, and the number represents the position information or category information in the vector output by the recommendation model. According to the digital label corresponding to the position of each element in the output vector, the transit city and intermodal transport mode represented by the category corresponding to the position can be obtained.

[0170] In some embodiments, a training sample includes multiple types of the following information: departure point location information, arrival point location information, age, and gender. The departure point location information may include the departure point latitude and the departure point longitude, and the arrival point location information may include the arrival point latitude and the arrival point longitude. Alternatively, it also includes the departure time of the first means of transport. The first means of transport is the fastest means of transport among at least two means of transport that passengers take from the departure point to the arrival point. For example, among the two means of transport, airplanes and trains, airplanes are the fastest means of transport.

[0171] The present application also provides a device for recommending air-rail transport solutions. Fig.10 As shown, the device may only include the first recommendation module.

[0172] Specifically, the first recommendation module is used to obtain the departure point information and the arrival point information in response to the user operation; call the trained recommendation model, and obtain M recommendation schemes corresponding to the departure point information and the arrival point information through the recommendation model.

[0173] Among them, as mentioned in the above exemplary description, each of the M recommended plans includes transfer point information and intermodal transport mode; wherein, M is a positive integer, and M≥3; the intermodal transport mode includes that the passenger first takes the first means of transportation from the departure point to the transfer point, and then transfers to the second means of transportation from the transfer point to the arrival point; or, the passenger first takes the second means of transportation from the departure point to the transfer point, and then transfers to the first means of transportation from the transfer point to the arrival point; the first means of transportation and the second means of transportation are of different categories.

[0174] In some embodiments, the device may further include an optimization module; after obtaining the M recommended solutions corresponding to the departure point information and the arrival point information, the optimization module is used to implement the following functions: according to the intermodal coefficient, the proportion of the first distance to the sum of the second distance and the third distance, and the predicted values ​​corresponding to the M recommended solutions output by the recommendation model, the evaluation scores corresponding to the M recommended solutions are calculated. And, according to the evaluation scores, the M recommended solutions are sorted, and the recommended solutions with evaluation scores lower than the threshold are filtered out to obtain Q recommended solutions; wherein Q is a positive integer, and M≥Q.

[0175] In some embodiments, the device further includes a query module and / or a transit calculation module.

[0176] The query module is used to query the schedule information corresponding to the first means of transportation and the second means of transportation respectively according to the intermodal transport mode of each scheme in the Q recommended schemes.

[0177] The transfer calculation module is used to calculate the transfer information of transferring from the first means of transportation to the second means of transportation at the transfer point in the Q recommended solutions respectively; the transfer information at least includes the transfer time.

[0178] In summary, a spatiotemporal composite modular air-rail transport scheme real-time recommendation algorithm is designed in the embodiment of the present application. First, multi-source heterogeneous data is used to enhance the model training to obtain more diverse candidate recommendation schemes. Secondly, an overall scheme optimization module (i.e., the optimization module in the above exemplary description) and a transit calculation module are added to the algorithm to further improve the real-time recommendation framework of air-rail transport schemes and solve the problem of unreasonable schemes and unreasonable combinations in the recommendation results. Among them, multi-source heterogeneous data enriches the quantity and type of data, thereby realizing the enhancement of the overall recommendation model of air-rail transport, solving the recommendation limitation problem of the model, and optimizing the overall scheme recommendation module (i.e., the first recommendation module).

[0179] It brings a more complete framework and obtains more reliable recommendation results. This modular structure has good scalability and supports optimization within the module and adjustment between modules. This time-space composite design is relatively simple for passengers to operate and use, and the system recommendation quality is high and the speed is fast. At the same time, the enhanced overall recommendation model for air-rail transport has greatly increased the number of recommended transit cities, further meeting the diverse travel needs of passengers.

[0180] An embodiment of the present application further provides an electronic device, the electronic device comprising: a processor, the processor being used to execute a computer program or instruction in a memory to implement a method as described in any of the above embodiments.

[0181] The electronic device involved in the present application may be one or more of the following devices: a smart phone, a portable computer (Tablet Personal Computer, Tablet PC), a laptop computer (laptop), a desktop computer (Desktop computer), a wearable device, an augmented reality (AR), a virtual reality (VR), a mixed reality (MR) and other extended reality (XR) devices, an ultra-mobile personal computer (UMPC), a netbook or a personal digital assistant (PDA) and other devices. The embodiments of the present application do not impose any special restrictions on the specific types of electronic devices.

[0182] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in this application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integration. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk Solid State Disk), etc.

[0183] In the embodiments of the present application, "at least one" refers to one or more, and "plurality" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. Among them, A and B can be singular or plural. The character " / " generally indicates that the previous and subsequent associated objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b and c can be represented by: a, b, c, ab, ac, bc, or abc, where a, b, c can be single or multiple.

[0184] The above are only preferred embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for recommending air-rail transport solutions, characterized in that: The method is applied to an electronic device, and the method comprises: In response to user operations, obtaining departure point information and arrival point information; Calling a trained recommendation model, and obtaining M recommendation schemes corresponding to the departure point information and the arrival point information through the recommendation model; each of the M recommendation schemes includes transfer point information and an intermodal transport mode; wherein M is a positive integer, and M≥3; the intermodal transport mode includes that a passenger first takes a first means of transportation and then transfers to a second means of transportation; or, a passenger first takes a second means of transportation and then transfers to the first means of transportation; the first means of transportation and the second means of transportation are means of transportation based on different modes of transportation; Among them, the trained recommendation model is obtained by enhancing the training of the recommendation model by using multi-source heterogeneous data composed of system intermodal data and autonomous intermodal data; the system intermodal data includes multiple intermodal orders; the autonomous intermodal data includes multiple groups of orders, one group of orders includes two non-intermodal orders corresponding to the same user; the time difference between the two non-intermodal orders is less than the predetermined duration, and the arrival point of one of the two non-intermodal orders and the departure point of the other order are in the same city.

2. The method according to claim 1, characterized in that: After obtaining M recommended solutions corresponding to the departure point information and the arrival point information, the method further includes: According to the intermodal coefficient, the proportion of the first distance to the sum of the second distance and the third distance, and the predicted values ​​corresponding to the M recommended solutions output by the recommendation model, the evaluation scores corresponding to the M recommended solutions are calculated; wherein the first distance is the distance between the departure point and the arrival point, the second distance is the distance between the departure point and the transfer point; the third distance is the distance between the transfer point and the arrival point; the intermodal coefficient is determined according to whether the distance corresponding to the fastest means of transportation between the first means of transportation and the second means of transportation is greater than the distance corresponding to the other means of transportation; According to the evaluation scores, the M recommended solutions are sorted, and the recommended solutions with evaluation scores lower than a threshold are filtered out to obtain Q recommended solutions; wherein Q is a positive integer, and M≥Q.

3. The method according to claim 2, characterized in that The method further comprises: Get information on recommended transit points; Determine the constant corresponding to the preferred transfer point; After obtaining M recommended solutions corresponding to the departure point information and the arrival point information, the method further includes: When it is determined that a transfer point among the M recommended schemes is a transfer point that needs to be recommended first, the evaluation score corresponding to the recommended scheme including the preferred transfer point is calculated according to the intermodal mode coefficient, the proportion of the first distance to the sum of the second distance and the third distance, the predicted value output by the recommendation model, and the constant corresponding to the preferred transfer point.

4. The method according to claim 2 or 3, characterized in that: After obtaining Q recommended solutions, the method further includes: According to the intermodal transport modes of each of the Q recommended solutions, query the flight information corresponding to the first transportation tool and the second transportation tool respectively; and / or, The transfer information of transferring from the first means of transportation to the second means of transportation or from the second means of transportation to the first means of transportation at the transfer point in the Q recommended solutions is calculated respectively; the transfer information at least includes the transfer time.

5. A model training method, characterized in that: The method is applied to electronic equipment; the method comprises: Acquire system intermodal transport data and autonomous intermodal transport data to form multi-source heterogeneous data; wherein the system intermodal transport data includes multiple intermodal transport orders; the autonomous intermodal transport data includes multiple groups of orders, one group of orders includes two non-intermodal transport orders corresponding to the same user; the time difference between the two non-intermodal transport orders is less than a predetermined time length, and the arrival point of one of the two orders and the departure point of the other order are in the same city; Based on the multi-source heterogeneous data, a plurality of training samples and sample labels are obtained, and the plurality of training samples and sample labels are used to perform enhanced training on the recommendation model to obtain a trained recommendation model; The trained recommendation model is used to: Predict M recommended plans corresponding to the departure point information and the arrival point information; each of the M recommended plans includes transfer point information and an intermodal transport mode; wherein M is a positive integer, and M≥3; the intermodal transport mode includes a passenger first taking a first means of transportation and then transferring to a second means of transportation; or a passenger first taking the second means of transportation and then transferring to the first means of transportation; the first means of transportation and the second means of transportation are means of transportation based on different modes of transportation.

6. The method according to claim 5, characterized in that The method further comprises: Combine the X transfer point labels and Y intermodal transport mode labels corresponding to multiple training samples to obtain X Y combination labels, and number the combination labels to get X Y tag IDs; the X Y tag IDs and the recommended model output data include X Y dimension, representing X Y categories; X Y tag IDs with the X Y categories correspond one to one.

7. The method according to claim 5, characterized in that A training sample among the multiple training samples includes multiple types of the following information: Departure point location information, arrival point location information, age, gender; Alternatively, it also includes the departure time of a first means of transportation; the first means of transportation is the fastest means of transportation among at least two means of transportation that the passenger takes from the departure point to the arrival point.

8. An air-rail transport scheme recommendation device, characterized in that: The device comprises: A first recommendation module, the first recommendation module is used to obtain departure point information and arrival point information in response to user operations; call a trained recommendation model, and obtain M recommendation schemes corresponding to the departure point information and the arrival point information through the recommendation model; each of the M recommendation schemes includes transfer point information and an intermodal transport mode; wherein M is a positive integer, and M≥3; the intermodal transport mode includes that the passenger first takes the first means of transportation and then transfers to the second means of transportation; or the passenger first takes the second means of transportation and then transfers to the first means of transportation; the first means of transportation and the second means of transportation are means of transportation based on different modes of transportation; Among them, the trained recommendation model is obtained by enhancing the training of the recommendation model by using multi-source heterogeneous data composed of system intermodal data and autonomous intermodal data; the system intermodal data includes multiple intermodal orders; the autonomous intermodal data includes multiple groups of orders, one group of orders includes two non-intermodal orders corresponding to the same user; the time difference between the two non-intermodal orders is less than the predetermined duration, and the arrival point of one of the two non-intermodal orders and the departure point of the other order are in the same city.

9. The device according to claim 8, characterized in that The device further includes an optimization module; after obtaining M recommended solutions corresponding to the departure point information and the arrival point information, the optimization module is used to: According to the intermodal coefficient, the proportion of the first distance to the sum of the second distance and the third distance, and the predicted values ​​corresponding to the M recommended solutions output by the recommendation model, the evaluation scores corresponding to the M recommended solutions are calculated; wherein the first distance is the distance between the departure point and the arrival point, the second distance is the distance between the departure point and the transfer point; the third distance is the distance between the transfer point and the arrival point; the intermodal coefficient is determined according to whether the distance corresponding to the fastest means of transportation between the first means of transportation and the second means of transportation is greater than the distance corresponding to the other means of transportation; According to the evaluation scores, the M recommended solutions are sorted, and the recommended solutions with evaluation scores lower than a threshold are filtered out to obtain Q recommended solutions; wherein Q is a positive integer, and M≥Q.

10. The device according to claim 9, characterized in that The device also includes: A query module, used for querying the schedule information corresponding to the first transportation means and the second transportation means respectively according to the intermodal transport mode of each of the Q recommended solutions; and / or, The transfer calculation module is used to calculate the transfer information of transferring from the first means of transportation to the second means of transportation or from the second means of transportation to the first means of transportation at the transfer point in the Q recommended solutions respectively; the transfer information at least includes the transfer time.

11. An electronic device, characterized in that: The electronic device comprises: A processor, wherein the processor is configured to execute a computer program or instruction in a memory to implement the method according to any one of claims 1 to 4 or 5 to 7.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein the program, when executed by a processor, implements the method according to any one of claims 1-4 or 5-7.

13. A computer program product, characterized in that The computer program product includes a program, and when the program is executed by an electronic device, the electronic device implements the method according to any one of claims 1 to 4 or 5 to 7.

Images