Route recommendation method, device, storage medium, electronic device and product

Abstract

This application discloses a route recommendation method, device, storage medium, electronic device, and product, relating to the field of artificial intelligence technology. This application can be applied to technologies such as map-based vehicle networking and blockchain. The method includes: obtaining travel mode information, travel route information, and route status information corresponding to a target object; combining the travel mode information and travel route information into combined travel information; performing feature extraction processing on the combined travel information to obtain target travel feature information; fusing the target travel feature information with the route status information to obtain fused feature information; and performing prediction processing based on the fused feature information to obtain predicted target travel information, wherein the target travel information is used to indicate a travel route recommended for the target object under the travel mode. This application can improve the accuracy and effectiveness of route recommendations.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and specifically to a route recommendation method, device, storage medium, electronic device, and product. Background Art

[0002] Route recommendation is the process of recommending travel routes. For example, on a map platform, travel methods and routes can be recommended for navigation objects. Currently, related technologies use a simple process to process the object's relevant information and then perform a multi-step analysis to determine the recommended travel method and route.

[0003] In the current related technologies, due to the unreliability of relevant information and simple information processing, there is a problem of low accuracy in the analysis of travel modes and travel routes, resulting in poor route recommendation effects. Summary of the Invention

[0004] The embodiments of the present application provide a route recommendation method and related devices, which can effectively improve the accuracy of route recommendations and enhance the effect of route recommendations.

[0005] To solve the above technical problems, the embodiments of the present application provide the following technical solutions:

[0006] According to one embodiment of the present application, a route recommendation method includes: obtaining travel mode information, travel route information and route status information corresponding to a target object; combining the travel mode information and the travel route information into combined travel information; performing feature extraction processing on the combined travel information to obtain target travel feature information; fusing the target travel feature information with the route status information to obtain fused feature information; performing prediction processing based on the fused feature information to obtain predicted target travel information, wherein the target travel information is used to indicate a travel route under the travel mode recommended to the target object.

[0007] According to one embodiment of the present application, a route recommendation device includes: an acquisition module for acquiring travel mode information, travel route information and route status information corresponding to a target object; a combination module for combining the travel mode information and the travel route information into combined travel information; an extraction module for performing feature extraction processing on the combined travel information to obtain target travel feature information; a fusion module for performing fusion processing on the target travel feature information and the route status information to obtain fused feature information; and an analysis module for performing prediction processing based on the fused feature information to obtain predicted target travel information, wherein the target travel information is used to indicate a travel route under the travel mode recommended to the target object.

[0008] In some embodiments of the present application, the extraction module is used to: perform convolution processing on the combined travel information based on the preset convolution matrix to obtain the target travel characteristic information, each element in the preset convolution matrix corresponds to two predetermined travel modes, and each element is the weight of the object in the sample object set selecting the two predetermined travel modes corresponding to the element within a predetermined time period.

[0009] In some embodiments of the present application, the device also includes a generation module for a preset convolution matrix, including: an object determination unit, used to determine the object corresponding to each of the predetermined travel modes from the sample object set; a weight determination unit, used to determine the weight of the objects of each preset travel mode group selected within a predetermined time period according to the objects corresponding to each of the predetermined travel modes, wherein two of the predetermined travel modes form a preset travel mode group; a matrix generation unit, used to generate a matrix element for the weight corresponding to each of the preset travel mode groups to obtain the preset convolution matrix.

[0010] In some embodiments of the present application, the weight determination unit is used to: obtain the number of objects that select each preset travel mode group within a predetermined time period based on the objects corresponding to each of the predetermined travel modes; calculate the ratio of the number of objects that select each preset travel mode group to the target number to obtain the weight corresponding to each of the preset travel mode groups, and the target number is the total number of objects corresponding to all the predetermined travel modes.

[0011] In some embodiments of the present application, the fusion module includes: a pre-fusion unit, used to perform a Cartesian product operation on the target travel feature information and the route status information to obtain pre-fusion feature information; and an information generation unit, used to generate the fusion feature information based on the pre-fusion feature information.

[0012] In some embodiments of the present application, the information generating unit is configured to: perform a fully connected operation on the pre-fused feature information to obtain operation result feature information; and use the operation result feature information as the fused feature information.

[0013] In some embodiments of the present application, the travel mode information includes the operation information of the target object on each target travel mode in the device, and the travel route information includes the operation information of the target object on each target travel route in the device; the combination module is used to: perform Cartesian product operation on the operation information corresponding to each target travel mode and the operation information corresponding to the target travel route under each target travel mode to obtain the combined operation information under each target travel mode; and generate the combined travel information based on the combined operation information under each target travel mode.

[0014] In some embodiments of the present application, the analysis module includes: a model analysis unit, which is used to use a route analysis model to perform prediction processing based on the fused feature information to obtain a prediction result; and a result analysis unit, which is used to obtain predicted target travel information based on the prediction result.

[0015] In some embodiments of the present application, the route analysis model is a multi-classification model, and the prediction result includes at least one score, each score corresponds to a type of travel information, and each travel information includes a travel mode and a travel route; the result analysis unit is used to: determine the travel information whose score meets the preset recommendation conditions based on the score corresponding to each travel information; and use the travel information that meets the preset recommendation conditions as the target travel information.

[0016] In some embodiments of the present application, the device also includes a training module for a route analysis model, which is used to: obtain sample information and scheduled travel information corresponding to each object in the target object set, the sample information including travel mode information, travel route information and route status information, and the scheduled travel information is used to indicate the scheduled travel mode and travel route; combine the travel mode information corresponding to each object with the travel route information to obtain the combined travel information corresponding to each object; perform feature extraction processing on the combined travel information corresponding to each object to obtain the target travel feature information corresponding to each object; fuse the target travel feature information corresponding to each object with the route status information to obtain the fused feature information corresponding to each object; use the target analysis model to perform prediction processing based on the fused feature information corresponding to each object to obtain the predicted travel information corresponding to each object, and the predicted travel information is used to indicate the predicted travel mode and travel route; adjust the parameters in the target analysis model according to the predicted travel information and the scheduled travel information to obtain the trained route analysis model.

[0017] In some embodiments of the present application, the route status information includes status information corresponding to each target travel route, and the status information includes at least one of route traffic condition information, cost information, travel time information, speed limit information, and queue length information.

[0018] According to another embodiment of the present application, a computer-readable storage medium stores a computer program thereon. When the computer program is executed by a processor of a computer, the computer is caused to execute the method described in the embodiment of the present application.

[0019] According to another embodiment of the present application, an electronic device includes: a memory storing a computer program; and a processor reading the computer program stored in the memory to execute the method described in the embodiment of the present application.

[0020] According to another embodiment of the present application, a computer program product or computer program includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in various optional implementations described in the embodiments of the present application.

[0021] In an embodiment of the present application, travel mode information, travel route information and route status information corresponding to a target object are obtained; the travel mode information and the travel route information are combined into combined travel information; feature extraction processing is performed on the combined travel information to obtain target travel feature information; the target travel feature information and the route status information are fused to obtain fused feature information; prediction processing is performed based on the fused feature information to obtain predicted target travel information, and the target travel information is used to indicate a travel route under the travel mode recommended to the target object.

[0022] In this way, by combining the travel mode information, travel route information and route status information corresponding to the target object, performing feature extraction processing and fusion processing to generate fused feature information, the target travel information can be accurately predicted based on the fused feature information. According to the target travel information, the travel mode and travel route under the travel mode recommended to the target object can be accurately indicated, thereby effectively improving the accuracy of route recommendation and improving the route recommendation effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0024] Figure 1 A schematic diagram of a system to which embodiments of the present application can be applied is shown.

[0025] Figure 2 A flowchart of a route recommendation method according to an embodiment of the present application is shown.

[0026] Figure 3 A flowchart of performing route recommendation according to a related art is shown.

[0027] Figure 4 A flowchart of applying the implementation of the present application to perform route recommendation in a scenario is shown.

[0028] Figure 5Show Figure 4 Framework diagram of the route recommendation system for corresponding scenarios.

[0029] Figure 6 A block diagram of a route recommendation device according to another embodiment of the present application is shown.

[0030] Figure 7 A block diagram of an electronic device according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0031] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0032] It can be understood that in the specific implementation of this application, related data such as travel mode information, travel route information and route status information are involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0033] Figure 1 FIG. 1 shows a schematic diagram of a system 100 to which an embodiment of the present application can be applied. Figure 1 As shown, the system 100 may include a terminal 101 and a server 102 .

[0034] Terminal 101 can be any device, including but not limited to mobile phones, computers, intelligent voice interaction devices, smart home appliances, car terminals, VR / AR devices, smart watches, computers, etc.

[0035] Server 102 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers. It can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, as well as big data and artificial intelligence platforms.

[0036] In some implementations, the terminal 101 and the server 102 may be node devices in a blockchain network.

[0037] In one implementation of this example, the terminal 101 or the server 102 may: obtain travel mode information, travel route information, and route status information corresponding to the target object; combine the travel mode information and the travel route information into combined travel information; perform feature extraction processing on the combined travel information to obtain target travel feature information; fuse the target travel feature information with the route status information to obtain fused feature information; perform prediction processing based on the fused feature information to obtain predicted target travel information, where the target travel information is used to indicate the travel route under the travel mode recommended to the target object.

[0038] In some implementations, travel mode information, travel route information, and route status information may be obtained from a map vehicle networking platform (such as a map application).

[0039] Figure 2 The flowchart of the route recommendation method according to an embodiment of the present application is schematically shown. The execution subject of the route recommendation method can be any terminal, such as Figure 1 The terminal 101 or server 102 is shown.

[0040] like Figure 2 As shown, the route recommendation method may include steps S210 to S250.

[0041] Step S210, obtaining travel mode information, travel route information, and route status information corresponding to the target object;

[0042] Step S220, combining the travel mode information and the travel route information into combined travel information;

[0043] Step S230, performing feature extraction processing on the combined travel information to obtain target travel feature information;

[0044] Step S240: fusing the target travel feature information with the route status information to obtain fused feature information;

[0045] Step S250 , performing prediction processing based on the fused feature information to obtain predicted target travel information, where the target travel information is used to indicate a travel route under a travel mode recommended to the target object.

[0046] The target object is the object for which the travel mode and travel route are to be recommended, such as a user who searches for a navigation route on a map platform.

[0047] Travel mode information refers to information related to travel modes, such as driving, taxi, bus, subway, walking, cycling, etc. in the map. Travel mode information includes operation information of the target object for each target travel mode in the map on the device.

[0048] Travel route information refers to information related to travel routes. A travel route is a route between a starting point and a destination. Each travel mode can include multiple (e.g., three) travel routes. For example, walking can correspond to three different travel routes. Travel route information includes, for example, information about the target subject's actions on each target travel route on a map on the device.

[0049] Route status information refers to the status information of each travel route. The travel route status information may include the status information corresponding to the target travel route. The status information includes at least one of the following information: route traffic condition information, cost information, travel time information, speed limit information, queue length information, etc.

[0050] By detecting the operation of a target object on a device, it is possible to determine at least one target travel mode corresponding to the target object and the target route for each target travel mode, and obtain travel mode information for the target travel mode and travel route information for the target travel route. Furthermore, real-time route status information for each target travel route can be obtained from road network-related devices associated with the travel route.

[0051] After obtaining the travel mode information, travel route information and route status information corresponding to the target object, fusion feature information is generated through combination, feature extraction processing and fusion processing, and prediction processing is performed based on the fusion feature information to obtain target travel information. The target travel information can indicate the travel route under the travel mode recommended to the target object.

[0052] In this way, based on steps S210 to S250, by combining the travel mode information, travel route information and route status information corresponding to the target object, performing feature extraction processing and fusion processing to generate fused feature information, the target travel information can be accurately predicted based on the fused feature information, and the travel mode and travel route recommended to the target object under the travel mode can be accurately indicated according to the target travel information, thereby effectively improving the accuracy of route recommendation and improving the route recommendation effect.

[0053] The following describes the specific process of each step when making route recommendations.

[0054] In step S210, the travel mode information, travel route information and route status information corresponding to the target object are obtained.

[0055] If the target object separately agrees to the authorization, the operations of the target object in the device (such as clicking or inputting, etc., the input may include any input method, such as voice input or text input, etc.) are detected, and at least one target travel mode corresponding to the target object and the target route under each target travel mode can be determined, and the travel mode information of the target object under the target travel mode, the travel route information under the target travel route, and the route status information under the target travel route can be obtained.

[0056] For example, the target object can perform operations such as clicking and input in the map application. The target object can enter the starting point and end point of the trip. The map can generate multiple target travel modes under the starting point and end point (such as driving, taxi, bus, subway, walking, cycling, etc.) and multiple target travel routes under each target travel mode.

[0057] By detecting the target object's actions for each target travel mode in a map application, travel mode information can be obtained. By detecting the target object's actions for each target travel route in a map, travel route information can be obtained. Furthermore, real-time route status information for each target travel route can be obtained from road network-related systems.

[0058] In one embodiment, the travel mode information may include at least one of the following operation information: click behavior under the travel mode plan, input starting point, input end point, user travel history information, user active behavior on the APP / mini program side, module payment behavior, module payment type, etc.

[0059] In one embodiment, the travel route information may include at least one of the following operation information: clicking on the travel route plan, clicking on the route exploration behavior, exiting the route exploration behavior, clicking on the real-scene navigation information, exiting the real-scene navigation information, scanning the code information, etc.

[0060] In one embodiment, route status information may include at least one of the following: route traffic condition information, cost information, travel time information, speed limit information, queue length information, and other status information. Using at least one of these information as route status information can better reflect the impact of route status information on the target person's travel, improve the timeliness of travel route recommendations, and better reflect the travel route that is suitable for the target person.

[0061] In one embodiment, route status information includes route traffic condition information. The method for obtaining the route traffic condition information may include: obtaining the location information of traffic lights in each target travel route (wherein, for a subway route, the location information of each station in each subway route may be obtained); dividing each target travel route into nodes according to the location information to obtain a road node network for all target travel routes under each target travel mode; using image acquisition devices (such as cameras, etc.) at nodes in the road node network to collect image data in real time in the target travel route; performing feature extraction on the image data (such as feature extraction using a convolutional neural network) to extract image feature information, and obtaining route traffic condition information from the image feature information. In one embodiment, the route traffic condition information may include at least one of the following condition information: vehicle distance characteristics, road congestion characteristics, lane vehicle number characteristics, lane number characteristics, node average vehicle speed characteristics, collision occurrence characteristics, and the like.

[0062] For example, the device can obtain image data through an image acquisition device connected to a node in a road node network in the background, and use a convolutional neural network to extract image feature information from the image data to obtain route traffic condition information.

[0063] In subsequent steps, the integration of route traffic condition information with travel characteristics can more effectively reflect the impact of real-time road conditions on the target traveler's route selection. Furthermore, a road node network is constructed, with traffic lights or subway stations along each route serving as nodes for the target travel route. This effectively considers the impact of each road segment's node conditions on the target traveler's trip, improving the analysis accuracy and generalization of recommended travel routes. This can particularly enhance the accuracy and generalization of route analysis models used for route prediction.

[0064] In step S220 , the travel mode information and the travel route information are combined into combined travel information.

[0065] In one embodiment, the travel mode information includes operation information of the target subject on each target travel mode in the device, and the travel route information includes operation information of the target subject on each target travel route in the device. Step S220 combines the travel mode information and the travel route information into combined travel information, including:

[0066] Cartesian product operations are performed on the operation information corresponding to each target travel mode and the operation information corresponding to the target travel route under each target travel mode to obtain combined operation information under each target travel mode; and combined travel information is generated based on the combined operation information under each target travel mode.

[0067] For example, a target travel mode includes driving, which includes three target travel routes: A1, A2, and A3. A Cartesian product operation is performed on the operation information corresponding to the driving mode with the operation information corresponding to A1, A2, and A3 to obtain the combined operation information for the driving mode. Similarly, combined operation information for other target travel modes can be obtained.

[0068] After obtaining the combined operation information under each target travel mode, the set of combined operation information under all target travel modes can be used as the combined travel information, or the combined operation information under all target travel modes can be post-processed according to the rules and used as the combined travel information.

[0069] In one embodiment, step S220 combines the travel mode information and the travel route information into combined travel information, including: directly splicing the operation information corresponding to each target travel mode and the operation information corresponding to the target travel route under each target travel mode to obtain the combined operation information under each target travel mode; and using the combined operation information under all target travel modes as the combined travel information.

[0070] In step S230 , feature extraction processing is performed on the combined travel information to obtain target travel feature information.

[0071] In one embodiment, step S230 performs feature extraction processing on the combined travel information to obtain target travel feature information, including: performing convolution processing on the combined travel information based on a preset convolution matrix to obtain target travel feature information, wherein each element in the preset convolution matrix corresponds to two predetermined travel modes, and each element is the weight of the two predetermined travel modes corresponding to the element selected by the object in the sample object set within a predetermined time period.

[0072] In an example, there are five scheduled travel modes (indicated by serial numbers 1 to 5), and the preset convolution matrix is ​​as shown below.

[0073]

[0074] An element W in the preset convolution matrix 11 The corresponding two scheduled travel modes are the same as 1 and 1, and the element W 12 The corresponding two scheduled travel modes are different, 1 and 2, where W11 represents the sample object set that chooses W within the scheduled time period. 11 The corresponding weights of the two scheduled travel modes 1, W 12 Indicates that the objects in the sample object set select W within a predetermined time period 12 The corresponding weights of the two scheduled travel modes 1 and 2 are obtained by analogy, and other matrix elements are obtained accordingly.

[0075] The sample object set may include at least one object. The sample object set may be a pre-collected set of objects. The travel modes historically selected by the objects in the sample object set may be recorded. From the records, the predetermined travel modes selected by each object in its historical travels may be queried.

[0076] In this embodiment, the weights of the objects in the sample object set under each travel mode are used to construct a preset convolution matrix. The preset convolution matrix can effectively reflect the object's preference for travel mode. By performing convolution processing on the combined travel information through the preset convolution matrix, the target travel feature information extracted can better reflect the target object's preference for travel mode.

[0077] In another embodiment, step S230, performing feature extraction processing on the combined travel information to obtain target travel feature information, may include: performing convolution processing on the combined travel information based on a candidate convolution matrix to obtain target travel feature information, and each element in the candidate convolution matrix may be set based on experience.

[0078] In one embodiment, a preset convolution matrix is ​​generated according to the following method: determining the objects corresponding to each predetermined travel mode from a sample object set; determining the weights of the objects selected in each preset travel mode group within a predetermined time period based on the objects corresponding to each predetermined travel mode, wherein two predetermined travel modes form a preset travel mode group; and generating a matrix element for each weight corresponding to the preset travel mode group to obtain a preset convolution matrix.

[0079] The sample object set may include at least one object. The sample object set may be a pre-selected set of objects. The travel modes historically selected by the objects in the sample object set may be recorded with the individual consent and authorization of the objects. From the records, the scheduled travel modes selected by each object in historical travel may be queried, and further, the corresponding objects under each scheduled travel mode may be found.

[0080] In an example, the scheduled travel modes include 5 types (indicated by serial numbers 1 to 5), and the objects under the scheduled travel modes 1 to 5 can be determined respectively, and the weight W of the objects selected in each preset travel mode group within the scheduled time period can be further determined. 11 To W 55 Then, the weight corresponding to each preset travel mode group is generated into a matrix element, and a 5*5 dimensional symmetrical preset convolution matrix is ​​obtained.

[0081] In one embodiment, the weight of the objects that select each preset travel mode group within a predetermined time period is determined based on the objects corresponding to each predetermined travel mode, including: obtaining the number of objects that select each preset travel mode group within the predetermined time period based on the objects corresponding to each predetermined travel mode; calculating the ratio of the number of objects that select each preset travel mode group to the target number to obtain the weight corresponding to each preset travel mode group, where the target number is the total number of objects corresponding to all predetermined travel modes.

[0082] Based on the collected travel records, the number n of objects that choose each preset travel mode group within a predetermined time period (such as a week or a day, etc.) can be determined. ij , i is the i-th scheduled travel mode, j is the j-th scheduled travel mode, and the preset travel mode group (i, j) includes the scheduled travel modes i and j. The total number of objects corresponding to all scheduled travel modes, that is, the target number, can be N. Then, calculate n ij The ratio of N is the weight n corresponding to the preset travel mode group (i, j) ij / N.

[0083] In one example, there are five predefined travel modes (numbered 1 to 5), and the number of objects selected for each predefined travel mode group is n. ij , the total number of objects corresponding to all scheduled travel modes, that is, the target number is Then, calculate n ij and The ratio of the preset travel mode group (i, j) is obtained.

[0084] In step S240, the target travel feature information is fused with the route status information to obtain fused feature information.

[0085] In one embodiment, step S240, the target travel feature information and the route status information are fused to obtain fused feature information, including: performing a Cartesian product operation on the target travel feature information and the route status information to obtain pre-fused feature information; and generating fused feature information based on the pre-fused feature information.

[0086] In this embodiment, a Cartesian product operation is performed on the target travel feature information and the route status information to obtain pre-fused feature information. This effectively pre-fuses the target travel feature information and the route status information to obtain pre-fused feature information. The pre-fused feature information can effectively reflect the target subject's preference for travel mode and travel route. Generating fused feature information based on the pre-fused feature information can effectively reflect the target subject's preference for travel mode and travel route.

[0087] The fused feature information is generated based on the pre-fused feature information. The pre-fused feature information can be directly used as the fused feature information, or the pre-fused feature information can be processed by a full-connection operation and used as the fused feature information.

[0088] In another embodiment, step S240, fusing the target travel feature information with the route status information to obtain fused feature information, may include: directly multiplying the target travel feature information with a feature vector of the route status information object to obtain the fused feature information.

[0089] In one embodiment, generating fused feature information based on pre-fused feature information includes: performing a full-connection operation on the pre-fused feature information to obtain operation result feature information; and using the operation result feature information as the fused feature information.

[0090] In this way, the pre-fused feature information obtained by the Cartesian product operation is further processed by the full connection operation, and the feature information of the operation result is used as the fused feature information, which can further effectively reflect the target object's preference for travel mode and travel route.

[0091] In step S250 , prediction processing is performed based on the fused feature information to obtain predicted target travel information, where the target travel information is used to indicate a travel route under a travel mode recommended to the target object.

[0092] The predicted target travel information (which may be a target travel route classification label) may simultaneously indicate a recommended travel mode (such as driving) and a recommended travel route under the travel mode.

[0093] In one embodiment, step S250 performs prediction processing based on the fused feature information to obtain predicted target travel information, including: using a route analysis model to perform prediction processing based on the fused feature information to obtain a prediction result; and obtaining the predicted target travel information according to the prediction result.

[0094] The route analysis model is a pre-trained model based on machine learning or deep learning. The fused feature information is input into the route analysis model, which performs analysis and outputs prediction results. Based on the prediction results, the predicted target travel information can be obtained. The target travel information is used to indicate the travel route under the travel mode recommended to the target object.

[0095] In this way, a single route analysis model can simultaneously analyze the recommended travel mode and route for the target user, resulting in high recommendation efficiency. In contrast, related technologies use multiple models for multi-step analysis, first analyzing the recommended travel mode and then using a second model to analyze the recommended route based on the travel mode analysis results, resulting in low analysis efficiency.

[0096] In another embodiment, step S250, performing prediction processing based on the fused feature information to obtain predicted target travel information, may include: using a preset function to calculate the feature vector corresponding to the fused feature information to obtain a calculation result, and querying the travel information corresponding to the calculation result from the preset information table as the target travel information.

[0097] In one embodiment, the route analysis model is a multi-classification model, and the prediction result includes at least one score, each score corresponds to a type of travel information, and each travel information includes a travel mode and a travel route; obtaining predicted target travel information based on the prediction result includes: determining travel information whose score meets preset recommendation conditions based on the score corresponding to each travel information; and using the travel information that meets the preset recommendation conditions as target travel information.

[0098] The route analysis model is a multi-classification model, that is, the route analysis model can perform multi-classification analysis and obtain multiple analysis results. In one example, the route analysis model is a softmax multi-classification model.

[0099] The multi-classification route analysis model can perform predictions based on the fused feature information. The resulting predictions are multi-classification results, meaning they can include at least one (e.g., 16) score, which can be expressed as a probability. Each score corresponds to a type of trip information. Trip information with higher scores is more likely to be recommended. Each trip information includes a combination of travel mode and route.

[0100] When determining travel information whose scores meet the preset recommendation conditions based on the scores corresponding to each travel information, the travel information with the highest score can be determined as the travel information with the preset recommendation conditions, or the travel information with a score higher than a predetermined threshold can be determined as the travel information with the preset recommendation conditions.

[0101] In one embodiment, a route analysis model is trained according to the following steps: obtaining sample information and scheduled travel information corresponding to each object in a target object set, the sample information including travel mode information, travel route information and route status information, the scheduled travel information being used to indicate the scheduled travel mode and travel route; combining the travel mode information corresponding to each object with the travel route information to obtain the combined travel information corresponding to each object; performing feature extraction processing on the combined travel information corresponding to each object to obtain target travel feature information corresponding to each object; fusing the target travel feature information corresponding to each object with the route status information to obtain fused feature information corresponding to each object; using a target analysis model to perform prediction processing based on the fused feature information corresponding to each object to obtain predicted travel information corresponding to each object, the predicted travel information being used to indicate the predicted travel mode and travel route; adjusting the parameters in the target analysis model according to the predicted travel information and the scheduled travel information to obtain a trained route analysis model.

[0102] The target object set includes at least one object. Sample information corresponding to each object in the target object set and scheduled travel information corresponding to the sample information can be obtained from a predetermined location such as a database. The sample information includes travel mode information, travel route information and route status information. The scheduled travel information is used to indicate the scheduled travel mode and travel route.

[0103] Among them, within the predetermined time period, each object may correspond to at least one sample information, each sample information corresponds to a predetermined travel information, and each sample information may be the travel mode information, travel route information and route status information collected for a trip of the object.

[0104] The travel mode information corresponding to each object is combined with the travel route information to obtain the combined travel information corresponding to each object; feature extraction processing is performed on the combined travel information corresponding to each object to obtain the target travel feature information corresponding to each object; the target travel feature information corresponding to each object is fused with the route status information to obtain the fused feature information corresponding to each object; and a target analysis model is used to perform prediction processing based on the fused feature information corresponding to each object to obtain the predicted travel information corresponding to each object. The predicted travel information is used to indicate the predicted travel mode and travel route. Furthermore, the parameters in the target analysis model can be adjusted according to the predicted travel information and the scheduled travel information through gradient descent to obtain a trained route analysis model.

[0105] This approach only requires training a single route analysis model, which can simultaneously analyze both the recommended travel mode and route for the target user, resulting in high recommendation efficiency. Conventional technologies, on the other hand, employ multiple models for multi-step analysis, first analyzing the recommended travel mode and then, based on the travel mode analysis results, using a second model to analyze the recommended route, resulting in low analysis efficiency.

[0106] In one embodiment, the travel mode information corresponding to each object includes the object's operation information on each target travel mode in the device, and the travel route information includes the object's operation information on each target travel route in the device; combining the travel mode information corresponding to each object with the travel route information to obtain the combined travel information corresponding to each object may include: performing a Cartesian product operation on the operation information corresponding to each target travel mode in the travel mode information and travel route information of each object at the same travel time, and the operation information corresponding to the target travel route under each target travel mode, to obtain the combined operation information under each target travel mode; generating combined travel information based on the combined operation information under each target travel mode to obtain the combined travel information corresponding to each object.

[0107] Travel mode information and travel route information for the same travel time, such as travel mode information and travel route information collected for a specific trip of the subject. After obtaining the combined operation information for each target travel mode, the combined operation information for all target travel modes can be collected as the combined travel information, or the combined operation information for all target travel modes can be post-processed according to the rules to obtain the combined travel information.

[0108] In one embodiment, the travel mode information and travel route information corresponding to each object are combined to obtain the combined travel information corresponding to each object, which may include: directly splicing the operation information corresponding to each target travel mode and the operation information corresponding to the target travel route under each target travel mode in the travel mode information and travel route information of each object at the same travel time to obtain the combined operation information under each target travel mode; taking the combined operation information of each object under all target travel modes at the same travel time as the combined travel information to obtain the combined travel information corresponding to each object.

[0109] In one embodiment, feature extraction processing is performed on the combined travel information corresponding to each object to obtain target travel feature information corresponding to each object, including: for each combined travel information corresponding to each object, convolution processing is performed on the combined travel information based on a preset convolution matrix to obtain each corresponding target travel feature information, each element in the preset convolution matrix corresponds to two predetermined travel modes, and each element is the weight of the two predetermined travel modes corresponding to the element selected by the object in the sample object set within a predetermined time period.

[0110] A preset convolution matrix is ​​constructed using the weights of each travel mode for each subject in the sample set. This preset convolution matrix can effectively reflect the subject's preference for travel mode. By convolving the combined travel information with this preset convolution matrix, the extracted target travel feature information can better reflect the target subject's preference for travel mode, effectively improving the accuracy of the route analysis model's recommended route analysis. In one embodiment, the sample set is the aforementioned target set, which further improves the accuracy of route recommendations.

[0111] In one embodiment, a preset convolution matrix is ​​generated according to the following method: determining the objects corresponding to each predetermined travel mode from a sample object set; determining the weights of the objects selected in each preset travel mode group within a predetermined time period based on the objects corresponding to each predetermined travel mode, wherein two predetermined travel modes form a preset travel mode group; and generating a matrix element for each weight corresponding to the preset travel mode group to obtain a preset convolution matrix.

[0112] In one embodiment, the weight of the objects that select each preset travel mode group within a predetermined time period is determined based on the objects corresponding to each predetermined travel mode, including: obtaining the number of objects that select each preset travel mode group within the predetermined time period based on the objects corresponding to each predetermined travel mode; calculating the ratio of the number of objects that select each preset travel mode group to the target number to obtain the weight corresponding to each preset travel mode group, where the target number is the total number of objects corresponding to all predetermined travel modes.

[0113] Based on the collected travel records, the number n of objects that choose each preset travel mode group within a predetermined time period (such as a week or a day, etc.) can be determined. ij , i is the i-th scheduled travel mode, j is the j-th scheduled travel mode, and the preset travel mode group (i, j) includes the scheduled travel modes i and j. The total number of objects corresponding to all scheduled travel modes, that is, the target number, can be N. Then, calculate n ij The ratio of N is the weight n corresponding to the preset travel mode group (i, j) ij / N.

[0114] In one example, there are five predefined travel modes (numbered 1 to 5), and the number of objects selected for each predefined travel mode group is n. ij , the total number of objects corresponding to all scheduled travel modes, that is, the target number is Then, calculate n ij and The ratio of the preset travel mode group (i, j) is obtained.

[0115] In another embodiment, feature extraction processing is performed on the combined travel information corresponding to each object to obtain target travel feature information corresponding to each object, which may include: for each combined travel information corresponding to each object, convolution processing is performed on the combined travel information based on a candidate convolution matrix to obtain target travel feature information corresponding to each object, and each element in the candidate convolution matrix can be set based on experience.

[0116] In one embodiment, the target travel feature information corresponding to each object is fused with the route status information to obtain the fused feature information corresponding to each object, including: performing a Cartesian product operation on the target travel feature information corresponding to the same travel time of each object and the route status information to obtain pre-fused feature information; generating fused feature information based on the pre-fused feature information to obtain the fused feature information corresponding to each object.

[0117] The fused feature information is generated based on the pre-fused feature information. The pre-fused feature information can be directly used as the fused feature information, or the pre-fused feature information can be processed by a full-connection operation and used as the fused feature information.

[0118] In another embodiment, the target travel characteristic information corresponding to each object is fused with the route status information to obtain the fused characteristic information corresponding to each object, which may include: directly multiplying the target travel characteristic information corresponding to the same travel time of each object with the characteristic vector of the route status information object to obtain the fused characteristic information corresponding to each object.

[0119] In one embodiment, generating fused feature information based on pre-fused feature information includes: performing a full-connection operation on the pre-fused feature information to obtain operation result feature information; and using the operation result feature information as the fused feature information.

[0120] The above embodiment is further described below with reference to a process of recommending travel modes and travel routes to a target object in an application scenario.

[0121] Figure 3 The following is a flowchart of the route recommendation scheme used in related technologies in this scenario. Figure 4 and Figure 5 A flowchart is shown for recommending travel modes and travel routes by applying the aforementioned embodiments of the present application in this scenario.

[0122] First, see Figure 3 In the related art, a double-layer softmax multi-classification algorithm is used to recommend travel modes and travel routes. This related art uses a double-layer softmax multi-classification algorithm to analyze the recommended routes in two layers and multiple steps.

[0123] The first layer, in step S310, "input travel mode training test samples", which are samples from the previous period (T-1); in step S320, train the "optimal travel mode model W" based on the softmax multi-classification algorithm; in step S330, "input travel mode prediction samples", which are samples from the current period (T); in step S340, use the optimal travel mode model W to "predict the user's preferred travel mode i" based on the travel mode prediction samples.

[0124] Then, the process enters the second layer. In step S350, based on the prediction results of the optimal travel mode model W for the travel mode, "user route plan training and testing samples" are input. These samples are samples from the previous period (T-1). In step S360, the "optimal route plan model S" based on the softmax multi-classification algorithm is trained. In step S370, "travel route plan prediction samples" are input. These samples are samples of the user in the current period (T). In step S380, the optimal route plan model S is used to "predict the user's preferred travel route plan j" based on the travel route plan prediction samples. In step S390, the "optimal travel mode and route (i, j)" predicted by the two models are recommended to the user.

[0125] In this related technical solution, after simple processing of the relevant information of the object, a multi-step analysis is performed to determine the recommended travel mode and route. Problems with this solution include at least the following: on the one hand, due to the unreliability of the relevant information used and the simplistic information processing, the analysis accuracy of the travel mode and route is low, resulting in poor route recommendation results; on the other hand, the user's preferred travel mode i is first trained and predicted, and then the user's preferred travel route plan j is trained and predicted under this travel mode, thereby obtaining the user's preferred travel mode and route (i, j). However, the training and prediction method of the two-layer multi-classification algorithm has a high complexity in the training and prediction model, which reduces the model efficiency.

[0126] Further Reading Figure 4 and Figure 5 In this scenario, the process of recommending travel modes and travel routes by using the aforementioned embodiments of this application may include: (1) training a route analysis model, and (2) predicting travel routes based on the route analysis model. The route analysis model in this scenario adopts a multi-classification model based on the softmax multi-classification algorithm. In this scenario, the process of recommending travel modes and travel routes by using the aforementioned embodiments of this application may avoid Figure 3 The problems existing in the related technology are shown.

[0127] See Figure 4As shown in FIG, (1) training the route analysis model includes: a feature processing stage and a model training stage. (2) applying the route analysis model to recommend travel routes to the target object includes: a feature processing stage and a model prediction stage.

[0128] Combine Figure 5 As shown, when (1) training the route analysis model, the feature processing stage can execute steps S410 to S4450.

[0129] Step S410: Obtain sample information and scheduled travel information corresponding to each object in the target object set. The sample information includes travel mode information, travel route information, and route status information. The scheduled travel information is used to indicate the scheduled travel mode and travel route.

[0130] The target object set includes at least one object. Sample information corresponding to each object in the target object set and the scheduled travel information corresponding to the sample information can be legally obtained from a predetermined location such as a database. The sample information includes travel mode information, travel route information, and route status information. The scheduled travel information is used to indicate the scheduled travel mode and travel route.

[0131] Within a predetermined time period, each subject can be associated with at least one sample information item, each of which corresponds to a scheduled trip information item. Each sample information item can include travel mode information, travel route information, and route status information collected for a single trip. The scheduled trip information item can be a travel route classification label indicating the travel mode and route selected by the subject for that trip. In this scenario, there are five target travel modes: taxi, driving, bus / subway, walking, and cycling. Each target travel mode in this scenario corresponds to three target travel routes.

[0132] The travel mode information includes the object's operation information for each target travel mode in the device, that is, the travel mode information may include at least one of the following operation information: the click behavior of the object under the travel mode plan in the device (such as the corresponding page of the taxi plan in the map application), the input starting point, the input end point, the user's travel history information, the user's active behavior on the APP / mini program side, the module payment behavior, the module payment type, etc.

[0133] Travel route information includes the subject's operations on each target travel route in the device. Specifically, travel route information may include at least one of the following: the subject's clicks on the travel route plan in the device, clicks on route exploration, exits route exploration, clicks on real-time navigation information, exits real-time navigation information, and code scanning information. Route status information may include at least one of the following: route traffic conditions, cost information, travel time information, speed limit information, queue length information, and other status information.

[0134] The route status information includes route traffic condition information, and the method for obtaining the route traffic condition information may include: obtaining the location information of the traffic lights in each target travel route (wherein, for a subway route, the location information of each station in each subway route can be obtained); dividing each target travel route into nodes according to the location information, and obtaining a road node network of all target travel routes under each target travel mode; using an image acquisition device (such as a camera, etc.) of a node in the road node network to collect image data in real time in the target travel route; performing feature extraction on the image data (in this scenario, such as Figure 4 The two convolutional layers used for feature extraction (shown in the figure) can extract image feature information, from which route traffic condition information can be obtained. Route traffic condition information can include at least one of the following: vehicle spacing characteristics, road congestion characteristics, number of vehicles in a lane, number of lanes, average vehicle speed at a node, and collision occurrence characteristics.

[0135] For 5 travel modes, each of which has 3 travel route plans, there are a total of 5*3=15 travel plans. When the subject clicks on the jth (j=1, 2, ..., 15) travel plan and enters the navigation, the travel route classification label can be recorded as j, and 0 represents other travel plans except 1 to 15. Specifically, 0 represents other travel plans except 1 to 15, 1 represents travel route plan 1 for driving, 2 represents travel route plan 2 for driving, 3 represents travel route plan 3 for driving, 4 represents travel route plan 1 for taxi, 5 represents travel route plan 2 for taxi, 6 represents travel route plan 3 for taxi, 7 represents travel route plan 1 for walking, 8 represents travel route plan 2 for walking, 9 represents travel route plan 3 for walking, 10 represents travel route plan 1 for cycling, 11 represents travel route plan 2 for cycling, 12 represents travel route plan 3 for cycling, 13 represents travel route plan 1 for public transportation, 14 represents travel route plan 2 for public transportation, and 15 represents travel route plan 3 for public transportation.

[0136] Furthermore, 16 types of travel route classification labels (0-15) and corresponding route classification label mapping tables can be output (the mapping table records the correspondence between each travel route classification label and the occurrence plan, which includes the occurrence method and travel route). The scheduled travel information corresponding to each sample information can be a travel route classification label indicating the travel method and travel route selected by the subject for the trip.

[0137] Step S420 : Combining the travel mode information and travel route information corresponding to each object to obtain combined travel information corresponding to each object.

[0138] The travel mode information and travel route information corresponding to each object are combined to obtain the combined travel information corresponding to each object, including: performing a Cartesian product operation on the travel mode information and travel route information of each object at the same travel time, the operation information corresponding to each target travel mode, and the operation information corresponding to the target travel route under each target travel mode to obtain the combined operation information under each target travel mode; generating the combined travel information based on the combined operation information under each target travel mode to obtain the combined travel information corresponding to each object.

[0139] Travel mode information and travel route information for the same travel time, such as travel mode information and travel route information collected for a specific trip of the subject. After obtaining the combined operation information for each target travel mode, the combined operation information for all target travel modes can be collected as the combined travel information, or the combined operation information for all target travel modes can be post-processed according to the rules to obtain the combined travel information.

[0140] Step S430 : performing feature extraction processing on the combined travel information corresponding to each object to obtain target travel feature information corresponding to each object.

[0141] Perform feature extraction processing on the combined travel information corresponding to each object to obtain target travel feature information corresponding to each object, including: for each combined travel information corresponding to each object, perform convolution processing on the combined travel information based on a preset convolution matrix to obtain each corresponding target travel feature information, each element in the preset convolution matrix corresponds to two predetermined travel modes, and each element is the weight of the two predetermined travel modes corresponding to the element selected by the object in the sample object set within a predetermined time period.

[0142] In this scenario, based on the preset convolution matrix, the following Figure 4 and 5 The two convolutional layers shown perform convolution processing on the combined travel information to obtain the characteristic information of each corresponding target trip.

[0143] The preset convolution matrix is ​​generated as follows: determining the objects corresponding to each predetermined travel mode from a sample object set; determining the weights of the objects that select each predetermined travel mode group within a predetermined time period based on the objects corresponding to each predetermined travel mode, where two predetermined travel modes form a predetermined travel mode group; and generating a matrix element from the weights corresponding to each predetermined travel mode group to obtain the preset convolution matrix. The sample object set may be the target object set in the aforementioned step.

[0144] In this scenario, there are five scheduled travel modes (numbered 1 to 5, representing taxi, driving, bus and subway, walking, and cycling respectively). The objects under scheduled travel modes 1 to 5 can be determined respectively, and the weight W of the objects who choose each preset travel mode group within the scheduled time period can be further determined. 11 To W 55 Then, the weight corresponding to each preset travel mode group is generated into a matrix element, and the 5*5 symmetrical preset convolution matrix shown below is obtained.

[0145]

[0146] Among them, an element W in the preset convolution matrix 11 The corresponding two scheduled travel modes are the same as 1 and 1, and the element W 12 The corresponding two scheduled travel modes are different, 1 and 2, where W11 represents the sample object set that chooses W within the scheduled time period. 11 The corresponding weights of the two scheduled travel modes 1, W 12 Indicates that the objects in the sample object set select W within a predetermined time period 12 The corresponding weights of the two scheduled travel modes 1 and 2 are obtained by analogy, and other matrix elements are obtained accordingly.

[0147] The weight of the objects that select each preset travel mode group within the predetermined time period is determined based on the objects corresponding to each predetermined travel mode, including: obtaining the number of objects that select each preset travel mode group within the predetermined time period based on the objects corresponding to each predetermined travel mode; calculating the ratio of the number of objects that select each preset travel mode group to a target number to obtain the weight corresponding to each preset travel mode group, where the target number is the total number of objects corresponding to all predetermined travel modes.

[0148] The number n of objects that select each preset travel mode group within a predetermined time period (such as a week or a day, etc.) can be determined. ij , i is the i-th scheduled travel mode, j is the j-th scheduled travel mode, and the preset travel mode group (i, j) includes the scheduled travel modes i and j. The total number of objects corresponding to all scheduled travel modes, that is, the target number, can be N. Then, calculate n ij The ratio of N is the weight n corresponding to the preset travel mode group (i, j) ij / N.

[0149] In this scenario, there are five pre-set travel modes (numbered 1 to 5, respectively, representing taxi, driving, bus and subway, walking, and cycling). The number of objects selected in each pre-set travel mode group is n. ij , the total number of objects corresponding to all scheduled travel modes, that is, the target number is Then, calculate n ij and The ratio of the preset travel mode group (i, j) is obtained.

[0150] Step S440 : fusing the target travel feature information corresponding to each object with the route status information to obtain fused feature information corresponding to each object.

[0151] The target travel feature information corresponding to each object is fused with the route status information to obtain fused feature information corresponding to each object, including: performing a Cartesian product operation on the target travel feature information corresponding to the same travel time of each object and the route status information to obtain pre-fused feature information; generating fused feature information based on the pre-fused feature information to obtain fused feature information corresponding to each object.

[0152] Generating fused feature information based on pre-fused feature information includes: performing full-connection operation processing on the pre-fused feature information to obtain operation result feature information; and using the operation result feature information as fused feature information.

[0153] In this scenario, participation Figure 4 and 5 As shown, at the collection layer, the target travel feature information corresponding to the same travel time of each object can be processed with the route status information by a Cartesian product operation to obtain pre-fusion feature information; at the fully connected layer, a fully connected operation can be performed to obtain the operation result feature information.

[0154] Each object can correspond to at least one fused feature information. For example, if an object travels three times within a predetermined time period, the sample information for each trip is processed to generate a fused feature information, and thus the object corresponds to three fused feature information. Each sample information corresponds to a predetermined trip information, and furthermore, each fused feature information corresponds to a predetermined trip information (i.e., a travel route classification label, such as 1 or 3, etc.).

[0155] Further, combined Figure 5 As shown, when (1) training the route analysis model, step S450 can be executed in the model training phase.

[0156] Step S450 uses the target analysis model to perform prediction processing based on the fused feature information corresponding to each object, obtaining predicted travel information corresponding to each object. The predicted travel information indicates the predicted travel mode and travel route. The parameters in the target analysis model are adjusted based on the predicted travel information and the scheduled travel information to obtain a trained route analysis model. The parameters in the target analysis model can be adjusted based on the predicted travel information and the scheduled travel information using gradient descent to obtain a trained route analysis model that meets the requirements.

[0157] All the fusion feature information generated in step S440 (generated based on the sample information of the historical period (T-1 period) collected in advance) can be divided according to the ratio (such as training sample: test sample = 8:2) as the training sample of the current period (T period) and test samples

[0158] The training samples The fusion feature information X in T-1 Substitute the target analysis model (softmax multi-classification model): The predicted travel information P corresponding to each fused feature information corresponding to each object can be obtained.

[0159] The gradient descent method can be used to adjust the parameter W in the target analysis model according to the predicted travel information P and scheduled travel information corresponding to each fusion feature information. i Update and adjust to train the route analysis model. In the model testing phase, input test samples into the route analysis model. The fusion feature information X in T-1 , calculate the score of each category of the object in the 16 travel route classification labels (0 to 15), and calculate evaluation indicators such as recall rate, precision rate, AUC under the test sample. When the route analysis model meets the standards, the route analysis model can be saved.

[0160] Further, combined Figure 5 As shown, when (2) the route analysis model is applied to recommend a travel route to the target object, the feature processing stage can execute steps S410 to S440.

[0161] Step S410: Acquire travel mode information, travel route information, and route status information corresponding to the target object.

[0162] Travel mode information may include at least one of the following: clicks on travel mode options, starting point and destination input, user travel history, user activity on the app / mini-program, module payment behavior, and module payment type. Travel route information may include at least one of the following: clicks on travel route options, clicks on route exploration, exiting route exploration, clicks on real-time navigation information, exiting real-time navigation information, and code scanning information. Route status information may include at least one of the following: route traffic conditions, cost information, travel time information, speed limit information, and queue length information.

[0163] Route status information includes route traffic condition information. Methods for obtaining this route traffic condition information may include: obtaining the location information of traffic lights in each target travel route (wherein, for subway routes, the location information of each station in each subway route can be obtained); dividing each target travel route into nodes according to the location information to obtain a road node network for all target travel routes under each target travel mode; using image acquisition devices (such as cameras, etc.) at nodes in the road node network to collect image data in real time in the target travel route; performing feature extraction on the image data (such as through a convolutional neural network) to extract image feature information, and obtaining route traffic condition information from the image feature information. Route traffic condition information may include at least one of the following: vehicle spacing characteristics, road congestion characteristics, lane vehicle number characteristics, lane number characteristics, node average vehicle speed characteristics, collision occurrence characteristics, and other condition information.

[0164] Step S420: combining the travel mode information and the travel route information into combined travel information.

[0165] The travel mode information includes the operation information of the target object on each target travel mode in the device, and the travel route information includes the operation information of the target object on each target travel route in the device; the travel mode information and the travel route information are combined into combined travel information, including: performing Cartesian product operation on the operation information corresponding to each target travel mode and the operation information corresponding to the target travel route under each target travel mode to obtain the combined operation information under each target travel mode; and generating the combined travel information based on the combined operation information under each target travel mode.

[0166] Step S430: performing feature extraction processing on the combined travel information to obtain target travel feature information.

[0167] Performing feature extraction processing on the combined travel information to obtain target travel feature information, including: performing convolution processing on the combined travel information based on a preset convolution matrix to obtain target travel feature information, wherein each element in the preset convolution matrix corresponds to two predetermined travel modes, and each element is a weight of the two predetermined travel modes corresponding to the element selected by an object in a sample object set within a predetermined time period.

[0168] Step S440: fusing the target travel feature information with the route status information to obtain fused feature information.

[0169] The target travel feature information and the route status information are fused to obtain fused feature information, including: performing a Cartesian product operation on the target travel feature information and the route status information to obtain pre-fused feature information; and generating fused feature information based on the pre-fused feature information.

[0170] Generating fused feature information based on pre-fused feature information includes: performing full-connection operation processing on the pre-fused feature information to obtain operation result feature information; and using the operation result feature information as fused feature information.

[0171] Further, combined Figure 5 As shown, when (2) the route analysis model is applied to recommend a travel route to the target object, step S460 can be executed in the model prediction stage.

[0172] Step S460 , performing prediction processing based on the fused feature information to obtain predicted target travel information, where the target travel information is used to indicate a travel route under a travel mode recommended to the target object.

[0173] Prediction processing is performed based on the fused feature information to obtain predicted target travel information, including: using a route analysis model, performing prediction processing based on the fused feature information to obtain a prediction result; and obtaining the predicted target travel information according to the prediction result.

[0174] Among them, the route analysis model is a softmax multi-classification model, which uses the fusion feature information corresponding to the target object as Figure 4 The predicted sample X shown T , substitute into the softmax multi-classification model: The model can calculate and output the score of each of the 16 travel route classification labels (0 to 15) of the target object. Furthermore, the prediction result includes at least one score, each score corresponds to a type of travel information (i.e., travel route classification label), and each travel information includes travel mode and travel route.

[0175] Obtaining predicted target travel information according to the prediction result, including: determining travel information whose score meets preset recommendation conditions according to the score corresponding to each travel information; and using the travel information that meets the preset recommendation conditions as target travel information.

[0176] Specifically, according to the score corresponding to each travel information, determining the travel information whose score meets the preset recommendation conditions may include: determining the travel information whose score is higher than a predetermined threshold as the travel information of the preset recommendation conditions (i.e., the target travel information). For example, given a predetermined threshold of 0.5, when P i If the value is ≥0.5, it is classified as Class A, otherwise it is classified as Class B. Based on the travel information that meets the preset recommendation conditions (i.e., target travel information), a travel route classification data list corresponding to the target object can be output.

[0177] The travel route classification data list and the route classification label mapping table are intersected and associated according to the travel route classification labels in the travel route classification data list to obtain the travel route classification labels associated from the route classification label mapping table. According to the travel plan associated with the travel route classification label (the occurrence plan includes the occurrence method and the travel route), the travel route under the travel mode recommended to the target object by the target travel information can be determined, that is, the travel mode preferred by the target object and the corresponding travel route under the travel mode pointed to by the target travel information.

[0178] It should be noted that if Figure 6 In the system framework diagram shown, the convolutional layer, the aggregation layer, the fully connected layer and the route classification model can be combined into a recommendation model, or the convolutional layer, the aggregation layer, the fully connected layer and the route classification model can be independent structures.

[0179] In this way, by applying the embodiments of the present application to perform route recommendations in this scenario, a recommendation scheme for the optimal travel mode and route based on the state fusion multi-classification method is implemented, which can have at least the following beneficial effects:

[0180] By combining the travel mode information, travel route information and route status information corresponding to the target object, performing feature extraction processing and fusing processing to generate fused feature information, the target travel information can be accurately predicted based on the fused feature information. According to the target travel information, the travel mode and travel route under the travel mode recommended to the target object can be accurately indicated, thereby effectively improving the accuracy of route recommendation and improving the route recommendation effect.

[0181] Only one route analysis model needs to be trained, and it can simultaneously analyze the recommended travel mode and travel route for the target person, achieving high recommendation efficiency. In contrast, related technologies use multiple models for multi-step analysis, first analyzing the recommended travel mode and then using a second model to analyze the recommended travel route based on the travel mode analysis results, resulting in low analysis efficiency.

[0182] By effectively considering the impact of each road segment's node conditions on the target person's travel, the accuracy and generalization of recommended travel routes can be improved. This is particularly true when using route analysis models for route prediction, as it can improve the accuracy and generalization of route analysis models.

[0183] The preset convolution matrix is ​​constructed using the weights of the objects in the sample object set under each travel mode. The preset convolution matrix can effectively reflect the object's preference for travel mode. By performing convolution processing on the combined travel information through the preset convolution matrix, the target travel feature information extracted can better reflect the target object's preference for travel mode, and can effectively improve the accuracy of the recommended route analysis of the route analysis model.

[0184] To facilitate better implementation of the route recommendation method provided in the embodiment of the present application, the embodiment of the present application also provides a route recommendation device based on the above route recommendation method. The meanings of the terms herein are the same as those in the above route recommendation method, and the specific implementation details can be referred to the description in the method embodiment. Figure 6 A block diagram of a route recommendation device according to an embodiment of the present application is shown.

[0185] like Figure 6 As shown, the route recommendation device 500 may include an acquisition module 510 , a combination module 520 , an extraction module 530 , a fusion module 540 and an analysis module 550 .

[0186] The acquisition module 510 can be used to obtain the travel mode information, travel route information and route status information corresponding to the target object; the combination module 520 can be used to combine the travel mode information and the travel route information into combined travel information; the extraction module 530 can be used to perform feature extraction processing on the combined travel information to obtain target travel feature information; the fusion module 540 can be used to fuse the target travel feature information with the route status information to obtain fused feature information; the analysis module 550 can be used to perform prediction processing based on the fused feature information to obtain predicted target travel information, and the target travel information is used to indicate the travel route under the travel mode recommended to the target object.

[0187] In some embodiments of the present application, in some embodiments of the present application, the extraction module is used to: perform convolution processing on the combined travel information based on the preset convolution matrix to obtain the target travel characteristic information, each element in the preset convolution matrix corresponds to two predetermined travel modes, and each element is the weight of the object in the sample object set selecting the two predetermined travel modes corresponding to the element within a predetermined time period.

[0188] In some embodiments of the present application, the device also includes a generation module for a preset convolution matrix, including: an object determination unit, used to determine the object corresponding to each of the predetermined travel modes from the sample object set; a weight determination unit, used to determine the weight of the objects of each preset travel mode group selected within a predetermined time period according to the objects corresponding to each of the predetermined travel modes, wherein two of the predetermined travel modes form a preset travel mode group; a matrix generation unit, used to generate a matrix element for the weight corresponding to each of the preset travel mode groups to obtain the preset convolution matrix.

[0189] In some embodiments of the present application, the weight determination unit is used to: obtain the number of objects that select each preset travel mode group within a predetermined time period based on the objects corresponding to each of the predetermined travel modes; calculate the ratio of the number of objects that select each preset travel mode group to the target number to obtain the weight corresponding to each of the preset travel mode groups, and the target number is the total number of objects corresponding to all the predetermined travel modes.

[0190] In some embodiments of the present application, the fusion module includes: a pre-fusion unit, used to perform a Cartesian product operation on the target travel feature information and the route status information to obtain pre-fusion feature information; and an information generation unit, used to generate the fusion feature information based on the pre-fusion feature information.

[0191] In some embodiments of the present application, the information generating unit is configured to: perform a fully connected operation on the pre-fused feature information to obtain operation result feature information; and use the operation result feature information as the fused feature information.

[0192] In some embodiments of the present application, the travel mode information includes the operation information of the target object on each target travel mode in the device, and the travel route information includes the operation information of the target object on each target travel route in the device; the combination module is used to: perform Cartesian product operation on the operation information corresponding to each target travel mode and the operation information corresponding to the target travel route under each target travel mode to obtain the combined operation information under each target travel mode; and generate the combined travel information based on the combined operation information under each target travel mode.

[0193] In some embodiments of the present application, the analysis module includes: a model analysis unit, which is used to use a route analysis model to perform prediction processing based on the fused feature information to obtain a prediction result; and a result analysis unit, which is used to obtain predicted target travel information based on the prediction result.

[0194] In some embodiments of the present application, the route analysis model is a multi-classification model, and the prediction result includes at least one score, each score corresponds to a type of travel information, and each travel information includes a travel mode and a travel route; the result analysis unit is used to: determine the travel information whose score meets the preset recommendation conditions based on the score corresponding to each travel information; and use the travel information that meets the preset recommendation conditions as the target travel information.

[0195] In some embodiments of the present application, the device also includes a training module for a route analysis model, which is used to: obtain sample information and scheduled travel information corresponding to each object in the target object set, the sample information including travel mode information, travel route information and route status information, and the scheduled travel information is used to indicate the scheduled travel mode and travel route; combine the travel mode information corresponding to each object with the travel route information to obtain the combined travel information corresponding to each object; perform feature extraction processing on the combined travel information corresponding to each object to obtain the target travel feature information corresponding to each object; fuse the target travel feature information corresponding to each object with the route status information to obtain the fused feature information corresponding to each object; use the target analysis model to perform prediction processing based on the fused feature information corresponding to each object to obtain the predicted travel information corresponding to each object, and the predicted travel information is used to indicate the predicted travel mode and travel route; adjust the parameters in the target analysis model according to the predicted travel information and the scheduled travel information to obtain the trained route analysis model.

[0196] In some embodiments of the present application, the route status information includes status information corresponding to each target travel route, and the status information includes at least one of route traffic condition information, cost information, travel time information, speed limit information, and queue length information.

[0197] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiment of the application, the features and functions of two or more modules or units described above can be concretized in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units to be concretized.

[0198] In addition, an embodiment of the present application further provides an electronic device, which may be a terminal or a server, such as Figure 7 , which shows a schematic diagram of the structure of the electronic device involved in the embodiment of the present application, specifically:

[0199] The electronic device may include one or more processing core processors 601, one or more computer-readable storage media memories 602, a power supply 603, an input unit 604 and other components. Those skilled in the art will understand that Figure 7 The electronic device structure shown in the figure does not constitute a limitation of the electronic device, and may include more or fewer components than shown in the figure, or combine certain components, or arrange components differently.

[0200] Processor 601 is the control center of the electronic device, connecting the various components of the entire computer device using various interfaces and lines. By running or executing software programs and / or modules stored in memory 602 and accessing data stored in memory 602, it performs various functions of the computer device and processes data, thereby detecting the electronic device. Optionally, processor 601 may include one or more processing cores; preferably, processor 601 may integrate an application processor and a modem processor, wherein the application processor primarily processes the operating system, user interfaces, and application programs, while the modem processor primarily handles wireless communications. It is understood that the modem processor may not be integrated into processor 601.

[0201] The memory 602 can be used to store software programs and modules. The processor 601 executes various functional applications and data processing by running the software programs and modules stored in the memory 602. The memory 602 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 602 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage device. Accordingly, the memory 602 may also include a memory controller to provide the processor 601 with access to the memory 602.

[0202] The electronic device also includes a power supply 603 for supplying power to various components. Preferably, the power supply 603 can be logically connected to the processor 601 via a power management system, thereby enabling the power management system to manage charging, discharging, and power consumption. The power supply 603 can also include one or more DC or AC power supplies, a recharging system, a power failure detection circuit, a power converter or inverter, a power status indicator, and other arbitrary components.

[0203] The electronic device may further include an input unit 604, which may be used to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal input related to user settings and function control.

[0204] Although not shown, the electronic device may further include a display unit, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 601 in the electronic device will load the executable files corresponding to the processes of one or more computer programs into the memory 602 according to the following instructions, and the processor 601 will run the computer programs stored in the memory 602, thereby realizing the various functions of the aforementioned embodiments of the present application. For example, the processor 601 may execute:

[0205] Acquire travel mode information, travel route information, and route status information corresponding to a target object; combine the travel mode information and travel route information into combined travel information; perform feature extraction processing on the combined travel information to obtain target travel feature information; fuse the target travel feature information with the route status information to obtain fused feature information; perform prediction processing based on the fused feature information to obtain predicted target travel information, wherein the target travel information is used to indicate a travel route under the travel mode recommended to the target object.

[0206] Those skilled in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be accomplished by a computer program, or by controlling related hardware through a computer program. The computer program may be stored in a computer-readable storage medium and loaded and executed by a processor.

[0207] To this end, an embodiment of the present application further provides a computer-readable storage medium, in which a computer program is stored. The computer program can be loaded by a processor to execute the steps of any method provided in the embodiment of the present application.

[0208] The computer-readable storage medium may include a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0209] Since the computer program stored in the computer-readable storage medium can execute the steps of any method provided in the embodiments of the present application, the beneficial effects that can be achieved by the method provided in the embodiments of the present application can be achieved. Please refer to the previous embodiments for details and will not be repeated here.

[0210] According to one aspect of the present application, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in the various optional implementations of the above-described embodiments of the present application.

[0211] Those skilled in the art will readily conceive of other embodiments of the present application after considering the specification and practicing the embodiments disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present application that follow the general principles of this application and include common knowledge or customary techniques in the art that are not disclosed herein.

[0212] It should be understood that the present application is not limited to the embodiments that have been described above and shown in the accompanying drawings, but various modifications and changes may be made without departing from the scope thereof.

Claims

1. A route recommendation method, characterized in that: include: Obtain the travel mode information, travel route information, and route status information corresponding to the target object; combining the travel mode information and the travel route information into combined travel information; Performing feature extraction processing on the combined travel information based on a preset convolution matrix to obtain target travel feature information; wherein the preset convolution matrix is ​​generated according to the following method: determining objects corresponding to each predetermined travel mode from a sample object set; determining weights for objects that select each preset travel mode group within a predetermined time period based on the objects corresponding to each predetermined travel mode, wherein two predetermined travel modes form a preset travel mode group; generating a matrix element from the weights corresponding to each preset travel mode group to obtain the preset convolution matrix; Fusing the target travel feature information with the route status information to obtain fused feature information; Prediction processing is performed based on the fused feature information to obtain predicted target travel information, where the target travel information is used to indicate a travel route under a travel mode recommended to the target object.

2. The route recommendation method according to claim 1, characterized in that: The performing feature extraction processing on the combined travel information based on a preset convolution matrix to obtain target travel feature information includes: The combined travel information is convolved based on the preset convolution matrix to obtain the target travel characteristic information, wherein each element in the preset convolution matrix corresponds to two predetermined travel modes, and each element is the weight of the object in the sample object set selecting the two predetermined travel modes corresponding to the element within a predetermined time period.

3. The route recommendation method according to claim 1, wherein: The step of determining the weight of the objects that select each preset travel mode group within a predetermined time period according to the objects corresponding to each predetermined travel mode includes: According to the objects corresponding to each of the predetermined travel modes, the number of objects that select each preset travel mode group within a predetermined time period is obtained; The ratio of the number of objects that select each preset travel mode group to the target number is calculated to obtain the weight corresponding to each preset travel mode group, where the target number is the total number of objects corresponding to all the preset travel modes.

4. The route recommendation method according to any one of claims 1 to 3, characterized in that: The fusing the target travel characteristic information with the route status information to obtain fused characteristic information includes: Performing a Cartesian product operation on the target travel feature information and the route state information to obtain pre-fusion feature information; The fused feature information is generated based on the pre-fused feature information.

5. The route recommendation method according to claim 4, characterized in that: The generating the fused feature information based on the pre-fused feature information includes: Performing a full-connection operation on the pre-fused feature information to obtain operation result feature information; The operation result feature information is used as the fusion feature information.

6. The route recommendation method according to any one of claims 1 to 3, characterized in that: The travel mode information includes the operation information of the target object on each target travel mode in the device, and the travel route information includes the operation information of the target object on each target travel route in the device; The combining of the travel mode information and the travel route information into combined travel information includes: Performing a Cartesian product operation on the operation information corresponding to each target travel mode and the operation information corresponding to the target travel route under each target travel mode to obtain combined operation information under each target travel mode; The combined travel information is generated based on the combined operation information under each of the target travel modes.

7. The route recommendation method according to any one of claims 1 to 3, characterized in that: The performing prediction processing based on the fused feature information to obtain predicted target travel information includes: Using a route analysis model, performing prediction processing based on the fused feature information to obtain a prediction result; The predicted target travel information is obtained according to the prediction result.

8. The route recommendation method according to claim 7, characterized in that: The route analysis model is a multi-classification model, the prediction result includes at least one score, each score corresponds to a type of travel information, and each type of travel information includes a travel mode and a travel route; The obtaining predicted target travel information according to the prediction result includes: Determining, based on the scores corresponding to each piece of travel information, travel information whose scores meet preset recommendation conditions; The travel information that meets the preset recommendation conditions is used as the target travel information.

9. The route recommendation method according to claim 7, characterized in that: The route analysis model is trained according to the following steps: Obtaining sample information and scheduled travel information corresponding to each object in the target object set, wherein the sample information includes travel mode information, travel route information, and route status information, and the scheduled travel information is used to indicate the scheduled travel mode and travel route; Combine the travel mode information and travel route information corresponding to each object to obtain combined travel information corresponding to each object; Perform feature extraction processing on the combined travel information corresponding to each object to obtain target travel feature information corresponding to each object; The target travel feature information corresponding to each object is fused with the route status information to obtain the fused feature information corresponding to each object; Using a target analysis model to perform prediction processing based on the fused feature information corresponding to each of the objects, to obtain predicted travel information corresponding to each object, wherein the predicted travel information is used to indicate a predicted travel mode and travel route; According to the predicted travel information and the scheduled travel information, the parameters in the target analysis model are adjusted to obtain the trained route analysis model.

10. The route recommendation method according to any one of claims 1 to 3, characterized in that: The route status information includes status information corresponding to each target travel route, and the status information includes at least one of route traffic condition information, cost information, travel time information, speed limit information, and queue length information.

11. A route recommendation device, characterized in that: include: An acquisition module is used to obtain the travel mode information, travel route information and route status information corresponding to the target object; a combining module, configured to combine the travel mode information and the travel route information into combined travel information; an extraction module configured to perform feature extraction processing on the combined travel information based on a preset convolution matrix to obtain target travel feature information; wherein the preset convolution matrix is ​​generated by: determining an object corresponding to each predetermined travel mode from a sample object set; determining a weight for selecting an object in each preset travel mode group within a predetermined time period based on the object corresponding to each predetermined travel mode, wherein two predetermined travel modes form a preset travel mode group; and generating a matrix element from each weight corresponding to each preset travel mode group to obtain the preset convolution matrix; a fusion module, configured to fuse the target travel characteristic information with the route status information to obtain fused characteristic information; An analysis module is used to perform prediction processing based on the fused feature information to obtain predicted target travel information, where the target travel information is used to indicate a travel route under a travel mode recommended to the target object.

12. A computer-readable storage medium, characterized in that A computer program is stored thereon, and when the computer program is executed by a processor of a computer, the computer is caused to execute the method according to any one of claims 1 to 10.

13. An electronic device, characterized in that: include: a memory storing a computer program; A processor reads a computer program stored in a memory to execute the method according to any one of claims 1 to 10.

14. A computer program product, characterized in that The computer program product comprises a computer program, which implements the method of any one of claims 1 to 10 when executed by a processor.

Images