Route sorting model training method, route recommendation method, device and equipment

Abstract

The disclosed embodiments relate to a training method for a route ranking model, a route recommendation method, an apparatus, and a device. The method comprises inputting statistical features of a first recommended navigation route historically selected by a navigated object and bias features of multiple historical navigation routes into a first network to obtain a scoring result reflecting the influence of the bias features of the navigation route on the navigated object. The scoring result is added to the training process of a second network to help the second network learn the influence of the bias features of the navigation route on the navigated object. The preference features of the navigated object for each preset navigation strategy, the statistical features of the first recommended navigation route historically selected by the navigated object, and the route features of multiple historical navigation routes are used as training data for the second network to help the second network learn the true preferences of the navigated object. This ensures that the scoring results output by the second network match the true preferences of the navigated object, thereby improving the user experience.

Description

Technical Field

[0001] The embodiments of the present disclosure relate to the field of navigation technology, and more particularly to a training method for a route sorting model, a route recommendation method, an apparatus, and a device. Background Art

[0002] Related technologies provide map navigation applications that can recommend corresponding navigation routes based on the starting and ending locations entered by the user. However, the inventors of this application have discovered that if the recommended routes do not meet the user's individual preferences, this can negatively impact the user experience. Therefore, how to make the recommended routes more consistent with the user's individual preferences is a technical problem that those skilled in the art need to address. Summary of the Invention

[0003] In order to solve the above technical problems or at least partially solve the above technical problems, the embodiments of the present disclosure provide a training method for a route sorting model, a route recommendation method, an apparatus and a device.

[0004] A first aspect of an embodiment of the present disclosure provides a method for training a route sorting model, comprising: obtaining training data based on historical navigation data of a navigated object, the training data comprising: preference characteristics of the navigated object for a preset navigation strategy, statistical characteristics of the navigated object's historical selection of a first recommended navigation route, route characteristics of multiple historical navigation routes, and bias characteristics related to recommendation information of the historical navigation routes; inputting the statistical characteristics and the bias characteristics of the multiple historical navigation routes into a first network in the route sorting model, and having the first network perform a first scoring process on the multiple historical navigation routes; and The route features, statistical features and preference features of multiple historical navigation routes are input into the second network in the route sorting model, and the second network performs a second scoring process on the multiple historical navigation routes; the multiple historical navigation routes are scored and sorted based on the results of the first scoring process and the results of the second scoring process; and the parameters of the second network are adjusted according to the scoring and sorting results of the multiple historical navigation routes and the actual coverage ranking results of the multiple historical navigation routes, so that the scoring and sorting results of the multiple historical navigation routes by the second network match the actual coverage ranking results of the multiple historical navigation routes.

[0005] A second aspect of the embodiments of the present disclosure provides a route recommendation method, including:

[0006] Based on the navigation starting position, end position, and target navigation strategy, at least one navigation route is determined; based on the at least one navigation route, route characteristics of the at least one navigation route are determined; the route characteristics, as well as pre-obtained preference characteristics of the navigated object for the preset navigation strategy and statistical characteristics of the navigated object's historical selection of the first recommended navigation route, are input into a preset route sorting model, and a scoring and sorting result of the at least one navigation route is determined by the route sorting model; based on the scoring and sorting result of the at least one navigation route, a recommended route is determined, and the recommended route is displayed to the navigated object.

[0007] A third aspect of the present disclosure provides a model training device, including:

[0008] a processing module configured to process historical navigation data of a navigated object to obtain training data, the training data including: a preference characteristic of the navigated object for a preset navigation strategy, statistical characteristics of the navigated object's historical selection of a first recommended navigation route, route characteristics of multiple historical navigation routes, and bias characteristics associated with recommendation information of the historical navigation routes;

[0009] a first scoring module, configured to input the statistical features and the bias features of the plurality of historical navigation routes into a first network in the route sorting model, and to allow the first network to perform a first scoring process on the plurality of historical navigation routes;

[0010] a second scoring module, configured to input the route features, the statistical features, and the preference features of the plurality of historical navigation routes into a second network in the route sorting model, and to perform a second scoring process on the plurality of historical navigation routes by the second network;

[0011] a scoring and sorting module, configured to score and sort the plurality of historical navigation routes based on a result of the first scoring process and a result of the second scoring process;

[0012] A parameter adjustment module is used to adjust the parameters of the second network according to the scoring ranking results of the multiple historical navigation routes and the actual coverage ranking results of the multiple historical navigation routes, so that the scoring ranking results of the multiple historical navigation routes by the second network match the actual coverage ranking results of the multiple historical navigation routes.

[0013] A fourth aspect of the embodiments of the present disclosure provides a route recommendation device, including:

[0014] A first determining module is used to determine at least one navigation route based on the navigation starting position, the end position, and the target navigation strategy;

[0015] A second determining module is configured to determine a route feature of the at least one navigation route based on the at least one navigation route;

[0016] a scoring module for inputting the route characteristics, the pre-obtained preference characteristics of the navigated subject for the preset navigation strategy, and the statistical characteristics of the navigated subject's historical selection of the first recommended navigation route into a preset route ranking model, and determining a scoring ranking result for the at least one navigation route by the route ranking model;

[0017] The route recommendation module is used to determine a recommended route based on the scoring and sorting results of the at least one navigation route, and to present the recommended route to the navigated object.

[0018] A fifth aspect of an embodiment of the present disclosure provides a computer device, comprising: a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor can execute the method of the first aspect or the second aspect mentioned above.

[0019] A sixth aspect of an embodiment of the present disclosure provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the processor can execute the method of the first or second aspect above.

[0020] The seventh aspect of the embodiments of the present disclosure provides a computer program product, which includes a computer program stored in a computer-readable storage medium. When the computer program is read and executed by a processor in a computer device, the computer device can execute the method of the above-mentioned second aspect.

[0021] The technical solution provided by the embodiments of the present disclosure has the following advantages over the prior art:

[0022] In an embodiment of the present disclosure, the preference characteristics of the navigated object for a preset navigation strategy, the statistical characteristics of the navigated object's historical selection of the first recommended navigation route, the route characteristics of multiple historical navigation routes, and the bias characteristics related to the recommendation information of the historical navigation routes are obtained by processing the historical navigation data of the navigated object; the processed statistical characteristics and bias characteristics are input into a first network of a route ranking model, and the first network performs a first scoring process on the multiple historical navigation routes; the processed statistical characteristics, preference characteristics, and route characteristics are input into a second network of the route ranking model, and the second network performs a second scoring process on the multiple historical navigation routes, and scores and ranks the multiple historical navigation routes based on the results of the first scoring process and the results of the second scoring process, and adjusts the parameters of the second network based on the scoring and ranking results of the multiple historical navigation routes and the actual coverage ranking results of the multiple historical navigation routes, so that the scoring and ranking results of the historical navigation routes by the second network match the actual coverage ranking results of the historical navigation routes. In practice, the more frequently a navigated object selects the first recommended navigation route as its actual navigation route, the greater the influence of the bias characteristics of the navigation route on the navigated object. The disclosed embodiment, by inputting the statistical characteristics of the navigated object's historical selection of the first recommended navigation route and the bias characteristics of multiple historical navigation routes into a first network, can help the first network obtain a scoring result that reflects the influence of the bias characteristics on the navigated object. By adding the scoring results of the first network to the training process of the second network, the second network can be helped to learn the influence of the bias characteristics of the navigation route on the navigated object, so that the trained second network can eliminate this influence through route scoring, thereby improving the accuracy of the recommended routes. Furthermore, the disclosed embodiment, by using the navigated object's preference characteristics for each preset navigation strategy, the statistical characteristics of the navigated object's historical selection of the first recommended navigation route, and the route characteristics of multiple historical navigation routes as training data for the second network, can help the second network learn the navigated object's true preferences, ensuring that the scoring results output by the second network match the navigated object's true preferences, thereby enabling the recommended routes to meet the user's personalized needs and improving the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0024] In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0025] Figure 1is a schematic diagram of a training scenario of a route sorting model provided by an embodiment of the present disclosure;

[0026] Figure 2 is a flowchart of a method for training a route sorting model provided by an embodiment of the present disclosure;

[0027] Figure 3 This is a flow chart of a method for adjusting network parameters provided by an embodiment of the present disclosure;

[0028] Figure 4 is a flowchart of a route recommendation method provided by an embodiment of the present disclosure;

[0029] Figure 5 is a structural diagram of a model training device provided by an embodiment of the present disclosure;

[0030] Figure 6 Schematic diagram of the structure of a route recommendation device provided by an embodiment of the present disclosure. DETAILED DESCRIPTION

[0031] In order to more clearly understand the above-mentioned objectives, features and advantages of the present disclosure, the scheme of the present disclosure will be further described below. It should be noted that the embodiments of the present disclosure and the features therein can be combined with each other in the absence of conflict.

[0032] In the following description, many specific details are set forth to facilitate a full understanding of the present disclosure, but the present disclosure may also be implemented in other ways different from those described herein; it is obvious that the embodiments in the specification are only part of the embodiments of the present disclosure, rather than all of the embodiments.

[0033] There are two main route recommendation methods provided by related technologies. In one method, the road sections that the navigated object (for example, a person or vehicle using a navigation application for navigation) is familiar with are given priority consideration, and the routes with more familiar road sections are scored higher when the routes are scored and sorted; in the other method, multiple navigation strategies are provided to the navigated object in advance (for example, low toll, high-speed priority, no congestion, etc.), and the corresponding routes are recommended to the navigated object based on the navigation strategy selected by the navigated object. However, the former method can only realize personalized route recommendations in areas where the navigated object frequently moves, and is not universal. The latter method can only roughly reflect the navigated object's preference for a certain navigation strategy, but cannot accurately reflect the degree of preference of the navigated object for each navigation strategy, and thus cannot provide the navigated object with a route that meets the navigated object's true preferences. Moreover, regardless of the former or the latter method, the location where the recommended route is issued (for example, the first, alternative route 1, alternative route 2, not issued, etc.) has a strong guiding effect on the navigated object. Under these guiding effects, the route selected by the navigated object is often not the route that best meets the navigated object's preferences. As a result, in practice, the navigation route often does not match the route actually taken by the navigated object, resulting in a poor experience.

[0034] In order to solve the problems existing in related technologies, the present disclosure provides a training scheme for a route sorting model. Figure 1 is a schematic diagram of a training scenario of a route sorting model provided by an embodiment of the present disclosure, such as Figure 1 As shown, the route sorting model in the embodiment of the present disclosure includes a first network and a second network. The input data of the first network includes statistical features of the first recommended navigation route historically selected by the navigated subject, as well as bias features of the navigated subject's multiple historical navigation routes. The input data of the second network includes route features of the navigated subject's multiple historical navigation routes, statistical features of the first recommended navigation route historically selected by the navigated subject, and preferences of the navigated subject for multiple preset navigation strategies (e.g., low toll, highway priority, congestion avoidance, etc.).

[0035] The multiple historical navigation routes referred to in the embodiments of the present disclosure can be understood as all routes recalled by a navigation action, or part of all routes. A navigation action refers to a complete navigation process from recalling a route to completing navigation.

[0036] The statistical features referred to in the embodiments of the present disclosure are used to reflect the degree to which the navigated object accepts the first recommended navigation route. The more frequently the navigated object selects the first recommended navigation route, the greater the influence of the bias feature. In some implementations of the embodiments of the present disclosure, the statistical features may include at least one of the following features: the number of times the navigated object has historically selected the first recommended navigation route, the number of times the navigated object has historically selected alternative recommended navigation routes, and the number of times the first recommended navigation route is the navigation route with the highest actual coverage. The actual coverage can be represented by the proportion of the overlapping portion of the historical navigation route and the actual route in the actual route.

[0037] The bias features referred to in the embodiments of the present disclosure are related to the recommendation information of historical navigation routes. In some embodiments, the bias features may include at least information such as the display position of the historical navigation route on the navigation route recommendation page (e.g., first recommended route, alternative route 1, alternative route 2, not yet issued, etc.), and the similarity between the historical navigation route and each recommended navigation route (e.g., first recommended route, alternative route 1, and alternative route 2).

[0038] In the embodiment of the present disclosure, the preference characteristics of the navigated object for the preset navigation strategy can be represented by the ratio of the number of times the navigated object initiates navigation using the preset navigation strategy within a preset time period to the total number of times navigation is initiated within the preset time period.

[0039] The route features referred to in this disclosure may include route data features and route scene features.

[0040] in,

[0041] Route data features may include: mileage, estimated time of arrival, number of traffic lights, congestion information, road grade, and the ratio of the length of the road sections familiar to the navigation subject to the total length of historical navigation routes. Route scenario features may include: the time when the navigation subject initiated navigation and the distance to the navigation destination.

[0042] See also Figure 1In the training phase, the embodiment of the present disclosure simultaneously scores multiple historical navigation routes of the navigated object through the first network and the second network, and takes the sum of the scoring results of the first network and the scoring results of the second network corresponding to the historical navigation route as the final score of the historical navigation route. Then, the multiple historical navigation routes are scored and sorted according to the final scores of the multiple historical navigation routes. If the historical navigation route with the highest score does not match the actual route (for example, the route with the highest score is different from the actual route, or the actual coverage of the historical navigation route with the highest score is less than a preset threshold), the parameters of the second network are adjusted according to the scoring and sorting results of the above-mentioned multiple historical navigation routes and the actual coverage ranking results of these historical navigation routes, until the scoring and sorting results output by the second network match the actual coverage ranking results of the multiple historical navigation routes. Among them, the matching of the scoring and sorting results with the actual coverage ranking results can be exemplarily understood as the same as the scoring and sorting results with the actual coverage ranking results, or the similarity between the scoring and sorting results with the actual coverage ranking results is greater than a preset threshold.

[0043] In actual route recommendation scenarios, the second network trained by the embodiment of the present disclosure can be used alone. That is, in actual applications, the second network in the embodiment of the present disclosure can be used alone to score and sort navigation routes, and then recommend routes to users based on the scoring and sorting results of the second network.

[0044] In an embodiment of the present disclosure, statistical features of the first recommended navigation route historically selected by a navigated object and bias features of multiple historical navigation routes are input into a first network to obtain a scoring result reflecting the influence of the bias features of the navigation route on the navigated object. The scoring result is added to the training process of a second network to help the second network learn the influence of the bias features of the navigation route on the navigated object. The preference features of the navigated object for each preset navigation strategy, the statistical features of the first recommended navigation route historically selected by the navigated object, and the route features of multiple historical navigation routes are used as training data for the second network to help the second network learn the true preferences of the navigated object. This ensures that the scoring results output by the second network match the true preferences of the navigated object, thereby improving the user experience.

[0045] In order to better understand the technical solutions of the embodiments of the present disclosure, the technical solutions of the embodiments of the present disclosure are described below in conjunction with some exemplary embodiments.

[0046] Figure 2 This is a flow chart of a method for training a route sorting model provided by an embodiment of the present disclosure. Figure 2 As shown, the method can be executed by a computer device with computing and processing capabilities (eg, a server or a computing node in a data processing cluster).

[0047] like Figure 2 As shown, the method includes the following steps:

[0048] Step 201: Training data is obtained based on the historical navigation data of the navigated object. The training data includes: the navigated object's preference characteristics for a preset navigation strategy, statistical characteristics of the navigated object's historical selection of the first recommended navigation route, route characteristics of multiple historical navigation routes, and bias characteristics related to recommendation information of the historical navigation routes.

[0049] Step 202: Input the processed statistical features and the bias features of the plurality of historical navigation routes into a first network in a route ranking model, and perform a first scoring process on the plurality of historical navigation routes by the first network.

[0050] Step 203: Input the processed statistical features, preference features, and route features of the plurality of historical navigation routes into a second network in the route ranking model, and perform a second scoring process on the plurality of historical navigation routes by the second network.

[0051] Step 204: Score and sort the multiple historical navigation routes based on the results of the first scoring process and the results of the second scoring process.

[0052] Step 205: Adjust the parameters of the second network according to the scoring ranking results and the actual coverage ranking results of the multiple historical navigation routes, so that the scoring ranking results of the multiple historical navigation routes by the second network match the actual coverage ranking results of the multiple historical navigation routes.

[0053] In the embodiments of the present disclosure, historical navigation data can be understood as navigation data of a navigated object (e.g., a person or vehicle using a navigation application for navigation) within a preset time period. This historical navigation data includes navigation data of at least one navigation initiated by the navigated object. The navigation data of a navigation session includes at least one of the following data:

[0054] Multiple historical navigation routes planned based on the starting and ending positions of navigation, the navigation strategy selected by the navigated object (for example, low toll, highway priority, no highway, avoiding congestion, etc.), the route selected by the navigated object from the recommended navigation routes and the actual route taken, the corresponding sending positions of the historical navigation routes (for example, first, alternative route 1, alternative route 2, not sent, etc.), and the route features corresponding to each historical navigation route, wherein the route features include route data features and route scene features, wherein the route data features may include, for example, mileage, estimated arrival time, number of traffic lights, congestion information, road grade, the proportion of the length of the road section familiar to the navigated object in the historical navigation route to the total length of the historical navigation route, etc.; the route scene features may include, for example, the time when the navigated object initiated navigation, the straight-line distance between the starting point and the end point, that is, the distance to the navigation destination, the season when the navigation was initiated, the holiday category to which the navigation date belongs, etc., wherein the holiday category may include, for example, working days and non-working days, and the non-working days may include, for example, weekends and statutory holidays, etc.

[0055] The preset navigation strategy referred to in the embodiment of the present disclosure can be exemplarily understood as the preference options pre-set in the navigation application, such as low toll, high-speed priority, no high-speed, avoiding congestion, shortest distance, shortest time, etc. There can be multiple preset navigation strategies in the embodiment of the present disclosure, and the navigated object can select one or more preset navigation strategies as the target navigation strategy on the configuration interface provided by the navigation application.

[0056] In an embodiment of the present disclosure, a preference characteristic of a navigated object for a preset navigation strategy can be used to represent the degree of preference of the navigated object for the preset navigation strategy. For example, in one embodiment of the present disclosure, the number of times a navigated object initiated navigation using a preset navigation strategy in historical navigation data and the total number of times the navigated object initiated navigation in the historical navigation data can be counted, and the ratio of the number of times the navigated object initiated navigation using the preset navigation strategy to the total number of times can be used as a characteristic of the user's preference for the preset navigation strategy. For another example, in another embodiment of the present disclosure, the number of times a navigated object switched to a preset navigation strategy in historical navigation data can be counted, and the ratio of the number of times the navigated object switched to a preset navigation strategy to the total number of times the navigated object initiated navigation can be used as a characteristic of the user's preference for the preset navigation strategy. Since the action of a navigated object switching navigation strategies can often reflect the navigated object's preference for navigation strategies, the accuracy of the preference characteristic can be improved by counting the number of times a navigated object switched to a preset navigation strategy and using the ratio of the number of times the navigated object switched to a preset navigation strategy to the total number of times navigation was sent as a characteristic of the user's preference for the preset navigation strategy.

[0057] The bias feature referred to in the embodiment of the present disclosure is related to the recommendation information of the historical navigation route. The bias feature includes at least the display position of the historical navigation route on the navigation route recommendation page (for example, the first recommended route, alternative route 1, alternative route 2, not issued, etc.), and the similarity between the historical navigation route and each recommended navigation route (for example, the first recommended route, alternative route 1 and alternative route 2). Among them, the display position (i.e., the issuance position) of the historical navigation route on the navigation route recommendation page can be directly extracted from the historical navigation data of the navigated object. The similarity between the historical navigation route and each recommended route can be calculated according to a preset algorithm. For example, when a navigation application recommends three routes to a target, namely the first recommended route, alternative route 1, and alternative route 2, then for a particular navigation route (including the recommended route itself), its corresponding bias features include the location of issuance (e.g., first recommended navigation route, alternative route 1, alternative route 2, not issued, etc.), as well as its first similarity with the first recommended navigation route, its second similarity with alternative route 1, and its third similarity with alternative route 3. Of course, this is merely an example, not a limitation. In practice, the number of recommended routes and the method for calculating similarity can be set as needed, without being limited to a specific number of recommendations or a specific similarity calculation method.

[0058] The statistical features referred to in the embodiment of the present disclosure are used to reflect the degree to which the navigated object accepts the first recommended navigation route. The statistical features may include at least one of the following: the number of times the navigated object has historically selected the first recommended navigation route, the number of times the navigated object has historically selected an alternative recommended navigation route, and the number of times the first recommended navigation route is the navigation route with the largest actual coverage. In the embodiment of the present disclosure, the greater the frequency with which the navigated object selects the first recommended navigation route as the navigation route, the higher the degree to which the navigated object accepts the first recommended navigation route, and the greater the influence of the bias features of the route on the navigated object. By using the statistical features of the historical selection of the first recommended navigation route by the navigated object as training data, the route sorting model can be helped to learn the influence of the bias features on the navigated object, so that the route sorting model can eliminate the influence of the bias features of the route on the navigated object by improving the score of the route that meets the preferences of the navigated object.

[0059] The route ranking model in the embodiments of the present disclosure can be exemplarily understood as a route ranking model implemented based on a deep ranking model (Navigation-bias-Award Deep Ranking Net, abbreviated as NADRN), but is not limited to NADRN. For example, in some embodiments, the route ranking model referred to in the embodiments of the present disclosure can also be implemented based on other deep neural networks (such as the AutoInt model, etc.).

[0060] During the training phase, the route sorting model referred to in the embodiment of the present disclosure includes a first network and a second network. The first network and the second network are both scoring networks, which are used to score historical navigation routes. The input data of the first network include the statistical features of the first recommended navigation route historically selected by the navigated object and the bias features of multiple historical navigation routes. The input data of the second network include the route features of multiple historical navigation routes, the statistical features of the first recommended navigation route historically selected by the navigated object, and the preference features of the navigated object for multiple preset navigation strategies. The first network is used to help the second network learn the influence of the bias features of the navigation route on the navigated object, and the second network is used to learn the preference of the navigated object for the navigation route, and eliminate the influence of the bias features on the navigated object by scoring the navigation route.

[0061] In an exemplary embodiment, the first network and the second network can use a list method (or also called a listwise method) to process the input data. Specifically, after the input data is input into the first network and the second network, the first network and the second network perform batch normalization (BN) on the data, and then classify the normalized data based on a preset exponential linear unit (ELU) activation function to obtain the scoring results corresponding to each historical navigation route. This method uses the listwise method to score each historical navigation route, which can avoid the problem of missing comparison information between routes in the pairing method (or also called the pointwise method) and the problem that the pairwise method may not be able to select the route that best meets the user's preferences from a large number of historical navigation routes.

[0062] Furthermore, the sum of the scores of the first network and the second network can be used as the final score of the historical navigation route, and then the final scores of the multiple historical navigation routes input into the first network and the second network are ranked. If the highest-ranked historical navigation route does not match the actual route taken by the navigated object in this navigation (inconsistent or the actual coverage of the highest-ranked historical navigation route is less than a preset threshold), the parameters of the second network are adjusted according to the current score ranking result and the actual coverage ranking results of these historical navigation routes, so that the score ranking result output by the adjusted second network matches the actual coverage ranking results of these historical navigation routes (consistent or the ranking similarity is greater than the preset similarity).

[0063] In the embodiment of the present disclosure, by inputting the statistical features of the first recommended navigation route historically selected by the navigated object and the bias features of multiple historical navigation routes into the first network, a scoring result reflecting the impact of the bias features of the navigation route on the navigated object can be obtained. By adding this scoring result to the training process of the second network, the second network can be helped to learn the impact of the bias features of the navigation route on the navigated object, so that the trained second network can eliminate this impact through route scoring, thereby improving the accuracy of the recommended routes. In addition, by using the preference features of the navigated object for each preset navigation strategy, the statistical features of the first recommended navigation route historically selected by the navigated object, and the route features of multiple historical navigation routes as training data for the second network, the embodiment of the present disclosure can help the second network learn the true preferences of the navigated object, ensure that the scoring results output by the second network match the true preferences of the navigated object, and thereby enable the recommended routes to meet the personalized needs of the user, thereby improving the user experience.

[0064] Figure 3 This is a flow chart of a method for adjusting network parameters provided by an embodiment of the present disclosure, such as Figure 3 As shown, the method includes:

[0065] Step 301: for each of the plurality of historical navigation routes, determine the actual walking coverage of the historical navigation route based on the difference between the historical navigation route and the actual walking route.

[0066] Step 302: weighting the sum of the scores of the historical navigation routes based on the actual coverage of the historical navigation routes to obtain a weighted result.

[0067] Step 303: average the multiple weighted results corresponding to the multiple historical navigation routes to obtain an average result.

[0068] Step 304: Based on the averaging result, adjust the parameters of the second network so that the averaging result of the second network is greater than a preset threshold.

[0069] In the disclosed embodiments, the actual coverage rate of the historical navigation route can be exemplified by the ratio of the length of the overlap between the historical navigation route and the actual route to the total length of the actual route. For example, if the overlap between the historical navigation route and the actual route is 50 kilometers, and the total length of the actual route is 100 kilometers, the actual coverage rate of the historical navigation route is 50%. Of course, this is only an example and not a limitation.

[0070] In one implementation of the disclosed embodiment, the scoring results of the first network and the scoring results of the second network corresponding to the historical navigation route can be summed to obtain a corresponding scoring summation result. The actual walking coverage of the historical navigation route is then used as a weight to weight the scoring summation result of the historical navigation route to obtain a corresponding weighted result, and then the multiple weighted results corresponding to multiple historical navigation routes (which can be exemplarily understood as all recalled routes corresponding to the navigation action, or routes with actual walking coverage or current scoring ranking higher than a preset threshold among all recalled routes) are averaged to obtain an average result, and then the parameters of the second network are adjusted according to the above-mentioned average processing result until the average processing result is greater than the preset threshold. In practice, a larger average processing result indicates that historical navigation routes with higher actual coverage rates are ranked higher in the scoring ranking. In other words, the higher the similarity between the scoring ranking output by the second network and the actual coverage ranking results of multiple historical navigation routes, the larger the average processing result. When the scoring ranking results of multiple historical navigation routes are consistent with the actual coverage ranking results of these historical navigation routes, the average processing result reaches its maximum value. Therefore, in practice, when the average processing result is greater than a preset threshold, the training goal can be considered achieved, and training of the second network model can be stopped, and the parameters of the second network at this time can be fixed.

[0071] This embodiment adjusts the parameters of the second network based on the scoring and coverage ranking results of multiple historical navigation routes, so that the second network can give high scores to navigation routes with high coverage rates, thereby matching the recommended routes with the user's actual preferences and improving the user experience.

[0072] Figure 4 is a flow chart of a route recommendation method provided by an embodiment of the present disclosure, such as Figure 4 As shown, the method includes:

[0073] Step 401: Determine at least one navigation route based on the navigation start position, the end position, and the target navigation strategy.

[0074] Step 402: Based on the navigation routes, determine route features of the navigation routes.

[0075] Step 403: Input the route characteristics of the navigation routes, the pre-obtained preference characteristics of the navigated object for the preset navigation strategy, and the statistical characteristics of the navigated object's historical selection of the first recommended navigation route into a preset route ranking model, and the route ranking model determines the scoring and ranking results of the navigation routes.

[0076] Step 404: Determine a recommended route based on the scoring and ranking results of the navigation routes, and display the recommended route to the navigated object.

[0077] The route sorting model in the embodiment of the present disclosure can be based on the above Figure 2 The method of the embodiment is trained. The route features, pre-determined preference features, and statistical features of the navigation routes are input into the route ranking model. The route ranking model scores and ranks the planned navigation routes based on these input data. The top three navigation routes with the highest scores are recommended as the routes to be navigated.

[0078] The beneficial effects of this embodiment are Figure 2 The embodiments are similar and will not be described again here.

[0079] Figure 5 This is a schematic diagram of the structure of a model training device provided by an embodiment of the present disclosure. The device can be understood as Figure 2 The computer device or some functional modules in the computer device referred to in the embodiment. Figure 5 As shown, the model training device 50 includes:

[0080] a processing module 51 configured to process historical navigation data of a navigated subject to obtain training data, the training data including: a preference characteristic of the navigated subject for a preset navigation strategy, statistical characteristics of the navigated subject's historical selection of a first recommended navigation route, route characteristics of multiple historical navigation routes, and bias characteristics associated with recommendation information of the historical navigation routes;

[0081] a first scoring module 52, configured to input the statistical features and the bias features of the plurality of historical navigation routes into a first network in the route sorting model, and to allow the first network to perform a first scoring process on the plurality of historical navigation routes;

[0082] A second scoring module 53 is configured to input the route features, the statistical features, and the preference features of the plurality of historical navigation routes into a second network in the route sorting model, and to perform a second scoring process on the plurality of historical navigation routes by the second network;

[0083] A scoring and sorting module 54 is configured to score and sort the plurality of historical navigation routes based on the result of the first scoring process and the result of the second scoring process;

[0084] The parameter adjustment module 55 is used to adjust the parameters of the second network according to the scoring ranking results of the multiple historical navigation routes and the actual coverage ranking results of the multiple historical navigation routes, so that the scoring ranking results of the multiple historical navigation routes by the second network match the actual coverage ranking results of the multiple historical navigation routes.

[0085] In one embodiment, the processing module 51 is used to process the historical navigation data of the navigated object to obtain the preference characteristics of the navigated object for the preset navigation strategy, based on the historical navigation data of the navigated object, to obtain the number of times the navigated object initiated navigation based on each preset navigation strategy, and the total number of times the navigated object initiated navigation; determine the proportion of the number of times the navigated object initiated navigation based on each preset navigation strategy to the total number of times, and determine the proportion as the preference characteristics of the navigated object for the preset navigation strategy.

[0086] In one embodiment, the statistical features include at least one of the following: the number of times the navigated object historically selected the first recommended navigation route, the number of times the navigated object historically selected an alternative recommended navigation route, and the number of times the first recommended navigation route was the navigation route with the highest actual coverage.

[0087] In one embodiment, the route features include: route data features and route scene features;

[0088] The route data features include: mileage, estimated arrival time, number of traffic lights, congestion information, road grade, and the ratio of the length of the familiar road section to the length of the historical navigation route;

[0089] The route scenario features include: the time when the navigation object initiates navigation and the distance to the navigation destination.

[0090] In one embodiment, the processing module 51 is used to perform an operation of obtaining bias features related to recommendation information of the historical navigation route based on the historical navigation data of the navigated object: based on the historical navigation data of the navigated object, obtain the display position of the historical navigation route on the navigation route recommendation page, and the similarity between the historical navigation route and each recommended navigation route.

[0091] In one embodiment, the scoring and sorting module 54 is used to: for each historical navigation route, sum the results of the first scoring process and the results of the second scoring process corresponding to the historical navigation route to obtain the score sum result of the historical navigation route; and score and sort the score sum results of the multiple historical navigation routes.

[0092] In one embodiment, the parameter adjustment module 55 is used to determine, for each historical navigation route, the actual coverage rate of the historical navigation route based on the difference between the historical navigation route and the actual route; weight the score sum results of the historical navigation routes based on the actual coverage rate to obtain a weighted result; average the multiple weighted results corresponding to the multiple historical navigation routes to obtain an average result; and adjust the parameters of the second network based on the average result so that the average result is greater than a preset threshold.

[0093] The device provided in this embodiment can be used to perform Figure 2 and Figure 3 The method of the embodiment, its execution mode and beneficial effects are similar and will not be repeated here.

[0094] Figure 6 is a structural diagram of a route recommendation device provided by an embodiment of the present disclosure, such as Figure 6 As shown, the route recommendation device 60 includes:

[0095] A first determining module 61 is configured to determine at least one navigation route based on the navigation start location, the end location, and the target navigation strategy;

[0096] A second determining module 62 is configured to determine a route feature of the at least one navigation route based on the at least one navigation route;

[0097] a scoring module 63 for inputting the route characteristics, the pre-obtained preference characteristics of the navigated subject for the preset navigation strategy, and the statistical characteristics of the navigated subject's historical selection of the first recommended navigation route into a preset route ranking model, and determining a scoring ranking result for the at least one navigation route using the route ranking model;

[0098] The route recommendation module 64 is configured to determine a recommended route based on the scoring and ranking result of the at least one navigation route, and present the recommended route to the navigated object.

[0099] The device provided in this embodiment can be used to perform Figure 4 The method of the embodiment, its execution mode and beneficial effects are similar and will not be repeated here.

[0100] The present disclosure also provides a computer device, which includes a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor can perform the above Figure 2-Figure 4 The method of any embodiment.

[0101] The present disclosure also provides a computer-readable storage medium in which a computer program is stored. When the computer program is executed by a processor, the processor can perform the above Figure 2-Figure 4 The method of any embodiment.

[0102] The present disclosure also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. When the computer program is read and executed by a processor in a computer device, the computer device can perform the above-mentioned Figure 4 Method of embodiment.

[0103] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0104] The foregoing description is intended only to provide specific embodiments of the present disclosure, intended to enable those skilled in the art to understand and implement the present disclosure. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present disclosure. Therefore, the present disclosure is not intended to be limited to the embodiments described herein, but rather to be construed in the broadest manner consistent with the principles and novel features disclosed herein.

Claims

1. A method for training a route sorting model, comprising: Training data is obtained based on historical navigation data of the navigated object, the training data including: a preference characteristic of the navigated object for a preset navigation strategy, statistical characteristics of the navigated object's historical selection of a first recommended navigation route, route characteristics of multiple historical navigation routes, and bias characteristics associated with recommendation information of the historical navigation routes; Inputting the statistical features and the bias features of the plurality of historical navigation routes into a first network in the route sorting model, and allowing the first network to perform a first scoring process on the plurality of historical navigation routes; inputting the route features, the statistical features, and the preference features of the plurality of historical navigation routes into a second network in the route sorting model, and performing a second scoring process on the plurality of historical navigation routes by the second network; Sorting the plurality of historical navigation routes by scores based on a result of the first scoring process and a result of the second scoring process; According to the scoring and ranking results of the multiple historical navigation routes and the actual coverage ranking results of the multiple historical navigation routes, the parameters of the second network are adjusted so that the scoring and ranking results of the multiple historical navigation routes by the second network match the actual coverage ranking results of the multiple historical navigation routes.

2. The method according to claim 1, wherein The preference characteristics of the navigated object for the preset navigation strategy are obtained based on the historical navigation data of the navigated object, including: Based on historical navigation data of the navigated object, obtaining the number of times the navigated object initiates navigation based on each preset navigation strategy and the total number of times the navigated object initiates navigation; The ratio of the number of times the navigated object initiates navigation based on each preset navigation strategy to the total number of times is determined, and the ratio is determined as a preference feature of the navigated object for the preset navigation strategy.

3. The method according to claim 1, wherein The statistical features include at least one of the following: the number of times the navigated object has historically selected the first recommended navigation route, the number of times the navigated object has historically selected an alternative recommended navigation route, and the number of times the first recommended navigation route is the navigation route with the highest actual coverage.

4. The method according to claim 1, wherein The route features include: route data features and route scene features; The route data features include: Mileage, estimated time of arrival, number of traffic lights, congestion information, road grade, and the ratio of the length of familiar road sections to the length of historical navigation routes; The route scenario features include: the time when the navigation object initiates navigation and the distance to the navigation destination.

5. The method according to claim 1, wherein Based on the historical navigation data of the navigated object, bias features related to the recommendation information of the historical navigation route are obtained, including: Based on the historical navigation data of the navigated object, the display position of the historical navigation route on the navigation route recommendation page and the similarity between the historical navigation route and each recommended navigation route are obtained.

6. The method according to any one of claims 1 to 5, wherein The scoring and sorting the plurality of historical navigation routes based on the result of the first scoring process and the result of the second scoring process includes: For each historical navigation route, summing the result of the first scoring process and the result of the second scoring process corresponding to the historical navigation route to obtain a summed score result of the historical navigation route; The summed scores of the multiple historical navigation routes are ranked by score.

7. The method according to claim 6, wherein: The adjusting the parameters of the second network according to the scoring ranking results of the plurality of historical navigation routes and the actual coverage ranking results of the plurality of historical navigation routes includes: For each historical navigation route, determining the actual coverage rate of the historical navigation route based on the difference between the historical navigation route and the actual route; Performing weighted processing on the score sum results of the historical navigation routes based on the actual coverage rate to obtain a weighted result; Averaging the multiple weighted results corresponding to the multiple historical navigation routes to obtain an average result; Based on the averaging result, parameters of the second network are adjusted so that the averaging result is greater than a preset threshold.

8. A route recommendation method, comprising: Determine at least one navigation route based on the navigation start location, the end location, and the target navigation strategy; Determining, based on the at least one navigation route, route characteristics of the at least one navigation route; Inputting the route characteristics, the pre-obtained preference characteristics of the navigated subject for the preset navigation strategy, and the statistical characteristics of the navigated subject's historical selection of the first recommended navigation route into a preset route ranking model trained using the route ranking model training method according to any one of claims 1 to 7, and determining a scoring ranking result for the at least one navigation route by the route ranking model; Based on the scoring and ranking result of the at least one navigation route, a recommended route is determined, and the recommended route is displayed to the navigated object.

9. A training device for a route sorting model, comprising: a processing module configured to process historical navigation data of a navigated object to obtain training data, the training data including: a preference characteristic of the navigated object for a preset navigation strategy, statistical characteristics of the navigated object's historical selection of a first recommended navigation route, route characteristics of multiple historical navigation routes, and bias characteristics associated with recommendation information of the historical navigation routes; a first scoring module, configured to input the statistical features and the bias features of the plurality of historical navigation routes into a first network in the route sorting model, and to allow the first network to perform a first scoring process on the plurality of historical navigation routes; a second scoring module, configured to input the route features, the statistical features, and the preference features of the plurality of historical navigation routes into a second network in the route sorting model, and to perform a second scoring process on the plurality of historical navigation routes by the second network; a scoring and sorting module, configured to score and sort the plurality of historical navigation routes based on a result of the first scoring process and a result of the second scoring process; A parameter adjustment module is used to adjust the parameters of the second network according to the scoring ranking results of the multiple historical navigation routes and the actual coverage ranking results of the multiple historical navigation routes, so that the scoring ranking results of the multiple historical navigation routes by the second network match the actual coverage ranking results of the multiple historical navigation routes.

10. A route recommendation device, comprising: A first determining module is used to determine at least one navigation route based on the navigation starting position, the end position, and the target navigation strategy; A second determining module is configured to determine a route feature of the at least one navigation route based on the at least one navigation route; a scoring module, configured to input the route characteristics, the pre-obtained preference characteristics of the navigated subject for the preset navigation strategy, and the statistical characteristics of the navigated subject's historical selection of the first recommended navigation route, into a preset route ranking model trained using the route ranking model training method according to any one of claims 1 to 7, and determine a scoring ranking result for the at least one navigation route using the route ranking model; The route recommendation module is used to determine a recommended route based on the scoring and sorting results of the at least one navigation route, and to present the recommended route to the navigated object.

11. A computer device comprising: A processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the route sorting model training method according to any one of claims 1 to 7 or the route recommendation method according to claim 8.

12. A computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the route ranking model training method according to any one of claims 1 to 7 or the route recommendation method according to claim 8.

13. A computer program product, the computer program product comprising: A computer program, wherein the computer program is stored in a computer-readable storage medium. When the computer program is read and executed by a processor in a computer device, the computer device executes the route sorting model training method according to any one of claims 1 to 7 or the route recommendation method according to claim 8.

Images