Vehicle use recommendation method based on vehicle portrait
By building a vehicle portrait and similarity network, combining Kmeans++ clustering and vehicle-user binary graph projection method, the problem that existing vehicle management platforms are difficult to provide personalized vehicle recommendations is solved, and users' precise vehicle recommendations and vehicle resource management are achieved.
Patent Information
- Application Number
- CN202411553233.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-01
- Publication Date
- 2025-03-14
AI Technical Summary
The existing vehicle management platform is difficult to provide personalized and accurate vehicle recommendations when users perform work tasks, and cannot accurately match users' actual vehicle preferences, specific needs and work scenarios, resulting in a decrease in work efficiency and user satisfaction.
The vehicle-like recommendation method based on vehicle portrait is adopted, and the vehicle portrait and vehicle similarity network is constructed through data collection and cleaning, preliminary screening, and the vehicle-like similarity network is adjusted and personalized recommendation is performed by combining the Kmeans++ clustering algorithm and the vehicle-user binary graph projection method.
It realizes accurate vehicle recommendations for users, improves work efficiency and travel experience, optimizes vehicle resource management, reduces operating costs, and achieves accurate docking and efficient matching of vehicle resources and user needs.
Smart Images
Figure CN119622080A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of vehicle networking, and in particular to a vehicle recommendation method based on vehicle portraits. Background Art
[0002] As an important part of the intelligent transportation system, the current Internet of Vehicles technology is booming at an unprecedented speed. The core of Internet of Vehicles technology is to achieve comprehensive perception and intelligent interconnection of vehicles and related information. By integrating multi-source data of vehicles, such as vehicle registration information, usage information, etc., a detailed and multi-dimensional vehicle portrait can be constructed. This portrait not only includes the basic physical properties of the vehicle, but also incorporates dynamic information such as vehicle usage habits, maintenance records, etc., thus providing the possibility for in-depth understanding of the real needs of the vehicle and its users. In addition, personalized and customized services have become the new favorite of the market. For vehicle users, they are no longer satisfied with traditional and standardized travel services, but are more looking forward to getting a customized travel experience based on their preferences, needs and scenarios. Therefore, how to use vehicle portraits, combined with advanced recommendation algorithms, to achieve accurate recommendations for vehicle users provides a new research direction for vehicle management.
[0003] The defects and shortcomings of the existing technology are that, under the current technical background, if users need to call for vehicles when performing work tasks, the existing vehicle management platform often lacks personalization and accuracy, and it is difficult to accurately match the user's actual vehicle preferences, specific needs and work scenarios, which invisibly reduces work efficiency and user satisfaction. In view of this, it is urgent to develop an intelligent vehicle recommendation algorithm that can deeply analyze and understand each user's personalized preferences, real-time work needs and specific application scenarios, so as to provide customized vehicle recommendations, aiming to create a more efficient, comfortable and personal work travel experience for users. Summary of the invention
[0004] The purpose of the present invention is to solve the shortcomings of the prior art and propose a vehicle recommendation method based on vehicle portraits, which can optimize vehicle management and improve operational efficiency for enterprises, bring more convenient, comfortable and safe travel experience to the majority of vehicle users, and provide vehicle users with more personalized and intelligent travel services.
[0005] A vehicle recommendation method based on vehicle portraits of the present invention comprises the following steps:
[0006] S1. Data collection and cleaning: delete unnecessary empty records, handle outliers, fill missing data, remove duplicate fields, handle erroneous values, unify data formats, and standardize processing;
[0007] S2. Perform a preliminary screening based on the user's travel needs, destination, departure time, vehicle availability and travel mode, and select the vehicles that the user can use;
[0008] S3. Build a vehicle profile, determine one or more grouping criteria for the real-time operation data of the vehicle according to business requirements and data characteristics, group the data based on the selected characteristics, perform cluster analysis on the selected groups, and divide the data into different clusters;
[0009] S4. Construct a vehicle similarity network and present the positive and negative evaluation information collected by users after using the vehicle in a bipartite graph; users and vehicles are regarded as two independent sets, and the positive or negative evaluation of the user on the vehicle is represented by the edge connecting the two sets. The sign of the edge directly corresponds to the type of evaluation, which clearly shows the interactive relationship between users and vehicles;
[0010] S5 adjusts the vehicle similarity network;
[0011] S6. Realize personalized recommendations, optimize the vehicle similarity network based on vehicle portraits, and make recommendations based on the adjusted vehicle similarity network.
[0012] Preferably, in step S3, the vehicle data needs to be input into the Kmeans++ clustering algorithm to perform a clustering process. The algorithm will divide the data points into different clusters and generate vehicle portraits in parallel according to the similarity of the data and the set parameters.
[0013] Preferably, in step S4, the vehicle set in the bipartite graph is mapped to a new space based on a vehicle-user bipartite graph projection method, and the similarity between vehicles is quantified by counting and analyzing the positive and negative common evaluations of users on each vehicle;
[0014] The similarity between vehicle nodes is calculated by comprehensively evaluating the number of users who have made common evaluations, the proportion of common positive evaluations, and the impact of common negative evaluations;
[0015] Through the similarity score, a vehicle similarity network is built; the edge weight is a direct reflection of the similarity score. Through the similarity network, we can intuitively see which vehicles have a higher similarity in the minds of users.
[0016] Preferably, in step S5, the vehicle similarity network is adjusted using a similarity attenuation factor based on the vehicle similarities formed by the users' preferences for vehicles and the similarities of the vehicles at the attribute level to obtain a final vehicle similarity network.
[0017] Preferably, in step S6, the vehicle with the highest score given by the user most recently is obtained, and then the possibility of generating links between heterogeneous nodes without edges in the vehicle-user bipartite graph is calculated based on the adjusted vehicle similarity network, and the recommendation scores are sorted in descending order, and the first L vehicles are taken from the sorted vehicle list to form a recommendation list for recommendation, and finally the recommendation algorithm is evaluated using appropriate evaluation indicators.
[0018] The advantages and technical effects of the present invention are:
[0019] The present invention constructs a vehicle-user bipartite graph through vehicle-user interaction, user information, and vehicle information, and constructs a vehicle similarity network based on the vehicle-user bipartite graph projection method. At the same time, according to the Kmeans++ clustering results, the vehicle similarity network is adjusted using a similarity attenuation factor, and the possibility of generating links between heterogeneous nodes that do not have edges in the vehicle-user bipartite graph is calculated based on the adjusted vehicle similarity network, so as to complete the recommendation of a personalized vehicle list. The invention not only takes into account the historical behavior patterns of users in selecting vehicles and their preference characteristics for vehicles, but also adds real-time data analysis and prediction of vehicles, thereby achieving accurate prediction and personalized recommendation of users' vehicle selection. The invention not only provides enterprises with efficient vehicle management tools to help them accurately manage vehicle resources and optimize operating costs, but also provides convenient vehicle selection solutions for staff using vehicles, improves the work efficiency and travel experience of staff, and achieves accurate docking and efficient matching of vehicle resources with user needs. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] Figure 1 The figure is a flow chart of the vehicle recommendation method based on vehicle portrait of the present invention.
[0021] Figure 2 A schematic diagram of the process of extracting vehicle attribute information and constructing a vehicle portrait in the present invention.
[0022] Figure 3 The figure is a flow chart of constructing a vehicle similarity network through vehicle-user bipartite graph projection in the present invention.
[0023] Figure 4 The figure is a flow chart of optimizing the vehicle similarity network through the similarity attenuation factor and performing personalized recommendation according to the present invention. DETAILED DESCRIPTION
[0024] The embodiments of the invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the invention, and cannot be understood as limiting the invention.
[0025] The present invention aims to solve the problems that the current vehicle management platform lacks personalization and accuracy and cannot accurately match users' actual vehicle preferences, specific needs and work scenarios. To this end, the present invention provides a vehicle recommendation method based on vehicle portraits, comprising the following steps:
[0026] 1. Data collection and cleaning: delete unnecessary empty records, handle outliers, fill in missing data, deduplicate fields, handle erroneous values, unify data formats, and perform standardized processing.
[0027] 2. Conduct preliminary screening based on the user's travel needs, destination, departure time, vehicle availability and travel mode, and select vehicles that the user can use.
[0028] 3. Build vehicle portrait
[0029] The real-time operation data of the vehicle is used to determine one or more grouping criteria based on business needs and data characteristics, and the data is grouped based on the selected characteristics. Cluster analysis is performed on the selected groups to divide the data into different clusters. The vehicle data is input into the Kmeans++ clustering algorithm to perform the clustering process. The algorithm will divide the data points into different clusters and parallelize them into vehicle portraits based on the similarity of the data and the set parameters.
[0030] 4. Constructing a vehicle similarity network
[0031] The positive and negative evaluation information collected after users use the vehicle is presented in a bipartite graph. Users and vehicles are regarded as two independent sets, and the positive or negative evaluation of the vehicle by the user is represented by the edge connecting the two sets. The sign of the edge directly corresponds to the type of evaluation, clearly showing the interactive relationship between users and vehicles.
[0032] Based on the vehicle-user bipartite graph projection method, the vehicle set in the bipartite graph is mapped to a new space, and the similarity between vehicles is quantified by counting and analyzing the users' positive and negative common evaluations of each vehicle.
[0033] The similarity between vehicle nodes is calculated by performing a comprehensive evaluation through the number of users who jointly evaluate, the proportion of common positive evaluations, and the impact of common negative evaluations.
[0034] Through the similarity score, a vehicle similarity network is built. The edge weight is a direct reflection of the similarity score. Through the similarity network, we can intuitively see which vehicles have a high similarity in the minds of users.
[0035] 5. Adjust the vehicle similarity network
[0036] The vehicle similarity network is based on the actual evaluation behavior of users and the similarity of vehicles formed according to the preferences of users for vehicles, while the vehicle groups divided by the clustering algorithm reflect the similarity of vehicles at the attribute level. In order to combine the advantages of both aspects, the vehicle similarity network is adjusted using the similarity attenuation factor to obtain the final vehicle similarity network.
[0037] 6. Personalized recommendations
[0038] The vehicle similarity network is optimized based on vehicle portraits, and recommendations are made based on the adjusted vehicle similarity network. The vehicle with the highest score given by the user in the most recent time is obtained, and then the possibility of generating links between heterogeneous nodes that do not have edges in the vehicle-user bipartite graph is calculated based on the adjusted vehicle similarity network. The recommendation scores are sorted in descending order, and the first L vehicles are taken from the sorted vehicle list to form a recommendation list for recommendation. Finally, the recommendation algorithm is evaluated using appropriate evaluation indicators.
[0039] In order to more clearly illustrate the specific embodiments of the present invention, several embodiments are provided below:
[0040] Embodiment 1, as Figure 2 As shown, the vehicle portrait constructed in this embodiment can be implemented using an independent server or a server cluster composed of multiple servers, and the method includes:
[0041] S1. Data collection and cleaning:
[0042] Collect various types of original operational data generated by vehicles during operation. The collection methods include but are not limited to data information fed back in real time by on-board sensors, information sent by mobile communication devices, and information uploaded by personnel. The original data obtained through these channels usually covers multiple aspects of the vehicle. These original data may include but are not limited to basic vehicle information, vehicle technical specifications, vehicle size and weight, vehicle usage information, etc. Data cleaning after data collection can ensure the accuracy, completeness and consistency of the data.
[0043] S2. Preliminary vehicle screening:
[0044] A preliminary screening is conducted based on the user's travel needs, destination, departure time, vehicle availability and travel mode to select vehicles that the user can use.
[0045] S3. Build vehicle profile:
[0046] To meet specific business needs and make full use of the inherent characteristics of the data, design one or more grouping criteria. These grouping criteria are formulated based on business understanding, data availability, and analysis goals, and may include vehicle mileage, driving time, speed distribution, energy consumption, geographic location preferences, failure rate, and other dimensions. Through these criteria, complex vehicle operation data can be divided into subsets that are easier to manage and analyze.
[0047] Cluster analysis is performed on the selected groupings, and the vehicles are grouped according to their similarities or differences, so that the objects within the same group are as similar as possible, while the objects between different groups are as different as possible.
[0048] Kmeans++ clustering algorithm steps:
[0049] 1. Randomly select a point from the data set as the first cluster center.
[0050] 2. Calculate the shortest distance D(x) between each sample and the current cluster center, that is, the distance to the nearest cluster center. The larger the value, the greater the probability of being selected as the cluster center.
[0051] 3. Calculate the probability of each sample being selected as the next cluster center.
[0052]
[0053] Use the roulette wheel method to randomly select a center point as the next center point, and repeat steps 2, 3, and 4 until all center points are found.
[0054] The Kmeans++ algorithm divides data points into different clusters based on the similarity of the data and the number of clusters K. Each cluster represents a group of vehicles with similar characteristics, which may include similar driving patterns, energy consumption levels, fault tendencies, etc. Clustering forms vehicle portraits, that is, summarizing and describing the characteristics of different types of vehicles, thereby providing strong data support for vehicle management, maintenance optimization, market segmentation, etc.
[0055] In the second embodiment, a vehicle similarity network is constructed by projecting a vehicle-user bipartite graph.
[0056] S1. Data collection and cleaning: Perform data processing operations on vehicle-user interaction behavior data, user information, and vehicle information.
[0057] S2. Construct a vehicle-user bipartite graph:
[0058] Vehicle users and are regarded as two independent sets, and the user's positive or negative evaluation of the vehicle is represented by an edge connecting the two sets. The sign of the edge directly corresponds to the type of evaluation, clearly showing the interactive relationship between users and vehicles.
[0059] The interaction relationship between vehicles and users is represented by a symbolic bipartite network G(C, U, E), where C = {C1, C2, ..., C n ,} is the vehicle set, U={U1,U2,...,U n ,} is the user set, E is the edge set linking the vehicle node and the user node, E∈{1,-1} is the edge symbol, which represents the binary preference relationship of like and dislike. The preference relationship in the bipartite network is represented by the n×m dimensional adjacency matrix A=a i Indicates that when the vehicle C recommended to user U makes a positive evaluation, then e ij =1, if it is a negative evaluation, then e ij = -1, if user U has never used vehicle C, then e ij =0.
[0060] S3. Obtain vehicle similarity network through vehicle-user bipartite graph:
[0061] Based on the vehicle-user bipartite graph projection method, the vehicle set in the bipartite graph is mapped to a new space, and the similarity between vehicles is quantified by counting and analyzing the users' positive and negative common evaluations of each vehicle.
[0062] Through the similarity score, a vehicle similarity network is constructed. Each vehicle is a node, and the edges between nodes represent the similarity between vehicles. The weight of the edge is a direct reflection of the similarity score. The larger the weight, the more similar the two vehicles are in user evaluation. Through the vehicle similarity network, you can intuitively see which vehicles have a high similarity in the minds of users.
[0063] Using the bipartite graph link prediction method, the new link prediction task is achieved by projecting the two-mode network into the single-mode network. The vehicle similarity network is obtained through the vehicle-user bipartite graph to achieve subsequent personalized recommendations. The projected vehicle similarity network obtains the similarity relationship of the interested vehicles from the user's actual evaluation behavior. Vehicle C j The connection w between them is the network edge weight, which reflects the similarity between the vehicles of interest.
[0064] Recommend using the similarity of homogeneous nodes, w is node C i and C j The edge weight between them, that is, C i and C j The calculation formula of w is as follows.
[0065]
[0066] The denominator (U i ∪U j) is C i and C j The sum of the number of vehicles used by users, the sum of the numerators C i and C j The number of users who have the same vehicle rating.
[0067] Embodiment 3: Optimizing the vehicle similarity network through a similarity attenuation factor.
[0068] S1. Adjust the vehicle similarity network:
[0069] The vehicle similarity network of Example 2 is based on the actual evaluation behavior of users. The similarity network constructed solely based on the evaluation behavior of users may be affected by the one-sidedness and subjectivity of the evaluation and the limitations of the evaluator himself. In order to make up for this deficiency, the vehicle group information divided by the clustering algorithm is introduced. The clustering algorithm classifies vehicles with similar attributes into the same group by analyzing various attributes of the vehicle (such as engine performance, fuel consumption, safety performance, internal configuration, etc.). The similarity division based on attributes provides a more objective and comprehensive perspective for the vehicle similarity network, which helps to identify vehicles that are highly similar in actual attributes even if they are not directly reflected in the user evaluation.
[0070] By combining the advantages of user evaluation behavior and vehicle attribute similarity, a similarity attenuation factor is introduced. The connection strength between each vehicle node in the vehicle similarity network is dynamically adjusted. By continuously adjusting and optimizing the action mechanism of the similarity attenuation factor, a comprehensive vehicle similarity network that reflects the real preferences of users and takes into account the objective attributes of vehicles can be formed.
[0071] The vehicle similarity network and vehicle portrait are integrated, the attenuation factor γ is defined, and the vehicle similarity network is optimized.
[0072]
[0073] where γ c ∈(0,1), c ij =1 means that vehicles i and j belong to the same vehicle group, c ij = 0 means that vehicle i and vehicle j do not belong to the same vehicle group. When two vehicles belong to the same vehicle group, the similarity between the vehicles does not change, otherwise it decays to the original γ c Times. When γ c = 0, the similarity between heterogeneous vehicles is completely attenuated, and the recommendation algorithm will only recommend vehicles in the same vehicle group. c =1, the similarity between heterogeneous vehicles is not attenuated, that is, the vehicle portrait is not considered, and only the interaction behavior data between the vehicle and the user is used for recommendation.
[0074] S2. Personalized recommendation:
[0075] Through the optimized vehicle similarity network, personalized recommendations are made for users. First, the vehicle that the user has given the highest rating most recently is identified, which usually reflects the user's strongest current interests and preferences. Through the adjusted vehicle similarity network, vehicles that are similar to the vehicle with the highest rating given by the user but with which the user has not yet established a connection are found in the vehicle-user bipartite graph. By calculating the link probability between these potential vehicles and the user and generating the corresponding recommendation score, an ordered list of vehicle recommendations is obtained.
[0076] In the process of generating the recommendation list, the personalized needs and preferences of users are fully considered. Not only are they sorted according to the recommendation scores, but they are also filtered and fine-tuned based on the user's specific conditions. The recommendation list finally presented to the user is a personalized recommendation result that not only meets the user's interests but also meets their actual needs.
[0077] In order to verify the effectiveness and accuracy of the recommendation algorithm, a series of appropriate evaluation indicators are used to evaluate the recommendation results. These evaluation indicators include accuracy, recall, F1 score and other recommendation system evaluation indicators. Through continuous evaluation and optimization, the performance and effect of the recommendation algorithm are continuously improved to provide users with more high-quality and personalized vehicle recommendation services.
[0078] In summary, the present invention first performs a preliminary screening based on the user's travel needs, and then is divided into two parts. The first part: clustering the preprocessed vehicle data using the Kmeans++ algorithm, and the second part: using the user-vehicle interaction behavior, user information, and vehicle information to construct a vehicle-user bipartite graph, and then using the vehicle-user based bipartite graph projection method to construct a vehicle similarity network, and finally according to the clustering results, the vehicle similarity network is adjusted by the similarity attenuation factor to complete the recommendation of the personalized list. This invention not only takes into account the user's historical behavior patterns in selecting vehicles and the preference characteristics for vehicles, but also adds real-time data analysis and prediction of vehicles, and realizes accurate prediction and personalized recommendation of user selected vehicles. The present invention not only provides enterprises with efficient vehicle management tools to help them accurately manage vehicle resources and optimize operating costs, but also provides users who use vehicles with a convenient vehicle selection solution, improves the user's work efficiency and travel experience, and realizes the accurate docking and efficient matching of vehicle resources and user needs.
[0079] In the description of this specification, reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in the embodiment or example of the invention.
[0080] Although embodiments of the invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the claims and their equivalents.
Claims
1. A vehicle recommendation method based on vehicle portrait, characterized in that: The following steps are involved: S1. Data collection and cleaning: delete unnecessary empty records, handle outliers, fill missing data, remove duplicate fields, handle erroneous values, unify data formats, and standardize processing; S2. Perform a preliminary screening based on the user's travel needs, destination, departure time, vehicle availability and travel mode, and select the vehicles that the user can use; S3. Build a vehicle profile, determine one or more grouping criteria for the real-time operation data of the vehicle according to business requirements and data characteristics, group the data based on the selected characteristics, perform cluster analysis on the selected groups, and divide the data into different clusters; S4. Construct a vehicle similarity network and present the positive and negative evaluation information collected by users after using the vehicle in a bipartite graph; users and vehicles are regarded as two independent sets, and the positive or negative evaluation of the user on the vehicle is represented by the edge connecting the two sets. The sign of the edge directly corresponds to the type of evaluation, which clearly shows the interactive relationship between users and vehicles; S5 adjusts the vehicle similarity network; S6. Realize personalized recommendations, optimize the vehicle similarity network based on vehicle portraits, and make recommendations based on the adjusted vehicle similarity network.
2. The vehicle recommendation method based on vehicle portrait according to claim 1, characterized in that: In step S3, the vehicle data needs to be input into the Kmeans++ clustering algorithm to perform a clustering process. The algorithm will divide the data points into different clusters and generate vehicle portraits in parallel based on the similarity of the data and the set parameters.
3. The vehicle recommendation method based on vehicle portrait according to claim 1, characterized in that: In step S4, the vehicle set in the bipartite graph needs to be mapped to a new space based on the vehicle-user bipartite graph projection method, and the similarity between vehicles is quantified by counting and analyzing the positive and negative common evaluations of users on each vehicle; The similarity between vehicle nodes is calculated by comprehensively evaluating the number of users who have made common evaluations, the proportion of common positive evaluations, and the impact of common negative evaluations; Through the similarity score, a vehicle similarity network is built; the edge weight is a direct reflection of the similarity score. Through the similarity network, we can intuitively see which vehicles have a higher similarity in the minds of users.
4. The vehicle recommendation method based on vehicle portrait according to claim 1, characterized in that: In step S5, the vehicle similarity network is adjusted using a similarity attenuation factor according to the vehicle similarity formed by the users' preferences for vehicles and the similarity of the vehicles at the attribute level, so as to obtain a final vehicle similarity network.
5. The vehicle recommendation method based on vehicle portrait according to claim 1, characterized in that: In step S6, the vehicle with the highest score given by the user most recently needs to be obtained, and then the possibility of generating links between heterogeneous nodes without edges in the vehicle-user bipartite graph is calculated based on the adjusted vehicle similarity network, and the recommendation scores are sorted in descending order, and the first L vehicles are taken from the sorted vehicle list to form a recommendation list for recommendation, and finally the recommendation algorithm is evaluated using appropriate evaluation indicators.