Vehicle recommendation method, device, computer equipment and storage medium

Through the vector characterization model, the vehicle information and user historical traffic data are processed, the vehicle feature vectors and intention labels are generated, and the similarity is calculated for sorting and recommendations are sorted, which solves the problem of low vehicle recommendation accuracy and improves the exposure and marketing effect of long-tail car products.

CN112561663BActive Publication Date: 2024-11-15HANGZHOU SOUCHE DATA TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202011573596.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-12-24
Publication Date
2024-11-15
Estimated Expiration
2040-12-24

AI Technical Summary

Technical Problem

The accuracy of vehicle recommendations in the prior art is low, and the user's long tail preference cannot be effectively identified, resulting in too narrow recommendation lists and insufficient exposure to other models.

Method used

By obtaining vehicle information and user historical traffic data, a vector characterization model is used to generate vehicle feature vectors and intention labels, combined with neural network models to process appearance and click behavior sequences, and calculate vehicle similarity for sorting and recommendations.

Benefits of technology

It improves the accuracy of vehicle recommendations, promotes the marketing and exposure of long-tail car products, improves the overall marketing effect, and solves the disadvantage of the original model that can only be recommended for users to interact with.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN112561663B_ABST
    Figure CN112561663B_ABST
Patent Text Reader

Abstract

The present application relates to a vehicle recommendation method, device, computer equipment and storage medium, wherein the vehicle recommendation method includes: obtaining first vehicle information and user historical traffic information; processing the first vehicle information and the historical traffic information through a preset vector representation model to obtain a first feature vector of the first vehicle; reading the stored second feature vector, and sorting the first vehicle based on the first feature vector and the second feature vector, and determining the recommended vehicle according to the sorting result. The above-mentioned vehicle recommendation method, device, computer equipment and storage medium, through a neural network, convert the characteristics of multiple modes of the vehicle into vector statements representing similarity, and compare them with the user's purchased model to make recommendations, which solves the shortcomings and deficiencies of the original marketing process that only the models that the user has interacted with can be recommended, improves the overall marketing effect, and improves the accuracy of vehicle recommendations.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of deep learning, and in particular to a vehicle recommendation method, apparatus, computer device and storage medium. Background Art

[0002] With the booming development of e-commerce, online consumption accounts for an increasingly higher proportion of the total retail sales of social goods. The huge shopping advantages that online shopping provides to consumers are mainly reflected in breaking through time and space restrictions, shopping convenience, more product choices, competitive prices, rich product information, personalization and customization. Among them, using user behavior data to improve the efficiency of e-commerce operations and advertising delivery has attracted the attention of more and more e-commerce companies and advertisers.

[0003] In the automotive B2B marketing scenario, DMP (Data Management Platform) generates user preferences for car models and brands based on user browsing and clicking on vehicles as a portrait of car dealers, helping the platform to conduct personalized marketing for car dealers. In the marketing process, when SMS or push marketing is carried out, in order to promote high-intention models, the following tags are often used: recently sold models, recently browsed models, and recently clicked models. However, these tags are not flexible enough. Recent purchases and browsing do not necessarily represent future preferences. At the same time, these tags are too simple and crude, and users' long-tail preferences are ignored in this process. For example, vehicles that users have not interacted with are not because they have no intention to buy them, but are just due to temporary market conditions. In the end, the user's recommendation list becomes narrower and narrower, and other models on the B2B platform are not exposed enough.

[0004] Currently, no effective solution has been proposed for the problem of low vehicle recommendation accuracy in related technologies. Summary of the invention

[0005] Embodiments of the present application provide a vehicle recommendation method, apparatus, computer device, and storage medium to at least solve the problem of low vehicle recommendation accuracy in the related art.

[0006] In a first aspect, an embodiment of the present application provides a vehicle recommendation method, comprising:

[0007] Acquire first vehicle information of a first vehicle to be recommended and historical traffic information generated by a user browsing and clicking on second vehicle information, wherein the first vehicle information includes model information and appearance information of the first vehicle;

[0008] Processing the first vehicle information and the historical traffic information through a preset vector representation model to obtain a first feature vector of the first vehicle;

[0009] Read the stored second feature vector, and sort the first vehicle based on the first feature vector and the second feature vector, and determine the recommended vehicle according to the sorting result, wherein the second feature vector is generated by processing the historical transaction information of the second vehicle by the vector characterization model, and is used to characterize the user's intention label for the second vehicle.

[0010] In some embodiments, obtaining first vehicle information of a first vehicle to be recommended includes:

[0011] Acquire at least one of the following information about the model, series, and brand of the first vehicle as the model information;

[0012] acquiring a vehicle image of the first vehicle;

[0013] The vehicle image is input into a trained neural network model to obtain appearance information of the first vehicle, wherein the appearance information includes an appearance vector representing the similarity of vehicle appearances.

[0014] In some embodiments, the first vehicle information and the historical traffic information are processed by a preset vector representation model to obtain a first feature vector of the first vehicle, including:

[0015] Generate a user click behavior sequence based on the historical traffic information;

[0016] The first vehicle information and the user click behavior sequence are processed by a preset vector representation model to obtain a first feature vector of the first vehicle.

[0017] In some embodiments, generating a user click behavior sequence based on the historical traffic information includes:

[0018] Constructing a directed unweighted graph based on the historical traffic information, wherein the directed unweighted graph includes vehicle information and a vehicle click sequence;

[0019] An arbitrary starting node is selected in the directed unweighted graph, and a user click behavior sequence is generated by a random walk.

[0020] In some embodiments, the first vehicle information and the user click behavior sequence are processed by a preset vector representation model to obtain a first feature vector of the first vehicle, including:

[0021] The first vehicle information is used as input, and the adjacent vehicles of the first vehicle in the user click behavior sequence are used as output, the vector representation model is trained to obtain a hidden layer vector, and the hidden layer vector is used as a first feature vector of the first vehicle.

[0022] In some embodiments, before reading the stored second feature vector, the method further includes:

[0023] The vector representation model is trained based on the user click behavior sequence and the historical transaction information of the second vehicle to obtain a hidden layer vector of the second vehicle;

[0024] An average value of the hidden layer vector of the second vehicle is obtained based on the hidden layer vector of the second vehicle to obtain the second feature vector.

[0025] In some embodiments, the sorting of the first vehicles based on the first feature vector and the second feature vector, and determining a recommended vehicle according to the sorting result comprises:

[0026] Obtaining the cosine similarity between the first feature vector and the second feature vector;

[0027] sorting the first vehicles based on the cosine similarity;

[0028] A preset number of first vehicles are selected for recommendation based on the ranking from high to low.

[0029] In a second aspect, an embodiment of the present application provides a vehicle recommendation device, comprising:

[0030] An acquisition module, used to acquire first vehicle information of a first vehicle to be recommended and historical traffic information generated by a user browsing and clicking on second vehicle information, wherein the first vehicle information includes model information and appearance information of the first vehicle;

[0031] A first processing module, configured to process the first vehicle information and the historical traffic information through a preset vector representation model to obtain a first feature vector of the first vehicle;

[0032] A recommendation module is used to read the stored second feature vector, sort the first vehicle based on the first feature vector and the second feature vector, and determine the recommended vehicle according to the sorting result, wherein the second feature vector is generated by processing the historical transaction information of the second vehicle by the vector characterization model, and is used to characterize the user's intention label for the second vehicle.

[0033] In a third aspect, an embodiment of the present application provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the vehicle recommendation method as described in the first aspect above is implemented.

[0034] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the vehicle recommendation method as described in the first aspect above.

[0035] Compared with the related art, the vehicle recommendation method, apparatus, computer equipment and storage medium provided in the embodiments of the present application obtain the first vehicle information of the first vehicle to be recommended and the historical traffic information generated by the user browsing and clicking on the second vehicle information, wherein the first vehicle information includes the model information and appearance information of the first vehicle; the first vehicle information and the historical traffic information are processed by a preset vector representation model to obtain the first feature vector of the first vehicle; the stored second feature vector is read, and the first vehicle is sorted based on the first feature vector and the second feature vector, and the recommended vehicle is determined according to the sorting result, wherein the second feature vector is generated by the vector representation model through processing the historical transaction information of the second vehicle, and is used to represent the user's intention label for the second vehicle, which solves the shortcomings and deficiencies of low vehicle recommendation accuracy in the related art and improves the accuracy of vehicle recommendation.

[0036] Details of one or more embodiments of the present application are set forth in the following drawings and description to make other features, objects, and advantages of the present application more readily apparent. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0038] Figure 1 is a flow chart of a vehicle recommendation method according to an embodiment of the present application;

[0039] Figure 2 is a structural diagram of a trained neural network model of a vehicle recommendation method according to an embodiment of the present application;

[0040] Figure 3 is a schematic diagram of a directed unweighted graph of a vehicle recommendation method according to an embodiment of the present application;

[0041] Figure 4 is a structural diagram of a vector representation model of a vehicle recommendation method according to an embodiment of the present application;

[0042] Figure 5 is a flow chart of a vehicle recommendation method according to a preferred embodiment of the present application;

[0043] Figure 6 is a structural block diagram of a vehicle recommendation device according to an embodiment of the present application;

[0044] Figure 7 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. Based on the embodiments provided in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0046] Obviously, the drawings described below are only some examples or embodiments of the present application. For ordinary technicians in this field, the present application can also be applied to other similar scenarios based on these drawings without creative work. In addition, it can also be understood that although the efforts made in this development process may be complicated and lengthy, for ordinary technicians in this field related to the content disclosed in this application, some changes in design, manufacturing or production based on the technical content disclosed in this application are just conventional technical means, and should not be understood as insufficient content disclosed in this application.

[0047] Reference to "embodiments" in this application means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those of ordinary skill in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0048] Unless otherwise defined, the technical terms or scientific terms involved in this application should be understood by people with ordinary skills in the technical field to which this application belongs. The words "one", "a", "a", "the" and the like involved in this application do not indicate a quantitative limitation, and may represent the singular or plural. The terms "include", "comprise", "have" and any of their variations involved in this application are intended to cover non-exclusive inclusions; for example, a process, method, system, product or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units that are not listed, or may also include other steps or units inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The "multiple" involved in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that there may be three relationships, for example, "A and / or B" can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects before and after are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific ordering of the objects.

[0049] Before describing and illustrating the embodiments of the present application, the related technologies used in the present application are described as follows: the vector representation model is a neural network model using the word2vec algorithm. The neural network model using the word2vec algorithm is a group of related models used to generate word vectors. These models are shallow and two-layer neural networks used for training to reconstruct linguistic word texts. The network is represented by words, and it is necessary to guess the input words in adjacent positions. Under the assumption of the bag of words model in word2vec, the order of words is not important. After training, the word2vec model can be used to map each word to a vector, which can be used to represent the relationship between words. The vector is the hidden layer of the neural network.

[0050] Figure 1 FIG. 1 is a flow chart of a vehicle recommendation method according to an embodiment of the present application. Figure 1 As shown, the process includes:

[0051] Step S101, obtaining first vehicle information of a first vehicle to be recommended and historical traffic information generated by a user browsing and clicking on second vehicle information, wherein the first vehicle information includes vehicle model information and appearance information of the first vehicle.

[0052] Exemplarily, the user's historical traffic information is the interactive data generated when the user browses and clicks on the recommended vehicles within a preset time, such as the model, series, and brand of the vehicle that the user clicks to view, the order in which the user clicks to view the vehicles, etc. It can be understood that the user's historical traffic information can be obtained through the embedding point on the user side. In this embodiment, the first vehicle is the vehicle that the car dealer needs to push marketing, and its specific range can be determined according to the actual situation and is not limited here.

[0053] Step S102: Process the first vehicle information and the historical traffic information using a preset vector representation model to obtain a first feature vector of the first vehicle.

[0054] In this embodiment, a neural network model of the word2vec algorithm can be used as a vector representation model, and the first vehicle information is used as input, and the user's historical traffic information is used as output for joint training to obtain a first feature vector of the first vehicle. In other embodiments, other neural network models can also be used for training to obtain the first feature vector, which is not specifically limited here.

[0055] Step S103, read the stored second feature vector, and sort the first vehicle based on the first feature vector and the second feature vector, and determine the recommended vehicle according to the sorting result, wherein the second feature vector is generated by processing the historical transaction information of the second vehicle by the vector characterization model, and is used to characterize the user's intention label for the second vehicle.

[0056] Exemplarily, the historical transaction information of the second vehicle is the data of the vehicle purchased by the user within a preset time, including the model information and appearance information of the purchased vehicle. The historical transaction information is used as input, and the historical traffic information of the user is used as output. The neural network model that also uses the word2vec algorithm is jointly trained to obtain the second feature vector of the second vehicle. It can be understood that the neural network model used in this step only needs to be consistent with the neural network model used in step S102.

[0057] It can be understood that the first feature vector represents the comprehensive characteristics of the first vehicle, and the second feature vector represents the comprehensive characteristics of the second vehicle. By comparing the first feature vector with the second feature vector, it can be determined which vehicles in the first vehicle are closer to the user's purchasing needs and preferences. After quantification, they are sorted and the top-ranked vehicles are recommended to the user.

[0058] Through the above steps S101 to S103, the first vehicle information of the first vehicle to be recommended and the historical traffic information generated by the user browsing and clicking on the second vehicle information are obtained, wherein the first vehicle information includes the model information and appearance information of the first vehicle; the first vehicle information and the historical traffic information are processed by the preset vector representation model to obtain the first feature vector of the first vehicle; the stored second feature vector is read, and the first vehicle is sorted based on the first feature vector and the second feature vector, and the recommended vehicle is determined according to the sorting result, wherein the second feature vector is generated by the vector representation model by processing the historical transaction information of the second vehicle, and is used to represent the user's intention label for the second vehicle. In this way, through the vector representation model, the features of multiple modes of the vehicle are converted into vector statements representing similarity, and compared with the user's purchased model to make recommendations, rather than relying on the model for recommendation, the comparison is more comprehensive, which solves the shortcomings and deficiencies of the original marketing process that only the models that the user has interacted with can be recommended, promotes the marketing and exposure of long-tail automobile products, improves the overall marketing effect, and improves the accuracy of vehicle recommendations.

[0059] In some embodiments, obtaining first vehicle information of a first vehicle to be recommended includes the following steps:

[0060] Step 1, obtaining at least one of the model, series and brand of the first vehicle as model information.

[0061] Step 2, obtaining a vehicle image of the first vehicle; inputting the vehicle image into a trained neural network model to obtain appearance information of the first vehicle, wherein the appearance information includes an appearance vector representing the similarity of the vehicle appearance.

[0062] This embodiment obtains at least one of the information of vehicle model, vehicle series and brand as vehicle model information. In the case that the vehicle has not been sold, the vehicle series or brand can be used to replace the vehicle model features and spliced ​​with the image vector to solve the cold start problem of the product. In addition, the vehicle image is preprocessed first to obtain the appearance vector representing the similarity of the vehicle appearance as the appearance information, so that the vehicle recommendation is more accurate.

[0063] In this embodiment, the car pictures on the platform are annotated according to the model of the first vehicle. First, the banner image of the car display page is standardized through the car detection model, the foreground and the background are separated, and the rear view picture of the car is used as the vehicle image and input into the trained neural network model to obtain the picture vector. In this embodiment, the resnet-34 deep learning model can be used as the neural network model for training. Please refer to Figure 2 , Figure 2: is a structural diagram of a trained neural network model of a vehicle recommendation method according to an embodiment of the present application. In the case of the traditional softmax loss, center loss is added as a penalty term in the neural network model of this embodiment so that the distance between the image vectors of vehicles with similar appearance features is close; the distance between the image vectors of vehicles with different appearance features is far. Specifically, the loss function is as follows:

[0064]

[0065] Among them, L S represents the traditional softmax loss function, L C Represents the loss caused by the distance from each sample point to the class center, c yi Represents y i The vector of the class center of the category uses the mean of the vector as the class center, and λ represents the weight of the impact of centerloss loss on the entire loss function.

[0066] The 512-dimensional vector finally generated is used as the vehicle appearance information. In other embodiments, other deep learning models can be used to process vehicle images, as long as the image vector representing the vehicle similarity can be obtained.

[0067] It can be understood that the appearance information of the first vehicle may be a vehicle picture, or may be a picture vector obtained by processing based on the vehicle picture.

[0068] In some of the embodiments, processing the first vehicle information and the historical traffic information by a preset vector representation model to obtain a first feature vector of the first vehicle includes the following steps:

[0069] Step 1: Generate user click behavior sequence based on historical traffic information.

[0070] Step 2: Process the first vehicle information and the user click behavior sequence through a preset vector representation model to obtain a first feature vector of the first vehicle.

[0071] Exemplarily, based on the user's historical traffic information, that is, the interaction data generated when the user browses and clicks on the recommended vehicles within a preset time, such as the model, series, and brand of the vehicle that the user clicks to view, and the order in which the user clicks to view the vehicles, etc., multiple user click behavior sequences can be obtained, that is, a sequence formed with the vehicle model information as the node and the user's click order as the connecting line, which is used to characterize the user's preference.

[0072] In some of the embodiments, generating a user click behavior sequence based on historical traffic information includes the following steps:

[0073] Step 1: construct a directed unweighted graph based on the user's historical traffic information. The directed unweighted graph includes vehicle information and vehicle click sequence.

[0074] Step 2: Select any starting node in the directed unweighted graph and use random walk to generate the user click behavior sequence.

[0075] This embodiment generates a user click behavior sequence by establishing a directed unweighted graph and performing a random walk. The obtained user click behavior sequence reflects more comprehensive data, has a larger data volume, has a better effect on neural network training, and makes vehicle recommendations more accurate.

[0076] See also Figure 3 , Figure 3 is a schematic diagram of a directed unweighted graph of a vehicle recommendation method according to an embodiment of the present application. For example, a directed unweighted graph is defined as G = (V, ε), where V and ε represent the set of points and edges respectively. The purpose of graph vectorization is to learn a graph vector representation model Φ: V → R d , convert each node into a d-dimensional vector representation. The method for constructing a directed unweighted graph is as follows:

[0077] Get the user's click sequence on the vehicle, where the click sequence is truncated by day, and you will get each user's daily click sequence [car1, car2, car3, ...].

[0078] Each vehicle is embedded in the graph as a node V. The click sequence is traversed. When car1 is clicked and car2 is clicked, node 1 and node 2 are connected, and the direction is set to node 1 to node 2.

[0079] Traverse the click sequences of all users and construct a directed unweighted graph with click sequences as edges and vehicles as nodes.

[0080] For example, after the directed unweighted graph is constructed, any starting node is selected for random walk to generate multiple user click behavior sequences as training sets. Specifically, the random walk process is as follows:

[0081] Input parameters: graph G = (V, ε), number of iterations γ, length of each path t

[0082] Output: path collection Paths

[0083] for i=0in range(γ):

[0084] Arrange the vertices randomly, O

[0085] Initialize the path collection Paths Initialize a single path Path

[0086] for each vi∈O do:

[0087]

[0088] for k=0in range(t):

[0089] From v i Randomly pick a neighboring node and put it into the path. Path.append(random(neighbor(v i )))

[0090] node=random(neighbor(v i )

[0091] end for

[0092] Paths.append(Path)

[0093] end for

[0094] end for

[0095] Exemplarily, taking any node as the starting point and t as the path length, a random walk is performed on the directed unweighted graph. When the number of walked nodes reaches t, a user click behavior sequence is obtained. Then, any node is reselected for iteration, and the number of iterations is γ, and finally γ user click behavior sequences are obtained.

[0096] In some of the embodiments, the first vehicle information and the user click behavior sequence are processed by a preset vector representation model to obtain a first feature vector of the first vehicle, which includes the following steps: taking the first vehicle information as input, taking the adjacent vehicles of the first vehicle in the user click behavior sequence as output, training the vector representation model to obtain a hidden layer vector, and using the hidden layer vector as the first feature vector of the first vehicle.

[0097] This embodiment integrates various features of a vehicle through a vector representation model to obtain a feature vector that comprehensively represents the features of the vehicle, so that the feature expression is more accurate and the accuracy of vehicle recommendation is higher.

[0098] See also Figure 4 , Figure 4 It is a structural diagram of a vector representation model of a vehicle recommendation method according to an embodiment of the present application.

[0099] The vehicle model information includes vehicle features, brand features and series features, and the picture features are the appearance information.

[0100] In this embodiment, the input of the vector representation model is the first vehicle information of the first vehicle, that is, the vehicle model information and appearance information. Specifically, the vehicle model information can be the brand, series or specific model of the vehicle, or all three and the appearance information can be used as input at the same time; the output is the two adjacent vehicles of the first vehicle used as input in the user click behavior sequence, that is, the upstream vehicle and the downstream vehicle of the first vehicle in the user click behavior sequence.

[0101] By using the vehicle information, appearance information and user click behavior sequence of the first vehicle as a training set and training the vector representation model, a hidden layer vector corresponding to each input can be obtained, which can be used to represent the relationship between words, namely, the first feature vector.

[0102] In other embodiments, other neural network models may be used as long as they can obtain a feature vector based on the input feature synthesis.

[0103] In some of the embodiments, before reading the stored second feature vector, the following steps are also included:

[0104] Step 1: Train the vector representation model based on the user click behavior sequence and the historical transaction information of the second vehicle to obtain the hidden layer vector of the second vehicle.

[0105] Step 2: Obtain an average value of the hidden layer vector of the second vehicle based on the hidden layer vector of the second vehicle to obtain a second eigenvector.

[0106] This embodiment uses the same vector conversion method as the first vehicle for the vehicles in the user's historical transaction information to obtain a feature vector that represents the comprehensive characteristics of the vehicle, which is more closely matched when the features are subsequently compared with the first vehicle, thereby making the vehicle recommendation more accurate.

[0107] Exemplarily, the same vector conversion method is used for the second vehicle. First, the historical transaction information of the second vehicle, that is, the vehicle model information and appearance information, is obtained. Then, the brand, car series or model and appearance information are taken as input, and the user click behavior sequence is taken as output. The vector representation model is trained to obtain the hidden layer vector corresponding to each input second vehicle, and the hidden layer vectors of all second vehicles are averaged to obtain the second feature vector.

[0108] It can be understood that for each vehicle, it can be represented by extracting continuous embedded features. For example, the vehicle model, series, brand and appearance information are all 512-dimensional representation vectors. After weighted splicing through a neural network, a 1024-dimensional, 1536-dimensional or 2048-dimensional hidden layer vector is obtained to comprehensively represent the characteristics of the vehicle.

[0109] In some of the embodiments, sorting the first vehicle based on the first feature vector and the second feature vector, and determining the recommended vehicle according to the sorting result comprises the following steps:

[0110] Step 1: Obtain the cosine similarity between the first eigenvector and the second eigenvector.

[0111] Step 2: sort the first vehicles based on cosine similarity.

[0112] Step 3, selecting a preset number of first vehicles for recommendation based on the ranking from high to low.

[0113] This embodiment compares the similarity between the first vehicle and the second vehicle by calculating the cosine similarity between the first feature vector and the second feature vector, and performs quantitative comparison on the feature vectors, which is more intuitive, convenient for sorting and recommendation, and has higher recommendation accuracy.

[0114] For example, when operating B-end manufacturer marketing, a recommendation list is constructed based on historical purchases. According to the recommendation list, the latent vector of each second vehicle is obtained, and the central vector is obtained after averaging. Then, the latent vector of each first vehicle is calculated, and the cosine similarity between the latent vector of the first vehicle and the central vector is calculated. The first vehicles are sorted by similarity, and the top-N vehicles are selected to be put into the marketing list to increase the robustness and long-tail effect of the recommendation list. Specifically, the number of vehicles selected for recommendation can be set according to actual conditions.

[0115] In some of the embodiments, the input of the training set of the first neural network model may be one or more of the model, brand, or series and appearance information, and one or more of the model, brand, or series and appearance information may serve as multimodal features. When the model of the first vehicle is a new model, it has not accumulated a certain amount of browsing clicks, and it is impossible to obtain a feature vector based on the model features. At this time, we can use the brand or series features instead of the model features to form a hidden layer vector, and concatenate it with the appearance information to obtain a 1024-dimensional vector as the feature vector, which solves the cold start problem of the product very well.

[0116] See also Figure 5 , Figure 5 FIG. 1 is a flow chart of a method for obtaining a feature vector according to a preferred embodiment of the present application. Figure 5 As shown, the process includes:

[0117] Step S501, extracting vehicle appearance information.

[0118] Exemplarily, vehicle appearance information may be extracted based on the vehicle image, and the appearance information represents the vehicle appearance similarity.

[0119] Step S502, obtaining user historical traffic information.

[0120] Exemplarily, the user's historical traffic information is the interaction data generated when the user browses and clicks on the recommended vehicles within a preset time, such as the model, series, and brand of the vehicle that the user clicks to view, and the order in which the user clicks to view the vehicles.

[0121] Step S503, constructing a directed unweighted graph.

[0122] Exemplarily, based on the historical traffic information of users, the click sequences of all users are traversed to construct a directed unweighted graph with the click sequences as edges and the vehicles as nodes.

[0123] Step S504: Random walk to generate a user click behavior sequence.

[0124] Exemplarily, any starting node is selected in a directed unweighted graph, and a user click behavior sequence is generated by a random walk.

[0125] Step S505, input vehicle model information and appearance information.

[0126] Exemplarily, the vehicle model information and appearance information are input into the vector representation model.

[0127] Step S506: training a vector representation model.

[0128] Exemplarily, the vehicle model information and appearance information are used as input, and the upstream vehicle and downstream vehicle of the first vehicle in the user click behavior sequence are used as output to train the vector representation model.

[0129] Step S507, generating a feature vector.

[0130] Exemplarily, a hidden layer vector in the vector representation model is obtained as a feature vector.

[0131] This embodiment integrates various features of a vehicle through a vector representation model to obtain a feature vector that comprehensively represents the features of the vehicle, so that the feature expression is more accurate and the accuracy of vehicle recommendation is higher.

[0132] It should be noted that the steps shown in the above process or the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0133] This embodiment also provides a vehicle recommendation device, which is used to implement the above embodiments and preferred implementations, and will not be repeated here. As used below, the terms "module", "unit", "subunit", etc. can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, it is also possible and conceivable to implement it in hardware, or a combination of software and hardware.

[0134] Figure 6 is a structural block diagram of a vehicle recommendation device according to an embodiment of the present application, such as Figure 6 As shown, the device comprises:

[0135] The acquisition module 10 is used to obtain the first vehicle information of the first vehicle to be recommended and the historical traffic information generated by the user browsing and clicking on the second vehicle information, wherein the first vehicle information includes the model information and appearance information of the first vehicle.

[0136] The first processing module 20 is coupled to the acquisition module 10 and is used to process the first vehicle information and the historical traffic information through a preset vector representation model to obtain a first feature vector of the first vehicle.

[0137] The recommendation module 30 is coupled to the first processing module 20, and is used to read the stored second feature vector, and sort the first vehicle based on the first feature vector and the second feature vector, and determine the recommended vehicle according to the sorting result, wherein the second feature vector is generated by processing the historical transaction information of the second vehicle by a vector characterization model, and is used to characterize the user's intention label for the second vehicle.

[0138] In some of the embodiments, the acquisition module 10 is further used to acquire at least one of the model, series and brand of the first vehicle as the model information; acquire a vehicle image of the first vehicle; input the vehicle image into a trained neural network model to obtain the appearance information of the first vehicle, and the appearance information includes an appearance vector representing the similarity of the vehicle appearance. In some of the embodiments, the first processing module 20 is used to generate a user click behavior sequence based on historical traffic information; and process the first vehicle information and the user click behavior sequence through a preset vector representation model to obtain a first feature vector of the first vehicle.

[0139] In some of the embodiments, the first processing module 20 is used to construct a directed unweighted graph based on the user's historical traffic information, wherein the directed unweighted graph includes vehicle information and vehicle click order; an arbitrary starting node is selected in the directed unweighted graph, and a random walk method is used to generate a user click behavior sequence.

[0140] In some of the embodiments, the first processing module 20 is used to take the first vehicle information as input, take the adjacent vehicles of the first vehicle in the user click behavior sequence as output, train the vector representation model to obtain a hidden layer vector, and use the hidden layer vector as the first feature vector of the first vehicle.

[0141] In some of the embodiments, the recommendation module 30 is used to obtain the cosine similarity between the first feature vector and the second feature vector; sort the first vehicles based on the cosine similarity; and select a preset number of first vehicles for recommendation based on the sorting from high to low.

[0142] In some of the embodiments, the vehicle recommendation device is also used to train a vector representation model based on a user click behavior sequence and historical transaction information of a second vehicle to obtain a hidden layer vector of the second vehicle; and obtain an average value of the hidden layer vector of the second vehicle based on the hidden layer vector of the second vehicle to obtain a second feature vector.

[0143] It should be noted that the above modules can be functional modules or program modules, and can be implemented by software or hardware. For modules implemented by hardware, the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.

[0144] In addition, combined Figure 1 The vehicle recommendation method of the embodiment of the present application described above may be implemented by a computer device. Figure 7 Schematic diagram of the hardware structure of a computer device according to an embodiment of the present application.

[0145] The computer device may include a processor 71 and a memory 72 storing computer program instructions.

[0146] Specifically, the processor 71 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0147] Among them, the memory 72 may include a large-capacity memory for data or instructions. By way of example and not limitation, the memory 72 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 72 may include a removable or non-removable (or fixed) medium. Where appropriate, the memory 72 may be inside or outside the data processing device. In a specific embodiment, the memory 72 is a non-volatile memory. In a specific embodiment, the memory 72 includes a read-only memory (ROM) and a random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (Programmable Read-Only Memory, PROM for short), an erasable PROM (Erasable ProgrammableRead-Only Memory, EPROM for short), an electrically erasable PROM (Electrically Erasable ProgrammableRead-Only Memory, EEPROM for short), an electrically alterable ROM (Electrically Alterable Read-Only Memory, EAROM for short) or a flash memory (FLASH) or a combination of two or more of these. Under appropriate circumstances, the RAM can be a static random access memory (SRAM) or a dynamic random access memory (DRAM), wherein the DRAM can be a fast page mode dynamic random access memory (FPMDRAM), an extended data output dynamic random access memory (EDODRAM), a synchronous dynamic random access memory (SDRAM), etc.

[0148] The memory 72 may be used to store or cache various data files required for processing and / or communication, as well as possible computer program instructions executed by the processor 71 .

[0149] The processor 71 implements any one of the vehicle recommendation methods in the above embodiments by reading and executing computer program instructions stored in the memory 72 .

[0150] In some embodiments, the computer device may further include a communication interface 73 and a bus 70. Figure 7 As shown, the processor 71, the memory 72, and the communication interface 73 are connected via a bus 70 and communicate with each other.

[0151] The communication interface 73 is used to implement communication between the modules, devices, units and / or equipment in the embodiment of the present application. The communication interface 73 can also implement data communication with other components such as: external devices, image / data acquisition equipment, databases, external storage, and image / data processing workstations.

[0152] The bus 70 includes hardware, software or both, and couples the components of the computer device to each other. The bus 70 includes but is not limited to at least one of the following: a data bus, an address bus, a control bus, an expansion bus, and a local bus. By way of example and not limitation, bus 70 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses or a combination of two or more of these. Where appropriate, bus 70 may include one or more buses. Although embodiments of the present application describe and illustrate a particular bus, the present application contemplates any suitable bus or interconnect.

[0153] The computer device can execute the vehicle recommendation method in the embodiment of the present application based on the acquired computer program instructions, thereby realizing the combination Figure 1 Describe the vehicle recommendation method.

[0154] In addition, in combination with the vehicle recommendation method in the above embodiments, the present application embodiment can provide a computer-readable storage medium for implementation. The computer-readable storage medium stores computer program instructions; when the computer program instructions are executed by a processor, any one of the vehicle recommendation methods in the above embodiments is implemented.

[0155] The above-mentioned vehicle recommendation method, device, computer equipment and storage medium obtain the first vehicle information of the first vehicle to be recommended and the historical traffic information generated by the user browsing and clicking on the second vehicle information, wherein the first vehicle information includes the model information and appearance information of the first vehicle; the first vehicle information and the historical traffic information are processed by a preset vector representation model to obtain the first feature vector of the first vehicle; the stored second feature vector is read, and the first vehicle is sorted based on the first feature vector and the second feature vector, and the recommended vehicle is determined according to the sorting result, wherein the second feature vector is generated by the vector representation model by processing the historical transaction information of the second vehicle, and is used to represent the user's intention label for the second vehicle. In this way, the features of multiple modes of the vehicle are converted into vector statements representing similarity through the vector representation model, and compared with the user's purchased model to make recommendations, rather than relying on the model for recommendation. The comparison is more comprehensive, which solves the shortcomings and deficiencies of the original marketing process that only the models that the user has interacted with can be recommended, promotes the marketing and exposure of long-tail automobile products, improves the overall marketing effect, and improves the accuracy of vehicle recommendations. At the same time, the car series or brand can be used instead of the car model features and spliced ​​with the image vector, which effectively solves the cold start problem of the product.

[0156] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0157] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A vehicle recommendation method, characterized in that: include: Acquire first vehicle information of a first vehicle to be recommended and historical traffic information generated by a user browsing and clicking on second vehicle information, wherein the first vehicle information includes model information and appearance information of the first vehicle; Processing the first vehicle information and the historical traffic information through a preset vector representation model to obtain a first feature vector of the first vehicle; Reading a stored second feature vector, and sorting the first vehicle based on the first feature vector and the second feature vector, and determining a recommended vehicle according to the sorting result, wherein the second feature vector is generated by processing historical transaction information of the second vehicle by the vector representation model, and is used to represent the user's intention label for the second vehicle; The first vehicle information and the historical traffic information are processed by a preset vector representation model to obtain a first feature vector of the first vehicle, including: Generate a user click behavior sequence based on the historical traffic information; Processing the first vehicle information and the user click behavior sequence through a preset vector representation model to obtain a first feature vector of the first vehicle; The first vehicle information and the user click behavior sequence are processed by a preset vector representation model to obtain a first feature vector of the first vehicle, including: The first vehicle information is used as input, and the adjacent vehicles of the first vehicle in the user click behavior sequence are used as output, the vector representation model is trained to obtain a hidden layer vector, and the hidden layer vector is used as a first feature vector of the first vehicle.

2. The vehicle recommendation method according to claim 1, characterized in that: The obtaining first vehicle information of the first vehicle to be recommended includes: Acquire at least one of the following information about the model, series, and brand of the first vehicle as the model information; acquiring a vehicle image of the first vehicle; The vehicle image is input into a trained neural network model to obtain appearance information of the first vehicle, wherein the appearance information includes an appearance vector representing the similarity of vehicle appearances.

3. The vehicle recommendation method according to claim 1, characterized in that: Generating a user click behavior sequence based on the historical traffic information includes: Constructing a directed unweighted graph based on the historical traffic information, wherein the directed unweighted graph includes vehicle information and a vehicle click sequence; An arbitrary starting node is selected in the directed unweighted graph, and a user click behavior sequence is generated by a random walk.

4. The vehicle recommendation method according to claim 1, characterized in that: Before reading the stored second feature vector, the method further includes: The vector representation model is trained based on the user click behavior sequence and the historical transaction information of the second vehicle to obtain a hidden layer vector of the second vehicle; An average value of the hidden layer vector of the second vehicle is obtained based on the hidden layer vector of the second vehicle to obtain the second feature vector.

5. The vehicle recommendation method according to claim 1, characterized in that: The step of sorting the first vehicles based on the first feature vector and the second feature vector, and determining a recommended vehicle according to the sorting result includes: Obtaining the cosine similarity between the first feature vector and the second feature vector; sorting the first vehicles based on the cosine similarity; A preset number of first vehicles are selected for recommendation based on the ranking from high to low.

6. A vehicle recommendation device, characterized in that: include: An acquisition module, used to acquire first vehicle information of a first vehicle to be recommended and historical traffic information generated by a user browsing and clicking on second vehicle information, wherein the first vehicle information includes model information and appearance information of the first vehicle; A first processing module, configured to process the first vehicle information and the historical traffic information through a preset vector representation model to obtain a first feature vector of the first vehicle; a recommendation module, configured to read the stored second feature vector, sort the first vehicle based on the first feature vector and the second feature vector, and determine a recommended vehicle according to the sorting result, wherein the second feature vector is generated by processing the historical transaction information of the second vehicle by the vector representation model, and is used to represent the user's intention label for the second vehicle; The first processing module is further used to generate a user click behavior sequence based on the historical traffic information; process the first vehicle information and the user click behavior sequence through a preset vector representation model to obtain a first feature vector of the first vehicle; The first processing module is further used to take the first vehicle information as input, take the adjacent vehicles of the first vehicle in the user click behavior sequence as output, train the vector representation model to obtain a hidden layer vector, and use the hidden layer vector as a first feature vector of the first vehicle.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the vehicle recommendation method according to any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the vehicle recommendation method according to any one of claims 1 to 5 is implemented.