Vehicle prediction method, system and equipment combined with knowledge graph representation learning

Abstract

The invention discloses a vehicle prediction method, system and device combined with knowledge graph representation learning, and relates to the technical field of knowledge graphs, and the method comprises the steps: obtaining a target picture or keyword corresponding to a to-be-predicted vehicle, extracting a historical knowledge graph corresponding to the to-be-predicted vehicle, taking the historical knowledge graph as the input of a pre-training model, and carrying out the pre-training of the to-be-predicted vehicle; combining knowledge graph representation learning to obtain vector representation results of entities and relationships corresponding to the historical knowledge graph; and determining a prediction result corresponding to the logarithm to-be-predicted vehicle according to the vector representation result in combination with the target picture or the keyword. According to the method, the knowledge graph is combined, accurate identification and prediction of the to-be-predicted vehicle are realized, and the scheme is combined with knowledge graph representation learning and considers characteristics of target structure characteristics and the like in the knowledge graph, so that the scheme is higher in adaptability and has generalization ability.

Description

Background Art

[0002] For the increasingly rich brands and models, users cannot conduct horizontal comparison of vehicles based on a certain keyword. Based on this, it is crucial to associate vehicle prediction with a knowledge spectrum graph to enhance the user's information acquisition experience.

[0003] Graph representation learning is a research field that maps entities and relationships in a knowledge graph to a continuous vector space. It can improve the computational efficiency and application diversity of the knowledge graph and can also be used as pre-trained vector input for downstream applications. The methods of graph representation learning are mainly divided into translational distance models and semantic matching models. The former regards the relationship as a spatial transformation from the head entity to the tail entity, and the latter uses neural networks to model semantic similarity.

[0004] In recent years, with the continuous in-depth research of big data and deep learning, representation learning technology has been widely applied in fields such as natural language processing and image recognition. The purpose of representation learning is to represent the research object with low-dimensional dense vectors. The closer the objects are in the low-dimensional space, the more similar they are semantically.

[0005] Existing network representation learning models focus more on the information of nodes themselves, ignoring the rich semantic information on the edges, and the edges usually do not have directions. Knowledge graph representation learning is a subfield of network representation learning. Since the edges in the graph contain specific semantics and have directions, the design of the model is more complex.

[0006] A knowledge graph is a structured form of knowledge representation. It organizes knowledge into triples (h, r, t) in the form of entities, relationships, and attributes, where h and t represent the head entity and the tail entity respectively, and r represents the relationship between them. The knowledge graph can effectively store and manage large-scale knowledge and provide support for various artificial intelligence applications.

[0007] In order to make the knowledge graph compatible with machine learning models such as neural networks, it needs to be converted into low-dimensional dense vectors, that is, graph representation learning. At present, many graph representation learning methods have been proposed, mainly divided into two categories: translational distance models and semantic matching models. The translational distance models regard the relationship as a spatial transformation from the head entity to the tail entity, such as TransE, TransH, TransR, etc.; the semantic matching models use neural networks to model semantic similarity, such as RESCAL, DistMult, NTN, etc.

[0008] However, there are some deficiencies in existing graph representation learning methods, mainly including the following points:

[0009] (1) It ignores the target structural features in the knowledge graph, such as node degree, neighbor nodes, path length, etc. These features reflect the importance and complexity of entities and relationships in the graph and have an important impact on representation learning and downstream tasks.

[0010] (2) It lacks the ability to model multi-hop relationships and complex relationships, such as one-to-many, many-to-one, many-to-many, etc. These relationships are common in the real world, but existing methods are difficult to capture the semantic connections and logical inferences between them.

[0011] (3) It lacks adaptability to knowledge graphs of different types and domains, such as different scales, densities, heterogeneity, etc. These factors will affect the effect and generalization ability of representation learning. Summary of the Invention

[0012] The technical problem to be solved by the present invention is to address the deficiencies of the prior art, specifically for problems such as slow development cycle and long debugging time. Specifically, a vehicle prediction method, system, and device combining knowledge graph representation learning are provided, as follows:

[0013] 1) In the first aspect, the present invention provides a vehicle prediction method combining knowledge graph representation learning, and the specific technical solution is as follows:

[0014] Obtain the target image or keyword corresponding to the vehicle to be predicted, extract the historical knowledge graph corresponding to the vehicle to be predicted, use the historical knowledge graph as the input of the pre-trained model, and combine knowledge graph representation learning to obtain the vector representation results of the entities and relationships corresponding to the historical knowledge graph;

[0015] Combine the target image or keyword, and determine the prediction result corresponding to the vehicle to be predicted according to the vector representation results.

[0016] The beneficial effects of a vehicle prediction method combining knowledge graph representation learning provided by the present invention are as follows:

[0017] Combined with the knowledge graph, accurate identification and prediction of the vehicle to be predicted are achieved. Moreover, this solution combines knowledge graph representation learning, considers characteristics such as target structural features in the knowledge graph, making this solution more adaptable and more generalization ability.

[0018] On the basis of the above solution, the present invention can also be improved as follows.

[0019] Further, the training process of the pre-trained model is as follows:

[0020] For any target knowledge graph in the training set, calculate the target structural features corresponding to each entity and relationship in the target knowledge graph;

[0021] Initialize each entity and relationship in the target knowledge graph in combination with the target structural features to obtain the target vector representation corresponding to each entity and relationship;

[0022] Based on the target vector representation, perform training in combination with an optimization algorithm and a loss function, and complete the training when the number of iterations reaches a preset number or when the convergence condition is satisfied.

[0023] Furthermore, the target structural features include:

[0024] Node degree, neighbor nodes, and path length.

[0025] Furthermore, the loss function includes positive samples and negative samples. The positive samples represent the real triples in the target knowledge graph, and the negative samples represent the virtual triples in the target knowledge graph. The triples are composed of a head entity, a tail entity, and the relationship between the head entity and the tail entity.

[0026] 2) In the second aspect, the present invention also provides a vehicle prediction system combined with knowledge graph representation learning. The specific technical solution is as follows:

[0027] The extraction module is used to: obtain the target picture or keyword corresponding to the vehicle to be predicted, extract the historical knowledge graph corresponding to the vehicle to be predicted, use the historical knowledge graph as the input of the pre-trained model, and obtain the vector representation result of the entities and relationships corresponding to the historical knowledge graph in combination with knowledge graph representation learning;

[0028] The prediction module is used to: determine the prediction result corresponding to the vehicle to be predicted logarithmically in combination with the target picture or keyword according to the vector representation result.

[0029] On the basis of the above solution, the present invention can also be improved as follows.

[0030] Furthermore, the training process of the pre-trained model is as follows:

[0031] For any target knowledge graph in the training set, calculate the target structural features corresponding to each entity and relationship in the target knowledge graph;

[0032] Initialize each entity and relationship in the target knowledge graph in combination with the target structural features to obtain the target vector representation corresponding to each entity and relationship;

[0033] Based on the target vector representation, perform training in combination with an optimization algorithm and a loss function, and complete the training when the number of iterations reaches a preset number or when the convergence condition is satisfied.

[0034] Furthermore, the target structural features include:

[0035] Node degree, neighbor nodes, and path length.

[0036] Furthermore, the loss function includes positive samples and negative samples. The positive samples represent the true triples in the target knowledge graph, and the negative samples represent the virtual triples in the target knowledge graph. A triple consists of a head entity, a tail entity, and the relationship between the head entity and the tail entity.

[0037] 3) In a third aspect, the present invention further provides an electronic device, which includes a processor. The processor is coupled to a memory, and at least one computer program is stored in the memory. The at least one computer program is loaded and executed by the processor to enable the electronic device to implement any one of the above methods.

[0038] 4) In a fourth aspect, the present invention further provides a computer-readable storage medium, in which at least one computer program is stored. The at least one computer program is loaded and executed by a processor to enable a computer to implement any one of the above methods.

[0039] It should be noted that for the beneficial effects obtained by the technical solutions and corresponding possible implementation manners of the second to fourth aspects of the present invention, reference may be made to the technical effects of the first aspect and its corresponding possible implementation manners described above, and details will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] By reading the detailed description of the non-limiting embodiments with reference to the following drawings, other features, objects, and advantages of the present invention will become more apparent:

[0041] Figure 1 It is a schematic flowchart of a vehicle prediction method combining knowledge graph representation learning according to an embodiment of the present invention;

[0042] Figure 2 It is a structural framework diagram of an electronic device. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0043] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the drawings.

[0044] As Figure 1 shown, a vehicle prediction method combining knowledge graph representation learning according to an embodiment of the present invention includes the following steps:

[0045] S1. Obtain the target picture or keywords corresponding to the vehicle to be predicted, extract the historical knowledge graph corresponding to the vehicle to be predicted, use the historical knowledge graph as the input of the pre-trained model, and combine knowledge graph representation learning to obtain the vector representation results of the entities and relationships corresponding to the historical knowledge graph;

[0046] S2. Combine the target picture or keywords, and determine the prediction result corresponding to the vehicle to be predicted according to the vector representation result.

[0047] The beneficial effects of a vehicle prediction method combining knowledge graph representation learning provided by the present invention are as follows:

[0048] Combined with the knowledge graph, accurate identification and prediction of the vehicle to be predicted are realized. Moreover, this solution combines knowledge graph representation learning, considers characteristics such as the target structure features in the knowledge graph, making this solution more adaptable and more generalizable.

[0049] It should be noted that the vehicle to be predicted is a vehicle that the user wants to understand or obtain relevant information about. The information obtained includes but is not limited to: vehicle price, basic information of the same type of vehicle, etc. The basic information of the same type of vehicle refers to the body parameters, vehicle fuel consumption, price, etc. of the same type of vehicle.

[0050] The target picture can be a picture screenshot or downloaded by the user from the Internet, or a picture taken by the user himself. The keywords are at least two phrases obtained after extraction from the text content input by the user through the interaction interface.

[0051] In another embodiment of this solution, for the target picture or keywords given by the user, the above content needs to be text-extracted, that is, processed by a large language model to obtain optimized text content. Based on the above optimized text content, the historical knowledge graph corresponding to the optimized text content can be accurately extracted.

[0052] The process of performing corresponding processing through the large language model is specifically as follows:

[0053] Input the target picture into the large language model, and after recognition processing, obtain the text description corresponding to the target picture.

[0054] The process of segmenting the text description content and extracting the text content input by the user is as follows:

[0055] Determine the target length of the text description content or the text content. According to the comparison result between the target length and the preset length, determine the number of segments. If the target length is lower than the preset length, perform 1 segmentation on the target length. If the target length exceeds the preset length, it is necessary to further determine the multiple by which it exceeds the preparatory length. If the multiple exceeds 5 times, define the text description content or the text content as a paragraph and re-segment it according to the paragraph segmentation strategy. If the multiple does not exceed 5 times, add 2 more segmentations for each multiple exceeded. That is, in order to ensure that the true meaning of the text description content or the text content can be reflected to the greatest extent, the content exceeding 2 times the preset length needs to be segmented multiple times.

[0056] The paragraph segmentation strategy is specifically as follows:

[0057] Determine whether the user continuously inputs multiple groups of text content or continuously presents multiple target images. If so, integrate the continuously input text content or target images, determine the correlation between the current target image or the current text content and the previous text content or the previous target image (the correlation between target images is confirmed through the text description content), and when the correlation is higher than the preset correlation, extract the same third keyword from the first keyword corresponding to the previous target image or the previous text content used to calculate the correlation and the second keyword corresponding to the current target image or the current text content, and locate the position of the third keyword in the current target image or the current text content, and perform segmentation processing based on the position of the third keyword.

[0058] If the correlation is lower than the preset correlation, determine the fixed segmentation length, and segment any paragraph according to the fixed segmentation length to obtain at least two chunks.

[0059] It should be further noted that if there are commas or semicolons in the chunks obtained during the segmentation process, re-segmentation is required, that is, segment according to the commas or semicolons, and after segmentation, start re-segmenting according to the fixed segmentation length from the first text content after the comma or semicolon.

[0060] Furthermore, the training process of the pre-trained model is as follows:

[0061] For any target knowledge graph in the training set, calculate the target structural features corresponding to each entity and relationship in the target knowledge graph;

[0062] Initialize each entity and relationship in the target knowledge graph in combination with the target structural features to obtain the target vector representation corresponding to each entity and relationship;

[0063] Based on the target vector representation, perform training in combination with an optimization algorithm and a loss function, and complete the training when the number of iterations reaches the preset number of times or when the convergence condition is met.

[0064] Furthermore, the target structural features include:

[0065] Node degree, neighbor nodes, and path length.

[0066] Furthermore, the loss function includes positive samples and negative samples. The positive samples represent the real triples in the target knowledge graph, and the negative samples represent the virtual triples in the target knowledge graph. The triples are composed of a head entity, a tail entity, and the relationship between the head entity and the tail entity.

[0067] In the above embodiments, although the steps are numbered S1, S2, etc., they are only specific embodiments given by the present invention. Those skilled in the art can adjust the execution order of S1, S2, etc. according to the actual situation, which is also within the protection scope of the present invention. It can be understood that in some embodiments, it may include some or all of the above embodiments.

[0068] The present invention also provides a vehicle prediction system combined with knowledge graph representation learning. The specific technical solution is as follows:

[0069] The extraction module is used to: obtain the target picture or keyword corresponding to the vehicle to be predicted, extract the historical knowledge graph corresponding to the vehicle to be predicted, use the historical knowledge graph as the input of the pre-trained model, and obtain the vector representation results of the entities and relationships corresponding to the historical knowledge graph by combining knowledge graph representation learning;

[0070] The prediction module is used to: combine the target picture or keyword, and determine the prediction result corresponding to the vehicle to be predicted according to the vector representation results.

[0071] Based on the above solution, the present invention can also be improved as follows.

[0072] Furthermore, the training process of the pre-trained model is as follows:

[0073] For any target knowledge graph in the training set, calculate the target structural features corresponding to each entity and relationship in the target knowledge graph;

[0074] Initialize each entity and relationship in the target knowledge graph by combining the target structural features to obtain the target vector representation corresponding to each entity and relationship;

[0075] Based on the target vector representation, perform training by combining an optimization algorithm and a loss function, and complete the training when the number of iterations reaches a preset number or when the convergence condition is met.

[0076] Furthermore, the target structural features include:

[0077] Node degree, neighbor nodes, and path length.

[0078] Furthermore, the loss function includes positive samples and negative samples. The positive samples represent the true triples in the target knowledge graph, and the negative samples represent the virtual triples in the target knowledge graph. A triple consists of a head entity, a tail entity, and the relationship between the head entity and the tail entity.

[0079] Example 1: 1. For a given knowledge graph, calculate the target structural features of each entity and relationship and use them as additional input information. The target structural features can reflect the importance and complexity of entities and relationships in the graph and have an important impact on representation learning and downstream tasks. For example, the node degree can indicate how many other entities or relationships an entity or relationship is involved in; neighbor nodes can indicate which other entities or relationships an entity or relationship has direct connections with; the path length can indicate the distance between an entity or relationship and other entities or relationships. These features can be calculated through some simple statistical methods, and different combinations of features can be selected according to different application scenarios. In this example, the present invention assumes using node degree, neighbor nodes, and path length as the target structural features and concatenating them with the entity and relationship vectors as additional input information.

[0080] 2. For each entity and relationship, initialize a random vector and use it as the input of the graph neural network (GNN). GNN is a neural network model that can capture graph structure information. It updates the vector representation of each node through multi-layer information propagation and aggregation. There are various variants of GNN, such as graph convolutional network (GCN), graph attention network (GAT), relational graph convolutional network (R-GCN), etc. In this example, the present invention assumes using R-GCN as the GNN model. R-GCN is a GNN model that can handle heterogeneous relational graphs. It uses different weight matrices for different types of relationships for information propagation and aggregation, thereby enhancing the expressive power of the model. The information propagation and aggregation process of R-GCN can be expressed as:

[0081]

[0082] where \(h_i^{(l)}\) represents the vector representation of node \(i\) at the \(l\)-th layer, represents the set of relationship types, represents the set of nodes connected to node \(i\) by relationship \(r\), \(c_{i,r}\) represents the normalization factor, \(W\) rW(l) and W0(l) represent the weight matrices of relation r and self-loop at the l-th layer, and σ represents the activation function, such as ReLU or tanh. Through the multi-layer R-GCN, each node can fuse the information of its neighbor nodes and different relation types to obtain a richer and deeper vector representation.

[0083] 3. For each triple (h, r, t), use a semantic matching model, such as DistMult or NTN, to calculate its scoring function, which represents the possibility of the triple being valid. The scoring function contains the vector representations of entities and relations, as well as the target structural features. The semantic matching model is a graph representation learning method that uses neural networks to model semantic similarity. It can capture the complex semantic connections and logical inferences between entities and relations. There are various variants of the semantic matching model, such as RESCAL, DistMult, NTN, ConvE, etc. In this example, the present invention assumes that DistMult is used as the semantic matching model. DistMult is a semantic matching model based on a bilinear function. It treats both entities and relations as vectors and calculates the scoring function of the triple through the inner product:

[0084] f(h, r, t) = h T Rt

[0085] where h and t represent the vector representations of the head entity and the tail entity, and R represents the diagonal matrix of the relation. DistMult can effectively model symmetric relations and multi-ary relations, but it cannot model anti-symmetric relations and non-linear relations. To enhance the expressive power of DistMult, the present invention adds target structural features, such as node degree, neighbor nodes, path length, etc., to the scoring function, so as to obtain a more comprehensive and accurate triple evaluation.

[0086] 4. Use an optimization algorithm, such as Stochastic Gradient Descent (SGD), to minimize a loss function, such as cross-entropy or contrastive loss, to update the vector representations of entities and relations. The loss function contains positive samples and negative samples. Positive samples are the real triples existing in the knowledge graph, and negative samples are the fake triples obtained by randomly replacing the head entity or the tail entity. The optimization algorithm is a mathematical method for finding the optimal solution. It continuously adjusts the parameters to make the loss function reach the minimum value. Stochastic Gradient Descent (SGD) is a commonly used optimization algorithm. It calculates the gradient using only one or a small batch of samples each time and updates the parameters along the opposite direction of the gradient. SGD can effectively handle large-scale datasets and can escape from local optima. In this example, the present invention assumes that SGD is used as the optimization algorithm and contrastive loss is used as the loss function. Contrastive loss is a commonly used loss function. It distinguishes real triples and fake triples by maximizing the score of positive samples and minimizing the score of negative samples. The contrastive loss can be expressed as:

[0087]

[0088] Among them, S represents the set of positive samples, S' represents the set of negative samples, and γ represents a positive margin parameter. By minimizing the contrastive loss, the present invention can update the vector representations of entities and relationships, enabling them to better reflect the semantic and structural information in the knowledge graph. f(h, r, t) represents the loss function corresponding to the true triple, and f(h′, r′, t′) represents the loss function corresponding to the false triple.

[0089] 5. Repeat steps 2 to 4 until the preset number of iterations or convergence condition is reached to obtain the final vector representations of entities and relationships. The number of iterations or convergence condition is a termination condition used to control the training process, which can be adjusted according to different datasets and models. Generally, when the number of iterations reaches a certain value or the change in the loss function is less than a certain threshold, the present invention considers that the model has converged and stops training. Through multiple iterations, the present invention can obtain the final vector representations of entities and relationships, which can be used as low-dimensional dense vectors of the knowledge graph and can be applied to tasks such as question answering, reasoning, and completion of the knowledge graph.

[0090] It should be noted that the beneficial effects of the vehicle prediction system combining knowledge graph representation learning provided in the above embodiments are the same as those of the vehicle prediction method combining knowledge graph representation learning, and will not be elaborated here. In addition, when the system provided in the above embodiments implements its functions, only the division of the above functional modules is used as an example. In practical applications, the above functions can be allocated to different functional modules according to needs, that is, the system can be divided into different functional modules according to the actual situation to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be seen in the method embodiments and will not be elaborated here.

[0091] As Figure 2 shown, an electronic device 300 according to an embodiment of the present invention, the electronic device 300 includes a processor 320, the processor 320 is coupled to a memory 310, and at least one computer program 330 is stored in the memory 310. The at least one computer program 330 is loaded and executed by the processor 320 to enable the electronic device 300 to implement any of the above methods. Specifically:

[0092] The electronic device 300 can vary significantly due to different configurations or performances. It may include one or more processors 320 (Central Processing Units, CPUs) and one or more memories 310. Among them, at least one computer program 330 is stored in the one or more memories 310. The at least one computer program 330 is loaded and executed by the one or more processors 320, so that the electronic device 300 implements a vehicle prediction method combining knowledge graph representation learning provided in the above embodiments. Of course, the electronic device 300 may also have components such as a wired or wireless network interface, a keyboard, and an input / output interface for input and output. The electronic device 300 may also include other components for implementing the functions of the device, which will not be elaborated here.

[0093] A computer-readable storage medium according to an embodiment of the present invention stores at least one computer program, and the at least one computer program is loaded and executed by a processor so that a computer implements any one of the above methods.

[0094] Optionally, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), a magnetic tape, a floppy disk, an optical data storage device, etc.

[0095] In an exemplary embodiment, there is also provided a computer program product or a computer program. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the electronic device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the electronic device executes any one of the above methods.

[0096] It should be noted that the terms "first", "second", etc. in the specification and claims of this application are used to distinguish similar objects, and do not represent a limitation on a specific order or sequence. Under appropriate circumstances, the use order of similar objects may be interchanged so that the embodiments of the present application described here can be implemented in an order other than the illustrated or described order.

[0097] Those skilled in the art of the present technology know that the present invention can be implemented as a system, method, or computer program product. Therefore, the present disclosure can be specifically implemented in the following forms, namely: it can be entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, which is generally referred to as "circuit", "module", or "system" in this article. In addition, in some embodiments, the present invention can also be implemented in the form of a computer program product in one or more computer-readable media, which contain computer-readable program code.

[0098] Any combination of one or more computer-readable media can be adopted. The computer-readable media can be computer-readable signal media or computer-readable storage media. The computer-readable storage media can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive list) of the computer-readable storage media include: an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, the computer-readable storage media can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0099] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A vehicle prediction method combined with knowledge graph representation learning, characterized in that: include: Obtain the target image or keyword corresponding to the vehicle to be predicted, extract the historical knowledge graph corresponding to the vehicle to be predicted, use the historical knowledge graph as the input of the pre-training model, and combine the knowledge graph representation learning to obtain the vector representation results of the entities and relationships corresponding to the historical knowledge graph; In combination with the target image or keyword, a prediction result corresponding to the vehicle to be predicted is determined according to the vector representation result.

2. A vehicle prediction method combined with knowledge graph representation learning according to claim 1, characterized in that: The training process of the pre-trained model is: For any target knowledge graph in the training set, calculate the target structural features corresponding to each entity and relationship in the target knowledge graph; Initialize each entity and relationship in the target knowledge graph in combination with the target structural features to obtain a target vector representation corresponding to each entity and relationship; Based on the target vector representation, training is performed in combination with an optimization algorithm and a loss function, and the training is completed when the number of iterations reaches a preset number or when a convergence condition is met.

3. The vehicle prediction method combined with knowledge graph representation learning according to claim 1 is characterized in that: The target structural features include: Node degree, neighbor nodes, and path length.

4. The vehicle prediction method combined with knowledge graph representation learning according to claim 2 is characterized in that: The loss function includes positive samples and negative samples, the positive samples represent real triples in the target knowledge graph, and the negative samples represent virtual triples in the target knowledge graph, and the triples are composed of a head entity, a tail entity, and the relationship between the head entity and the tail entity.

5. A vehicle prediction system combined with knowledge graph representation learning, characterized in that: include: The extraction module is used to: obtain the target image or keyword corresponding to the vehicle to be predicted, extract the historical knowledge graph corresponding to the vehicle to be predicted, use the historical knowledge graph as the input of the pre-training model, and combine the knowledge graph representation learning to obtain the vector representation results of the entities and relationships corresponding to the historical knowledge graph; The prediction module is used to: determine the prediction result corresponding to the vehicle to be predicted based on the vector representation result in combination with the target image or keyword.

6. A vehicle prediction system combined with knowledge graph representation learning according to claim 5, characterized in that: The training process of the pre-trained model is: For any target knowledge graph in the training set, calculate the target structural features corresponding to each entity and relationship in the target knowledge graph; Initialize each entity and relationship in the target knowledge graph in combination with the target structural features to obtain a target vector representation corresponding to each entity and relationship; Based on the target vector representation, training is performed in combination with an optimization algorithm and a loss function, and the training is completed when the number of iterations reaches a preset number or when a convergence condition is met.

7. A vehicle prediction system combined with knowledge graph representation learning according to claim 5, characterized in that: The target structural features include: Node degree, neighbor nodes, and path length.

8. The vehicle prediction system combined with knowledge graph representation learning according to claim 6, characterized in that: The loss function includes positive samples and negative samples, the positive samples represent real triples in the target knowledge graph, and the negative samples represent virtual triples in the target knowledge graph, and the triples are composed of a head entity, a tail entity, and the relationship between the head entity and the tail entity.

9. An electronic device, characterized in that: The electronic device comprises a processor, the processor is coupled to a memory, at least one computer program is stored in the memory, and the at least one computer program is loaded and executed by the processor so that the electronic device implements the method according to any one of claims 1 to 4.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one computer program, and the at least one computer program is loaded and executed by a processor so that a computer implements the method according to any one of claims 1 to 4.

Images