Traffic data asset entity alignment method and device and electronic equipment
By using the BERT model in the alignment of the knowledge map entity for the study, combined with the weighted average technology and boundary similarity, the problem of alignment efficiency when the existing model is limited or the attribute information is low, and the high accuracy has been achieved. The physical alignment.
Patent Information
- Application Number
- CN202411420261.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-12
- Publication Date
- 2025-01-28
AI Technical Summary
When aligning knowledge graph entities, existing models rely on relationships and attribute triplets, and require a large number of seed alignments, and cannot be effectively aligned with limited seed count or low attribute information.
Embedded based on the BERT model, calculate the embedded of entity, relationship, and attributes, calculate the overall score of the entity through the weighted average technology, and use the boundary similarity to calculate the similarities of the entity, and the output predicting alignment results.
In the case of less number of seeds and less attribute information, high accuracy of knowledge map entity is achieved, which improves the efficiency and accuracy of alignment.
Smart Images
Figure CN119378658A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of traffic data asset entity alignment, and in particular to a traffic data asset entity alignment method, device and electronic equipment. Background Art
[0002] In recent years, KG (Knowledge Graphs) has received extensive attention and has been widely used in artificial intelligence applications such as question answering, semantic search, information retrieval, and web mining. KG contains a large amount of structured knowledge / facts. It is a multi-relation directed graph composed of entities as graph nodes and relationships as edges. It is an effective way to store facts in the real world as structured data in the form of triples. At present, there are entity alignment models based on KG embedding that represent different KGs as embeddings and find entity alignment by calculating the similarity between embeddings. Traditional models mostly rely on relationship and attribute triplets and require seeds (aligned triplets from two knowledge graphs) to align, but in general, the number of seeds is very limited. At the same time, although attributes are very useful in entity alignment tasks, existing models cannot align entities of two knowledge graphs when attribute information is less. Therefore, they need to be improved. Summary of the invention
[0003] Based on this, it is necessary to provide a method, device and electronic device for entity alignment of traffic data assets, as traditional models mostly rely on relationship and attribute triplets and require seed alignment. However, in general, the number of seeds is very limited. At the same time, although attributes are very useful in entity alignment tasks, existing models cannot perform entity alignment of two knowledge graphs when attribute information is less.
[0004] The present invention provides a method for aligning traffic data asset entities, the method comprising:
[0005] Merge and align two knowledge graphs based on the same predicate alignment algorithm;
[0006] Learn embedding representations for entities, relations, and attributes;
[0007] Call the BERT model to calculate the summary embedding, relationship embedding, and attribute embedding of entities in the two knowledge graphs;
[0008] The weighted average technique is used to calculate the overall scores of entities in the two knowledge graphs;
[0009] Calculate the similarity between two entities based on boundary similarity and output the predicted alignment result.
[0010] In one embodiment, the embedding representation learning of entities, relations and attributes includes:
[0011] Call the BERT model to calculate the summary embedding of the entity and generate a word vector representing the main meaning of the entity;
[0012] Learn relation embedding based on TransE and Graph Embedding algorithms and preserve the relation structure in the knowledge graph;
[0013] Use attribute embeddings to capture holistic features that match entities or attributes.
[0014] In one embodiment, the entity summary embedding is learned using three different scoring functions, and the calculation formula of the scoring function is as follows:
[0015] f s =||h sum +rt sum || l1 / l2 ,
[0016] f st =||h sum +rt|| l1 / l2 ,
[0017] f hs =||h+rt sum || l1 / l2 ,
[0018] Among them, h sum and t sum are the entity summary head and entity summary tail, h and t are the head entity and tail entity, r is the relationship between h and t, l1 / l2 represents the L1 / L2 norm, the L1 norm is the Manhattan norm, and the L2 norm is the Euclidean norm.
[0019] In one embodiment, the scoring function is applicable to a zero-shot scenario, and a summary-based representation is performed through a summary vector generated by BERT. The calculation formula for entity embedding learning based on the summary is as follows:
[0020] L SUM =f s +f st +f hs ,
[0021] Among them, L SUM is the loss function based on the summary entity representation, and the goal is to minimize L SUM .
[0022] In one embodiment, learning relation embedding based on TransE and Graph Embedding algorithm and retaining the relation structure in the knowledge graph includes:
[0023] Based on relation embedding learning, the relation structure is preserved by minimizing the objective function, the objective function includes the scoring function of the valid triples and the negative triples, and the calculation formula of the minimization objective function based on relation embedding is as follows:
[0024]
[0025] in, γ is the margin, f r (h, t) and f r (h', t') represent the scoring functions of valid triples and negative triples respectively, T r is the set of valid relation triples, T' r is the set of damaged relation triples, f r is the likelihood function of the relation triple, count(r) is the number of occurrences of relation r, and |T| is the total number of triples in the merged knowledge graph;
[0026] The optimization is performed based on the likelihood functions of valid relation triplets and damaged relation triplets.
[0027] In one embodiment, the weighted average technique is used to calculate the overall scores of entities in two knowledge graphs, including:
[0028] Compute the average embedding among entity summary embedding, relation embedding, and attribute embedding, and calculate the deviation of each embedding from the average, as follows:
[0029]
[0030] in, is the average value of embeddings, M is the number of embeddings, e (i) is the embedding of each component, ω i is the weight;
[0031] The entity summary embedding, relation embedding, and attribute embedding are linearly combined, and weights are assigned to each entity embedding item. The calculation formula is as follows:
[0032]
[0033] Among them, e cmb Represents a combined embedding.
[0034] In one embodiment, the calculating the similarity between two entities based on boundary similarity includes:
[0035] Calculate the similarity between the query entity and the target entity and find the aligned entity pair. The calculation formula is as follows:
[0036]
[0037] Among them, e1∈KG1 is the query entity, e2∈KG2 is the target entity;
[0038] Sort the ranks of the target entities that are actually aligned to the top of the rank list.
[0039] The present invention also provides a traffic data asset entity alignment device, comprising:
[0040] The alignment module is used to merge and align two knowledge graphs based on the same predicate alignment algorithm;
[0041] Learning module, used to learn embedding representations of entities, relations, and attributes;
[0042] The calling module is used to call the BERT model to calculate the summary embedding, relationship embedding, and attribute embedding of entities in the two knowledge graphs;
[0043] A calculation module, used to calculate the overall scores of entities in two knowledge graphs using a weighted average technique;
[0044] The output module is used to calculate the similarity between two entities based on boundary similarity and output the predicted alignment result.
[0045] The present invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method for aligning traffic data asset entities as described in any one of the above is implemented.
[0046] The present invention also provides a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the method for aligning traffic data asset entities as described in any one of the above is implemented.
[0047] The above-mentioned traffic data asset entity alignment method, device and electronic device adopts the method of embedding representation learning of entities, relationships and attributes, and then uses weighted average technology to calculate the overall score of entities in two knowledge graphs, and finally uses weighted average technology for integration, replacing the traditional embedding-based entity alignment method. It can align the two knowledge graphs with high accuracy with a small number of seeds, and at the same time, it is convenient to align the entities of the two knowledge graphs with less attribute information. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0049] Figure 1 A schematic diagram of a flow chart of a method for aligning a traffic data asset entity in an embodiment;
[0050] Figure 2 is a flow chart of an embedding learning method in one embodiment;
[0051] Figure 3 A flowchart of a relation embedding learning method in one embodiment is shown;
[0052] Figure 4 A schematic diagram of a method for calculating an overall score in one embodiment;
[0053] Figure 5 is a schematic diagram of a similarity calculation method flow in an embodiment;
[0054] Figure 6 A schematic diagram of a process architecture of a method for aligning a traffic data asset entity in an embodiment;
[0055] Figure 7 It is a schematic diagram of the structure of a traffic data asset entity alignment device in one embodiment;
[0056] Figure 8 The figure is a diagram showing the internal structure of an electronic device according to an embodiment. DETAILED DESCRIPTION
[0057] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0058] In recent years, KG (Knowledge Graphs) has received widespread attention and has been widely used in artificial intelligence applications such as question answering, semantic search, information retrieval and web mining. KG contains a large amount of structured knowledge or facts. It is a multi-relationship directed graph composed of entities as graph nodes and relationships as edges. It is an effective way to store facts in the real world as structured data in the form of triples. The relationship triple can be expressed as (h, r, t), where h and t are the head entity and tail entity respectively, and r is the relationship between h and t. If the object is text, it is called an attribute triple.
[0059] The entity alignment task refers to finding the same real entities in multiple KGs. In recent years, using KG embedding technology to solve the entity alignment problem has received increasing attention. The KG embedding model represents entities and relationships in a low-dimensional vector space while retaining the KG semantics. Currently, there are entity alignment models based on KG embedding that represent different KGs as embeddings and find entity alignment by calculating the similarity between embeddings.
[0060] Most existing models rely on relation and attribute triplets, which require seeds (aligned triplets from two knowledge graphs) to align, but the number of seeds is generally very limited. At the same time, although attributes are very useful in entity alignment tasks, existing models cannot align entities between two knowledge graphs when there is less attribute information.
[0061] Combine the following Figure 1-Figure 8 The traffic data asset entity alignment method, device and electronic device of the present invention are described.
[0062] like Figure 1 and Figure 6 As shown, in one embodiment, a method for aligning traffic data asset entities includes the following steps:
[0063] Step S100, merging and aligning the two knowledge graphs based on the same predicate alignment algorithm.
[0064] Step S200, embedding representation learning is performed on entities, relations and attributes.
[0065] Step S300, calling the BERT model to calculate the summary embedding, relationship embedding and attribute embedding of the entities in the two knowledge graphs.
[0066] Step S400, using weighted average technology to calculate the overall scores of entities in the two knowledge graphs.
[0067] Step S500, calculating the similarity between two entities based on boundary similarity, and outputting a predicted alignment result.
[0068] This traffic data asset entity alignment method adopts the embedding representation learning of entities, relationships and attributes, and then uses the weighted average technology to calculate the overall score of the entities in the two knowledge graphs. Finally, the weighted average technology is used for integration, replacing the traditional embedding-based entity alignment method. It can align the two knowledge graphs with high accuracy with a smaller number of seeds, and at the same time, it is convenient to align the entities of the two knowledge graphs with less attribute information.
[0069] In this embodiment, see Figure 2 , embedding representation learning for entities, relations, and attributes, including:
[0070] Step S210, calling the BERT model to calculate the summary embedding of the entity and generate a word vector representing the main meaning of the entity.
[0071] It should be noted that the summary is a text description associated with each entity. The Transformer's bidirectional encoder BERT is used to generate a set of word vectors from the summary of each specific entity. These word vectors can usually represent the main meaning of the entity.
[0072] Among them, the BERT (Bidirectional Encoder Representation from Transformer) model is built on the basis of multi-layer bidirectional Transformer encoders. By jointly adjusting the contextual information, deep bidirectional representations are pre-trained from unlabeled text. BERT is applicable to a wide range of tasks such as word embedding, question answering, and language reasoning without substantial task-specific architecture modifications. For each entity, the embedding of each word is obtained by providing the summary to the BERT model, and then the embedding of each word is summarized to obtain the entity embedding based on the summary of the specific entity.
[0073] Step S220, learning relation embedding based on TransE and Graph Embedding algorithm, and retaining the relation structure in the knowledge graph.
[0074] Step S230, using attribute embedding to capture overall features matching the entity or attribute.
[0075] It should be noted that although summary embedding is effective in entity alignment, due to the small number of attributes or no attribute values in the knowledge graph, entity summary embedding is learned using three different scoring functions. The calculation formula of the scoring function is as follows:
[0076] f s =||h sum +rt sum || l1 / l2 ,
[0077] f st =||h sum +rt|| l1 / l2 ,
[0078] f hs =||h+rt sum || l1 / l2 ,
[0079] Among them, h sum and t sumare the entity summary head and entity summary tail, h and t are the head entity and tail entity, r is the relationship between h and t, l1 / l2 represents the L1 / L2 norm, the L1 norm is the Manhattan norm, and the L2 norm is the Euclidean norm.
[0080] For the scoring function f s ,h sum and t sum It is the representation of the entity summary head and entity summary tail, and the scoring function f st Learning entity summary head representation, scoring function f st The tail of the entity summary is represented based on the relation embedding, f st With f hs The entity summary header and entity summary footer have completely opposite representations.
[0081] In this embodiment, since the zero-shot scenario focuses on the situation where at least one entity in the triple is not in the training sample, the representation method based on the relational structure cannot represent entities outside the training set and thus cannot handle the zero-shot scenario. Therefore, the scoring function is applicable to the zero-shot scenario, and the summary vector generated by BERT is used for summary-based representation. The calculation formula for entity embedding learning based on the summary is as follows:
[0082] L SUM =f s +f st +f hs ,
[0083] Among them, L SUM is the loss function based on the summary entity representation, and the goal is to minimize L SUM .
[0084] The summary vector generated by BERT is used for summary-based representation, which facilitates the problem of entity representation beyond the training set in zero-shot scenarios.
[0085] In this embodiment, see Figure 3 , learn relation embedding based on TransE and Graph Embedding algorithms, and retain the relation structure in the knowledge graph, including:
[0086] Step S221, based on relation embedding learning, the relation structure is preserved by minimizing the objective function, the objective function includes the scoring function of the valid triples and the negative triples, and the calculation formula of the minimization objective function based on relation embedding is as follows:
[0087]
[0088] in, γ is the margin, f r(h, t) and f r (h', t') represent the scoring functions of valid triples and negative triples respectively, T r is the set of valid relation triples, T'r is the set of damaged relation triples, f r is the likelihood function of the relation triple, count(r) is the number of occurrences of relation r, and |T| is the total number of triples in the merged knowledge graph.
[0089] Step S222, optimizing based on likelihood functions of valid relation triples and damaged relation triples.
[0090] In this embodiment, see Figure 4 , the weighted average technique is used to calculate the overall scores of entities in the two knowledge graphs, including:
[0091] Step S410, calculate the average embedding of entity summary embedding, relationship embedding and attribute embedding, and calculate the deviation of each embedding from the average value, the calculation formula is as follows:
[0092]
[0093] in, is the average value of embeddings, M is the number of embeddings, e (i) is the embedding of each component, ω i is the weight.
[0094] Step S420, linearly combine the entity summary embedding, the relationship embedding, and the attribute embedding, and assign a weight to each entity embedding item, and the calculation formula is as follows:
[0095]
[0096] Among them, e cmb Represents a combined embedding.
[0097] The overall score is obtained by merging the scores of three features, entity summary, relationship and attribute, in an integrated way, which ensures that if an embedding is far away from its average embedding, it will have a lower weight, thereby improving the accuracy of entity alignment.
[0098] In this embodiment, see Figure 5 , the similarity between two entities is calculated based on boundary similarity, including:
[0099] Step S510, calculate the similarity between the query entity and the target entity, and find the aligned entity pair. The calculation formula is as follows:
[0100]
[0101] Among them, e1∈KG1 is the query entity, e2∈KG2 is the target entity;
[0102] Step S520, arranging the ranks of the truly aligned target entities at the top of the rank list.
[0103] The traffic data asset entity alignment device provided by the present invention is described below. The traffic data asset entity alignment device described below and the traffic data asset entity alignment method described above can be referenced to each other.
[0104] Figure 6 The overall architecture of the entity alignment method for traffic data assets is presented. The implementation of the entity alignment method for traffic data assets is divided into three steps: predicate alignment between knowledge graphs KG1 and KG2; embedding representation learning of entities, relations, and attributes; entity alignment process. For relation embedding, since a unified vector space is required, the two KGs are merged by using the same predicate alignment method described in AttributeeE. Assuming that both knowledge graphs KG1 and KG2 have entity A (respectively denoted as AKG1 and AKG2), BERT is first used to calculate the summary embedding, relation embedding, and attribute embedding of AKG1 and AKG2; then, the overall scores of AKG1 and AKG2 are calculated by using the weighted average technique; finally, the boundary similarity is used to predict whether WashingtonKG1 and WashingtonKG2 are aligned.
[0105] like Figure 7 As shown, in one embodiment, a traffic data asset entity alignment device includes an alignment module 710 , a learning module 720 , a calling module 730 , a calculation module 740 and an output module 750 .
[0106] The alignment module 710 is used to merge and align two knowledge graphs based on the same predicate alignment algorithm.
[0107] The learning module 720 is used to learn embedded representations of entities, relations, and attributes.
[0108] The calling module 730 is used to call the BERT model to calculate the summary embedding, relationship embedding and attribute embedding of the entities in the two knowledge graphs.
[0109] The calculation module 740 is used to calculate the overall scores of the entities in the two knowledge graphs using a weighted average technique.
[0110] The output module 750 is used to calculate the similarity between two entities based on the boundary similarity and output the predicted alignment result.
[0111] The learning module 720 is specifically used for:
[0112] Call the BERT model to calculate the summary embedding of the entity and generate a word vector representing the main meaning of the entity;
[0113] Learn relation embedding based on TransE and Graph Embedding algorithms and preserve the relation structure in the knowledge graph;
[0114] Use attribute embeddings to capture holistic features that match entities or attributes.
[0115] It should be noted that the entity summary embedding is learned using three different scoring functions. The scoring function is suitable for zero-shot scenarios, and the summary vector generated by BERT is used for summary-based representation.
[0116] In this embodiment, relation embedding is learned based on TransE and Graph Embedding algorithms, and the relation structure in the knowledge graph is retained, which is specifically used for:
[0117] Based on relation embedding learning, the relation structure is preserved by minimizing the objective function, which includes the scoring functions of valid triples and negative triples.
[0118] The optimization is performed based on the likelihood functions of valid relation triplets and damaged relation triplets.
[0119] In this embodiment, the calculation module 740 is specifically used for:
[0120] Compute the average embedding among entity summary embeddings, relation embeddings, and attribute embeddings, and calculate the deviation of each embedding from the average;
[0121] The entity summary embedding, relation embedding, and attribute embedding are linearly combined and weights are assigned to each entity embedding.
[0122] In this embodiment, the output module 750 is specifically used for:
[0123] Calculate the similarity between the query entity and the target entity and find aligned entity pairs;
[0124] Sort the ranks of the target entities that are actually aligned to the top of the rank list.
[0125] The traffic data asset entity alignment device adopts the embedding representation learning of entities, relationships and attributes, and then uses the weighted average technology to calculate the overall score of the entities in the two knowledge graphs. Finally, the weighted average technology is used for integration, replacing the traditional embedding-based entity alignment method. It can align the two knowledge graphs with high accuracy with a small number of seeds, and at the same time, it is convenient to align the entities of the two knowledge graphs with less attribute information.
[0126] Figure 8 An example of a physical structure diagram of an electronic device is shown. The electronic device may be a smart terminal, and its internal structure diagram may be as follows: Figure 8 As shown. The electronic device includes a processor, a memory and a network interface connected via a system bus. The processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the electronic device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a method for aligning a traffic data asset entity is implemented, and the method includes:
[0127] Merge and align two knowledge graphs based on the same predicate alignment algorithm;
[0128] Learn embedding representations for entities, relations, and attributes;
[0129] Call the BERT model to calculate the summary embedding, relationship embedding, and attribute embedding of entities in the two knowledge graphs;
[0130] The weighted average technique is used to calculate the overall scores of entities in the two knowledge graphs;
[0131] Calculate the similarity between two entities based on boundary similarity and output the predicted alignment result.
[0132] Those skilled in the art will understand that the structure shown in Figure x is merely a block diagram of a partial structure related to the scheme of the present invention, and does not constitute a limitation on the electronic device to which the scheme of the present invention is applied. The specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0133] On the other hand, the present invention further provides a computer storage medium storing a computer program, which implements a method for aligning traffic data asset entities when the computer program is executed by a processor, the method comprising:
[0134] Merge and align two knowledge graphs based on the same predicate alignment algorithm;
[0135] Learn embedding representations for entities, relations, and attributes;
[0136] Call the BERT model to calculate the summary embedding, relationship embedding, and attribute embedding of entities in the two knowledge graphs;
[0137] The weighted average technique is used to calculate the overall scores of entities in the two knowledge graphs;
[0138] Calculate the similarity between two entities based on boundary similarity and output the predicted alignment result.
[0139] In another aspect, a computer program product or a computer program is provided, the computer program product or the computer program comprising computer instructions, the computer instructions being stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium, and when the processor executes the computer instructions, a method for aligning a traffic data asset entity is implemented, the method comprising:
[0140] Merge and align two knowledge graphs based on the same predicate alignment algorithm;
[0141] Learn embedding representations for entities, relations, and attributes;
[0142] Call the BERT model to calculate the summary embedding, relationship embedding, and attribute embedding of entities in the two knowledge graphs;
[0143] The weighted average technique is used to calculate the overall scores of entities in the two knowledge graphs;
[0144] Calculate the similarity between two entities based on boundary similarity and output the predicted alignment result.
[0145] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing related hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory.
[0146] By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.
[0147] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above 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.
[0148] The above-described embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the present invention. It should be pointed out that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the attached claims.
Claims
1. A method for aligning traffic data asset entities, characterized in that: The method comprises: Merge and align two knowledge graphs based on the same predicate alignment algorithm; Learn embedding representations for entities, relations, and attributes; Call the BERT model to calculate the summary embedding, relationship embedding, and attribute embedding of entities in the two knowledge graphs; The weighted average technique is used to calculate the overall scores of entities in the two knowledge graphs; Calculate the similarity between two entities based on boundary similarity and output the predicted alignment result.
2. The method for aligning traffic data asset entities according to claim 1, characterized in that: The embedding representation learning of entities, relations and attributes includes: Call the BERT model to calculate the summary embedding of the entity and generate a word vector representing the main meaning of the entity; Learn relation embedding based on TransE and Graph Embedding algorithms and preserve the relation structure in the knowledge graph; Use attribute embeddings to capture holistic features that match entities or attributes.
3. The method for aligning traffic data asset entities according to claim 2, characterized in that: The entity summary embedding is learned using three different scoring functions, and the calculation formula of the scoring function is as follows: f s =||h sum +r-t sum || l1 / l2 , f st =||h sum +r-t|| l1 / l2 , f hs =||h+r-t sum || l1 / l2 , Among them, h sum and t sum are the entity summary head and entity summary tail, h and t are the head entity and tail entity, r is the relationship between h and t, l1 / l2 represents the L1 / L2 norm, the L1 norm is the Manhattan norm, and the L2 norm is the Euclidean norm.
4. The method for aligning traffic data asset entities according to claim 3, characterized in that: The scoring function is applicable to the zero-shot scenario, and the summary vector generated by BERT is used for summary-based representation. The calculation formula for summary-based entity embedding learning is as follows: L SUM =f s +f st +f hs , Among them, L SUM is the loss function based on the summary entity representation, and the goal is to minimize L SUM .
5. The method for aligning traffic data asset entities according to claim 4, characterized in that: The TransE and Graph Embedding algorithms are used to learn relational embedding and retain the relational structure in the knowledge graph, including: Based on relation embedding learning, the relation structure is preserved by minimizing the objective function, the objective function includes the scoring function of the valid triples and the negative triples, and the calculation formula of the minimization objective function based on relation embedding is as follows: in, γ is the margin, f r (h, t) and f r (h', t') represent the scoring functions of valid triples and negative triples respectively, T r is the set of valid relation triples, T' r is the set of damaged relation triples, f r is the likelihood function of the relation triple, count(r) is the number of occurrences of relation r, and |T| is the total number of triples in the merged knowledge graph; The optimization is performed based on the likelihood functions of valid relation triplets and damaged relation triplets.
6. The method for aligning traffic data asset entities according to any one of claims 1 to 5, characterized in that: The weighted average technique is used to calculate the overall scores of entities in two knowledge graphs, including: Compute the average embedding among entity summary embedding, relation embedding, and attribute embedding, and calculate the deviation of each embedding from the average, as follows: in, is the average value of embeddings, M is the number of embeddings, e (i) is the embedding of each component, ω i is the weight; The entity summary embedding, relation embedding, and attribute embedding are linearly combined, and weights are assigned to each entity embedding item. The calculation formula is as follows: Among them, e cmb Represents a combined embedding.
7. The method for aligning traffic data asset entities according to claim 6, characterized in that: The calculating the similarity between two entities based on boundary similarity includes: Calculate the similarity between the query entity and the target entity and find the aligned entity pair. The calculation formula is as follows: Among them, e1∈KG1 is the query entity, e2∈KG2 is the target entity; Sort the ranks of the target entities that are actually aligned to the top of the rank list.
8. A traffic data asset entity alignment device, characterized in that: include: The alignment module is used to merge and align two knowledge graphs based on the same predicate alignment algorithm; Learning module, used to learn embedding representations of entities, relations, and attributes; The calling module is used to call the BERT model to calculate the summary embedding, relationship embedding, and attribute embedding of entities in the two knowledge graphs; A calculation module, used to calculate the overall scores of entities in two knowledge graphs using a weighted average technique; The output module is used to calculate the similarity between two entities based on boundary similarity and output the predicted alignment result.
9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for aligning traffic data asset entities according to any one of claims 1 to 7 are implemented.
10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method for aligning traffic data asset entities according to any one of claims 1 to 7 are implemented.