Conceptio › Archive › Google Patents

Google Patentsopen access

Method for training multivariate relationship generation model, electronic … — Beijing Baidu Netcom Science And Technology Co., Ltd. (US20220004892A1)

Beijing Baidu Netcom Science And Technology Co., Ltd. · Google Patents

Google Patents · Patents · License: Open Access

patent, google patents, intellectual property, US20220004892A1, Beijing Baidu Netcom Science And Technology Co., Ltd., Quan Wang, en, 2022

ABSTRACT

Abstract

A method for training a multivariate relationship generation model, an electronic device and a medium are provided. The technical solution includes: obtaining a plurality of knowledge text entries; performing semantic parsing on each knowledge text entry to obtain a plurality of entities and semantic information of each knowledge text entry; constructing a heterogeneous graph based on the plurality of entities and the semantic information; and training an initial artificial intelligence (AI) network model based on the heterogeneous graph to obtain a multivariate relationship generation model.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS

This application claims the benefit of priority to Chinese Application No. 202011520491.4, filed on Dec. 21, 2020, the contents of which are incorporated herein by reference in their entirety.

TECHNICAL FIELD

The present disclosure relates to a field of computer technology, specifically to a field of artificial intelligence (AI) technologies such as natural language processing(NLP), deep learning (DL), knowledge graph (KG), etc., and particularly to a method for training a multivariate relationship generation model, an electronic device and a medium.

BACKGROUND

Artificial intelligence (AI) is a discipline that studies computers to simulate certain thinking processes and intelligent behaviors (such as learning, reasoning, thinking, planning, etc.) of human beings, which covers hardware-level technologies and software-level technologies.

AI hardware technologies generally include technologies such as sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, etc.; AI software technologies mainly include computer vision technology, speech recognition technology, natural language processing (NLP) technology and machine learning (ML)/deep learning (DL), big data processing technology, knowledge graph (KG) technology, etc.

The knowledge graph is a structured knowledge base composed of entities and relations, which is configured to describe world, domain and language knowledge. Generally, the KG generally relates to a binary relationship between entities, and represents knowledge entries by a triple in the form of subject, predicate, object), for example, (Mary Curie, gained award, Nobel Prize in Physics).

SUMMARY

A method for training a multivariate relationship generation model, an electronic device, a storage medium and a computer program product are provided.

According to a first aspect, a method for training a multivariate relationship generation model is provided. The method includes: obtaining a plurality of knowledge text entries; performing semantic parsing on each knowledge text entry to obtain a plurality of entities and semantic information of each knowledge text entry; constructing a heterogeneous graph based on the plurality of entities and the semantic information; and training an initial artificial intelligence (AI) network model based on the heterogeneous graph to obtain a multivariate relationship generation model.

According to a second aspect, an electronic device is provided. The electronic device includes at least one processor; and a memory communicatively coupled to the at least one processor. The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor, so that the at least one processor can execute the method for training a multivariate relationship generation model according to embodiments of the present disclosure.

According to a third aspect, a non-transitory computer-readable storage medium storing computer instructions is provided. The computer instructions are configured to cause a computer to execute a method for training a multivariate relationship generation model in embodiments of the present disclosure.

It should be understood that, the content described in the part is not intended to recognize key or important features of embodiments of the present disclosure, nor intended to limit the scope of the present disclosure. Other features of the present disclosure will be easy to understand through the following specification.

BRIEF DESCRIPTION OF THE DRAWINGS

The drawings are intended to better understand the solution, and do not constitute a limitation to the disclosure.

FIG. 1 is a flow chart illustrating a method for training a multivariate relationship generation model according to a first embodiment of the present disclosure;

FIG. 2 is a schematic diagram illustrating a heterogeneous graph according to an embodiment of the present disclosure;

FIG. 3 is a flow chart illustrating a method for training a multivariate relationship generation model according to a second embodiment of the present disclosure;

FIG. 4 is a schematic diagram illustrating an initial artificial intelligence (AI) network model according to an embodiment of the present disclosure;

FIG. 5 is a block diagram illustrating an apparatus for training a multivariate relationship generation model according to a third embodiment of the present disclosure;

FIG. 6 is a block diagram illustrating an apparatus for training a multivariate relationship generation model according to a fourth embodiment of the present disclosure;

FIG. 7 is a block diagram illustrating an electronic device configured to implement a method for training a multivariate relationship generation model according to an embodiment of the present disclosure.

DETAILED DESCRIPTION

The exemplary embodiments of the present disclosure are described as below with reference to the accompanying drawings, which include various details of embodiments of the present disclosure to facilitate understanding, and should be considered as merely exemplary. Therefore, those skilled in the art should realize that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for clarity and conciseness, descriptions of well-known functions and structures are omitted in the following descriptions.

FIG. 1 is a flow chart illustrating a method for training a multivariate relationship generation model according to a first embodiment of the present disclosure.

It should be noted that, an execution subject of the method for training a multivariate relationship generation model in the embodiment may be an apparatus for training a multivariate relationship generation model, and the apparatus may be implemented by means of software and/or hardware and may be configured in an electronic device, and the electronic device may include but is not limited to a terminal, a server, etc.

The embodiments of the present disclosure relate to a field of artificial intelligence (AI) technologies such as natural language processing (NLP), deep learning (DL), knowledge graph (KG), etc.

Artificial Intelligence, abbreviated as AI in English, is a new science of technology that studies and develops theories, methods, technologies and application systems configured to simulate, extend and expand human intelligence.

Deep learning (DL) learns inherent law and representation hierarchy of sample data, and information obtained in the learning process is of great help in interpretation of data such as words, images and sounds. The final goal of DL is that the machine may have analytical learning ability like human beings, which may recognize data such as words, images, sounds, etc,

Natural language processing (NLP) studies all kinds of theories and methods that may achieve effective communication between human and computer by natural language.

Knowledge graph (KG) is a modern theory that combines theories and methods of applied mathematics, graphics, information visualization technology, information science and other disciplines with metrological citation analysis, co-occurrence analysis and other methods, and uses visual graphs to vividly display the core structure, development history, frontiers, and overall knowledge structure of the discipline to achieve multi-disciplinary integration.

As illustrated in FIG. 1 , the method for training a multivariate relationship generation model includes the following.

At S 101 , a plurality of knowledge text entries are obtained.

The knowledge text entry such as âMary Curie won the Nobel Prize in Physics together with Pierre Curie and Henry Becquerel in 1903â is a knowledge text entry represented by a quintuple relationship. The knowledge text entry may also relate to a relationship representation of any number of variants, for example, a ternary relationship representation, a quaternary relationship representation, a sextuple relationship representation, etc., which will not be limited here.

The relationship representation in each knowledge text entry mentioned above in the embodiments of the present disclosure may relate to any number of variants, so that when a plurality of knowledge text entries are taken as training data, feature dimensions of the training data may be effectively expanded, and an accuracy of model training may be guaranteed.

At S 102 , semantic parsing is performed on each knowledge text entry to obtain a plurality of entities and semantic information of each knowledge text entry.

After a plurality of knowledge text entries are obtained, semantic analysis may be performed on each knowledge text entry, to obtain respective entities and semantic information of the knowledge text entry.

Taking the above knowledge text entry âMarie Curie won the Nobel Prize in Physics together with Pierre Curie and Henry Becquerel in 1903â as an example, the entities may include, for example, âMary Curieâ and âGained Awardâ âAward-Gaining Timeâ âCo-winnerâ âCo-winnerâ, and correspondingly, the semantic information may be configured to describe the semantics corresponding to the knowledge text entry, or may also be configured to describe actual semantic content of the each entity, for example, â1903â, âNobel Prize in Physicsâ, âPierre Curieâ, âHenri Becquerelâ, etc., which will not be limited here.

In some embodiments, semantic parsing may be performed on the knowledge text entry to obtain the corresponding plurality of entities and semantic information as follows. Semantic parsing may be performed on the knowledge text entry to obtain a plurality of entities in the knowledge text entry; a prime attribute and a secondary attribute are determined from the plurality of entities; prime content corresponding to the prime attribute and secondary content corresponding to the secondary attribute are determined; and the prime content and secondary content are taken as the semantic information.

Taking the knowledge text entry âMarie Curie won the Nobel Prize in Physics with Pierre Curie and Henry Becquerel in 1903â as an example, the entities may include, for example, âMary Curieâ and âGained Awardâ âAward-Gaining Timeâ âCo-winnerâ âCo-winnerâ, and correspondingly, the prime attribute may be âGained Awardâ, and the secondary attributes may include âAward-Gaining Timeâ âCo-winnerâ âCo-winnerâ. Correspondingly, âNobel Prize in Physicsâ corresponding to the prime attribute âGained Awardâ may be referred to as the prime content. â1903â corresponding to the secondary attribute âAward-Gaining Timeâ, and âPierre Curieâ and âHenry Becquerelâ corresponding to the secondary attribute âCo-winnerâ may be referred to as the secondary content.

That is, in embodiments of the present disclosure, each multivariate knowledge text entry may be represented by a combination of a core triple (the core triple includes at least part of entities, and contents corresponding to the entities) and a secondary attribute-value pair, specifically, for example, the quintuple knowledge text entry âMarie Curie won the Nobel Prize in Physics with Pierre Curie and Henry Becquerel in 1903â may be represented by:

(Mary Curie, Gained Award, Nobel Prize in Physics)

Award-Gaining Time: 1903

Co-winner: Pierre Curie

Co-winner: Henry Becquerel

(Mary Curie, Gained Award, Nobel Prize in Physics) is a core triple, and âAward-Gaining Time: 1903â, âCo-winner: Pierre Coriâ, âCo-winner: Henry Becquerelâ are secondary attribute-value pairs supplementary to the core triple.

In the embodiments of the present disclosure, Mary Curie and Gained Award in the core triple may be abstracted as corresponding entities, and Mary Curie and Gained Award in the triple may be taken as the prime attribute which may be combined with the secondary attribute, or any other possible combinations are feasible, which will not be limited here.

Semantic parsing may be performed on the knowledge text entry to obtain a plurality of entities in the knowledge text entry; the prime attribute and the secondary attribute are determined from the plurality of entities; the prime content corresponding to the prime attribute and the secondary content corresponding to the secondary attribute are determined; and the prime content and secondary content are taken as the semantic information, thereby effectively expanding feature dimensions of the training data, and guaranteeing an accuracy of model training, and causing the trained multivariate relationship generation model to learn semantic integrity of the multivariate knowledge entries to a greater degree.

In other embodiments, semantic parsing may be performed on the knowledge text entry in other possible ways to obtain the corresponding plurality of entities and semantic information, for example, mathematical way, engineering way, which will not be limited here.

At S 103 , a heterogeneous graph is constructed based on the plurality of entities and the semantic information.

After semantic parsing is performed on the knowledge text entries to ob

CROSS-REFERENCE TO RELATED APPLICATIONS

This application claims the benefit of priority to Chinese Application No. 202011520491.4, filed on Dec. 21, 2020, the contents of which are incorporated herein by reference in their entirety.

TECHNICAL FIELD

BACKGROUND

SUMMARY

A method for training a multivariate relationship generation model, an electronic device, a storage medium and a computer program product are provided.

BRIEF DESCRIPTION OF THE DRAWINGS

The drawings are intended to better understand the solution, and do not constitute a limitation to the disclosure.

FIG. 1 is a flow chart illustrating a method for training a multivariate relationship generation model according to a first embodiment of the present disclosure;

FIG. 2 is a schematic diagram illustrating a heterogeneous graph according to an embodiment of the present disclosure;

FIG. 3 is a flow chart illustrating a method for training a multivariate relationship generation model according to a second embodiment of the present disclosure;

FIG. 4 is a schematic diagram illustrating an initial artificial intelligence (AI) network model according to an embodiment of the present disclosure;

FIG. 5 is a block diagram illustrating an apparatus for training a multivariate relationship generation model according to a third embodiment of the present disclosure;

FIG. 6 is a block diagram illustrating an apparatus for training a multivariate relationship generation model according to a fourth embodiment of the present disclosure;

DETAILED DESCRIPTION

FIG. 1 is a flow chart illustrating a method for training a multivariate relationship generation model according to a first embodiment of the present disclosure.

The embodiments of the present disclosure relate to a field of artificial intelligence (AI) technologies such as natural language processing (NLP), deep learning (DL), knowledge graph (KG), etc.

Natural language processing (NLP) studies all kinds of theories and methods that may achieve effective communication between human and computer by natural language.

As illustrated in FIG. 1 , the method for training a multivariate relationship generation model includes the following.

At S 101 , a plurality of knowledge text entries are obtained.

At S 102 , semantic parsing is performed on each knowledge text entry to obtain a plurality of entities and semantic information of each knowledge text entry.

(Mary Curie, Gained Award, Nobel Prize in Physics)

Award-Gaining Time: 1903

Co-winner: Pierre Curie

Co-winner: Henry Becquerel

At S 103 , a heterogeneous graph is constructed based on the plurality of entities and the semantic information.

After semantic parsing is performed on the knowledge text entries to obtain the corresponding plurality of entities and semantic information, and a heterogeneous graph may be constructed according to the entities and the semantic information.

The heterogeneous graph is a kind of graph model, and the graph model may be a graph model in deep learning, or a graph model of any other possible architectural form in the field of AI technologies, which is not limited here. More than one type of nodes and edges may exist in the heterogeneous graph, and different types of nodes are allowed to have features or attributes in different dimensions, that is, the heterogeneous graph is constructed based on entities and semantic information, and the types of nodes in the heterogeneous graph may be the same or different, and the heterogeneous graph may carry features or attributes in different dimensions associated with the knowledge text entries.

In some embodiments, the heterogeneous graph is constructed based on entities and semantic information as follows. The plurality of entities, the prime attribute, the secondary attribute, the prime content and the secondary content may be mapped to nodes; parts of speech of the plurality of entities, the prime attribute, the secondary attribute, the prime content and the secondary content are used to describe parts of speech of the nodes; an edge is constructed between nodes having a corresponding relationship; a part-of-speech type of each edge is formed based on the parts of speech of the nodes connected by the edge; and each edge is labeled with the part-of-speech type to construct the heterogeneous graph based on the nodes, the edges, and edge labels.

Whether two nodes have a corresponding relationship may be determined based on the semantic information or the part of speech of each node. For example, the corresponding relationship may refer to a semantic association relation, a part-of-speech association relation. When two nodes are determined to be associated in terms of semantic, it may be determine that the two nodes have a corresponding relationship. When two nodes are determined to be associated in terms of part of speech, it may be determine that the two nodes have a corresponding relationship.

The parts of speech may include, for example, subject, predicate, and object. An edge may connect two nodes. If the parts of speech corresponding to two nodes are subject and predicate, the pan-of-speech type of the corresponding edge is subject-predicate type.

For example, each multivariate knowledge text entry may be represented by a combination of a core triple (s, r, o) and secondary attribute-value pairs (a i , v i ), that is, ((s, r, o), {(a i ,v i )} i=1 m ), where m is the number of the secondary attribute-value pairs. In the embodiment of the present disclosure, a multivariate knowledge text entry such as ((s, r, o), {(a i , v i )} i=1 m ) may be represented by a heterogeneous graph. As illustrated in FIG. 2 , FIG. 2 is a schematic diagram illustrating a heterogeneous graph according to an embodiment of the present disclosure. In FIG. 2 , different shapes of edges correspond to different part-of-speech types, Îµ is a set of nodes in the graph, containing all entities (values) and relationships (attributes) in the corresponding multivariate knowledge text entry, that is, Îµ={r, s, o, a 1 , . . . , a m , v 1 , . . . , v m }. L is a set of edges in the graph, which contains four part-of-speech types, (2m+2) edges in total, that is:

subject-predicate edge (s, r), one in total;

object-predicate edge (a, r), one in total;

predicate-attribute edge (r, a i ), m in total;

attribute-value edge (a i , v i ), m in total.

In the heterogeneous graph in the embodiment of the present disclosure, the nodes and edges having corresponding part-of-speech types are included. Nodes relate to two types, i.e., entities and relationships (relationships include the prime attribute, the secondary attribute, the prime content, and the secondary content), and the part-of-speech types of the edges may be subject-predicate, object-predicate, predicate-attribute, and attribute-value. In the embodiment of the present disclosure, by constructing the heterogeneous graph based on entities and semantic information, the integrity of the knowledge entries may be retained to the greatest extent without losing semantic information, so that the heterogeneous graph completely retains the semantic information of the multivariate knowledge text entries, which may effectively assist in improving a modeling effect of the multivariate relationship generation model.

At S 104 , an initial artificial intelligence (AI) network model is trained based on the heterogeneous graph to obtain a multivariate relationship generation model.

After the heterogeneous graph is constructed based on entities and semantic information, an initial AI network model is trained according to the heterogeneous graph to obtain the multivariate relationship generation model, for example, each node and edge in the heterogeneous graph, as well as features or attributes in multiple dimensions carried by the nodes and edges may be input to the initial AI network model till the AI network model converges, and the trained AI network model is taken as the multivariate relationship generation model.

The AI network model may be, for example, a neural network model, a machine learning (ML) model, etc. In the embodiment of the present disclosure, a graph neural network model may be configured to train and obtain the multivariate relationship generation model, and a graph neural network technology is configured to model the heterogeneous graph, which may fully capture potential relations between various variants (entity, prime attribute, secondary attribute and corresponding contents) in the multivariate knowledge text entry.

FIG. 3 is a flow chart illustrating a method for training a multivariate relationship generation model according to a second embodiment of the present disclosure.

As illustrated in FIG. 3 , the method for training a multivariate relationship generation model includes the following.

At S 301 , a plurality of knowledge text entries are obtained.

At S 302 , semantic parsing is performed on each knowledge text entry to obtain a plurality of entities and semantic information.

At S 303 , a heterogeneous graph is constructed based on the plurality of entities and the semantic information.

The explanation of S 301 to S 303 may refer to the above embodiments, which will not be repeated here.

At S 304 , the heterogeneous graph is input to an input representation layer to obtain a vector representation corresponding to the nodes output by the input representation layer, the vector representation describing a semantic relationship among contents corresponding to the nodes.

That is, the initial AI network model in the embodiment of the present disclosure includes the input representation layer, a graph attention layer (there may be multiple graph attention layers, such as L layers, and the graph attention layers are arranged in sequence), and a linear prediction layer.

FIG. 4 is a schematic diagram illustrating an initial AI network model according to an embodiment of the present disclosure, which includes the input representation layer, L graph attention layers and the linear prediction layer.

After the heterogeneous graph is constructed based on entities and semantic information, the heterogeneous graph may be input to the input representation layer to obtain the vector representation corresponding to the nodes output by the input representation layer, so that the vector representation may describe the semantic relationship among contents corresponding to the nodes, thereby effectively integrating the heterogeneous graph and the AI network model, and causing the AI network model to quickly learn the semantic relationship among the contents corresponding to the nodes.

At S 305 , the heterogeneous graph is input to the graph attention layer to obtain an adaptation factor corresponding to the part-of-speech types of the edges output by the graph attention layer.

After the heterogeneous graph is input to the input representation layer to obtain the vector representation corresponding to the nodes output by the input representation layer, the vector representation output by the input representation layer may be further configured as labels of the nodes in the heterogeneous graph, and the labeled heterogeneous graph may be input to the graph attention layer to obtain the adaptation factor corresponding to the part-of-speech types of the edges output by the graph attention layer. The form of the adaptation factor may be a vector, and the adaptation factor may be configured to describe proportions of the part-of-speech types of its corresponding edges in the entire knowledge text entry.

After the heterogeneous graph is input to the graph attention layer to obtain the adaptation factor corresponding to the part-of-speech types of the edges output by the graph attention layer an attention intensity value of the graph attention layer may be further adjusted according to the adaptation factor, thereby forming a more accurate edge perception based self-attention mechanism to quickly and accurately model the heterogeneity of a graph.

At S 306 , a predictive multivariate relationship output by the linear prediction layer is obtained by inputting the vector representation and the adaptation factor to the linear prediction layer.

After the heterogeneous graph is input to the input representation layer to obtain the vector representation corresponding to the nodes output by the input representation layer, and the heterogeneous graph is input to the graph attention layer to obtain the adaptation factor corresponding to the part-of-speech types of the edges output by the graph attention layer, the predictive multivariate relationship output by the linear prediction layer may be obtained by inputting the vector representation and the adaptation factor to the linear prediction layer.

That is, in the training process, a plurality of knowledge text entries are configured to train the initial AI network model, that is, the initial AI network model is configured to predict a multivariate relationship for each knowledge text entry, and the multivariate relationship obtained by prediction may be referred to as the predictive multivariate relationship, and then the predictive multivariate relationship is judged accordingly till it is determined that the AI network model converges according to the predictive multivariate relationship.

For example, the multivariate knowledge reasoning is transformed into a node prediction problem on a heterogeneous graph, taking predicting a subject s in the multivariate knowledge entry ((s, r, o), {(a i , v i )} i=1 m ) for an example (the multivariate relationship obtained by prediction may be referred to as the predictive multivariate relationship), the input of the AI network model is a heterogeneous graph ((?, r, a), {(a i , v i )} i=1 m ) with missing nodes, and the output is a predictive multivariate relationship (including a predicted target node subject s).

At S 307 , the initial AI network model is trained based on the predictive multivariate relationship and labeled multivariate relationship to obtain the multivariate relationship generation model.

In some embodiments, a multivariate relationship may be labeled for each knowledge text entry in advance.

In some embodiments, the initial AI network model is trained based on the predictive multivariate relationship and the labeled multivariate relationship to obtain the multivariate relationship generation model as follows. A loss value between the predictive multivariate relationship and the labeled multivariate relationship is determined; if the loss value satisfies a preset loss threshold, the trained AI network model is taken as the multivariate relationship generation model, thereby achieving a better model training effect.

The loss value may be calculated from a loss function corresponding to the initial AI network model. In applications, the loss function is usually associated with an optimization problem as a learning criterion, that is, the model is solved and evaluated by minimizing the loss function, therefore, in the embodiment, the loss value between the predictive multivariate relationship and the labeled multivariate relationship may be determined to guide the training process.

After the loss value between the predictive multivariate relationship and the labeled multivariate relationship is determined, it may be determined in real time whether the loss value satisfies a preset loss threshold (the preset loss threshold may be pre-calibrated threshold of the loss value for determining whether the AI network model converges). If the loss value satisfies the preset loss threshold, the trained AI network model is taken as the multivariate relationship generation model, that is, the model training is completed, and the multivariate relationship generation model at this time satisfies a preset convergence condition.

In the embodiment, by obtaining a plurality of knowledge text entries, performing semantic parsing on each knowledge text entry to obtain a plurality of entities and semantic information, constructing a heterogeneous graph according to the entities and the semantic information, and training an initial artificial intelligence (AI) network model according to the heterogeneous graph to obtain a multivariate relationship generation model, the semantic integrity of the multivariate knowledge entries may be learned to a greater degree, thereby effectively improving an accuracy of a multivariate relationship representation of the multivariate relationship generation model, and achieving more accurate multivariate knowledge reasoning. After a heterogeneous graph is constructed based on entities and semantic information, the heterogeneous graph may be input to an input representation layer to obtain a vector representation corresponding to the nodes output by the input representation layer, the heterogeneous graph is input to a graph attention layer to obtain an adaptation factor corresponding to the part-of-speech types of the edges output by the graph attention layer, and a predictive multivariate relationship output by the linear prediction layer may be obtained by inputting the vector representation and the adaptation factor to the linear prediction layer, so that the vector representation may describe a semantic relationship among contents corresponding to the nodes, thereby effectively integrating the heterogeneous graph and the A 1 network model, causing the AI network model to quickly learn the semantic relationship among the contents corresponding to the nodes. After the heterogeneous graph is input to a graph attention layer to obtain an adaptation factor corresponding to the part-of-speech types of the edges output by the graph attention layer, an attention intensity value of the graph attention layer may be further adjusted according to the adaptation factor, thereby forming a more accurate self-attention mechanism of edge perception to quickly and accurately model the heterogeneity of a graph.

FIG. 5 is a block diagram illustrating an apparatus for training a multivariate relationship generation model according to a third embodiment of the present disclosure.

As illustrated in FIG. 5 , the apparatus 50 for training a multivariate relationship generation model includes: an obtaining module 501 , configured to obtain a plurality of knowledge text entries a parsing module 502 , configured to perform semantic parsing on each knowledge text entry to obtain a plurality of entities and semantic information of each knowledge text entry; a constructing module 503 , configured to construct a heterogeneous graph based on the plurality of entities and the semantic information; and a training module 504 , configured to train an initial artificial intelligence (AI) network model based on the heterogeneous graph to obtain a multivariate relationship generation model.

In some embodiments, the parsing module 502 is configured to: perform semantic parsing on a knowledge text entry to obtain the plurality of entities in the knowledge text entry; determine a prime attribute and a secondary attribute from the plurality of entities in the knowledge text entry; determine prime content corresponding to the prime attribute, and secondary content corresponding to the secondary attribute; and take the prime content and the secondary content as the semantic information of the knowledge text entry.

In some embodiments of the present disclosure, as illustrated in FIG. 6 , FIG. 6 is a block diagram illustrating an apparatus for training a multivariate relationship generation model according to a fourth embodiment of the present disclosure. The apparatus 60 for training a multivariate relationship generation model includes an obtaining module 601 , a parsing module 602 , a constructing module 603 , and a training module 604 , in which the constructing module 603 includes: a mapping sub-module 6031 , configured to map the plurality of entities, the prime attribute, the secondary attribute, the prime content and the secondary content to nodes, and describe parts of speech of the nodes based on parts of speech of the plurality of entities, the prime attribute, the secondary attribute, the prime content and the secondary content; a first constructing sub-module 6032 , configured to construct an edge between nodes having a corresponding relationship; a generating sub-module 6033 , configured to f form a part-of-speech type of each edge based on the parts of speech of the nodes connected by the edge; and a second constructing sub-module 6034 , configured to label each edge with the pan-of-speech type to construct the heterogeneous graph based on the nodes, the edges, and edge labels.

In some embodiments of the present disclosure, the initial artificial intelligence (AI) network model includes an input representation layer, a graph attention layer and a linear prediction layer. The training module 604 is configured to: input the heterogeneous graph to the input representation layer to obtain a vector representation corresponding to the nodes output by the input representation layer, the vector representation describing a semantic relationship among contents corresponding to the nodes; input the heterogeneous graph to the graph attention layer to obtain an adaptation factor corresponding to the part-of-speech types of the edges output by the graph attention layer; input the vector representation and the adaptation factor to the linear prediction layer to obtain a predictive multivariate relationship output by the linear prediction layer; and train the initial AI network model based on the predictive multivariate relationship and labeled multivariate relationship to obtain the multivariate relationship generation model.

In some embodiments, a training module 604 is configured to: adjust an attention intensity value of the graph attention layer based on the adaptation factor.

In some embodiments, a training module 604 is configured to: determine a loss value between the predictive multivariate relationship and the labeled multivariate relationship; if the loss value satisfies a preset loss threshold, taking the trained AI network model as the multivariate relationship generation model.

It is understandable that, the apparatus 60 for training a multivariate relationship generation model in FIG. 6 of the embodiment and the apparatus 50 for training a multivariate relationship generation model in the above embodiment, the obtaining module 601 and the obtaining module 501 in the above embodiment, the parsing module 602 and the parsing module 502 in the above embodiment, the constructing module 603 and the constructing module 503 in the above embodiment, the training module 604 and the training module 504 in the above embodiment, have the same functions and structures.

It needs to be noted that the foregoing explanation of the method for training a multivariate relationship generation model is also applied to an apparatus for training a multivariate relationship generation model in this embodiment, which will not be repeated here.

In the embodiment of the present disclosure, an electronic device, a readable storage medium and a computer program product are further provided according to embodiments of the present disclosure.

FIG. 7 is a block diagram illustrating an electronic device configured to implement a method for training a multivariate relationship generation model in the embodiment of the present disclosure. An electronic device is intended to represent various types of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. An electronic device may also represent various types of mobile apparatuses, such as personal digital assistants, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relations, and their functions are merely examples, and are not intended to limit the implementation of the disclosure described and/or required herein.

As illustrated in FIG. 7 , a device 700 includes a computing unit 701 , configured to execute various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 702 or loaded from a memory unit 708 to a random access memory (RAM) 703 . In a RAM 703 , various programs and data required for a device 700 may be stored. A computing unit 701 , a ROM 702 and a RAM 703 may be connected with each other by a bus 704 . An input/output(I/O) interface 705 is also connected to a bus 704 .

A plurality of components in the device 700 are connected to an I/ O interface 705 , and includes: an input unit 706 , for example, a keyboard, a mouse, etc.; an output unit 707 , for example various types of displays, speakers; a memory unit 708 , for example a magnetic disk, an optical disk; and a communications unit 709 , for example, a network card, a modem, a wireless transceiver. A communications unit 709 allows a device 700 to exchange information/data through a computer network such as internet and/or various types of telecommunication networks and other devices.

A computing unit 701 may be various types of general and/or dedicated processing components with processing and computing ability. Some examples of a computing unit 701 include but not limited to a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running a machine learning model algorithm, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. A computing unit 701 executes various methods and processes as described above, for example, a method for training a multivariate relationship generation model.

For example, in some embodiments, a method for training a multivariate relationship generation model may be further implemented as a computer software program, which is physically contained in a machine readable medium, such as a memory unit 708 . In some embodiments, a part or all of the computer program may be loaded and/or installed on the device 700 through a ROM 702 and/or a communication units 709 . When the computer program is loaded on a RAM 703 and executed by a computing unit 701 , one or more blocks in the method for training a multivariate relationship generation model as described above may be performed. Alternatively, in other embodiments, a computing unit 701 may be configured to execute a method for training a multivariate relationship generation model in other appropriate methods (for example, by virtue of a firmware).

Various implementation modes of systems and technologies described herein may be implemented in a digital electronic circuit system, an integrated circuit system, a field programmable gate array(FPGA), a dedicated application specific integrated circuit (ASIC), a system on a chip (SoC), a load programmable logic device (CPLD), a computer hardware, a firmware, a software, and/or combinations thereof. The various implementation modes may include: being implemented in one or more computer programs, and the one or more computer programs may be executed and/or interpreted on a programmable system including at least one programmable processor, and the programmable processor may be a dedicated or a general-purpose programmable processor that may receive data and instructions from a storage system, at least one input apparatus, and at least one output apparatus, and transmit the data and instructions to the storage system, the at least one input apparatus, and the at least one output apparatus.

A computer code configured to execute a method for training a multivariate relationship generation model in the present disclosure may be written with one or any combination of multiple programming languages. These programming languages may be provided to a processor or a controller of a general purpose computer, a dedicated computer, or other apparatuses for programmable data processing so that the function/operation specified in the flowchart and/or block diagram may be performed when the program code is executed by the processor or controller. A computer code may be executed completely or partly on the machine, executed partly on the machine as an

CLAIMS

Claims ( 18 )

What is claimed is:

1 . A method for training a multivariate relationship generation model, comprising:

obtaining a plurality of knowledge text entries; performing semantic parsing on each knowledge text entry to obtain a plurality of entities and semantic information of each knowledge text entry; constructing a heterogeneous graph based on the plurality of entities and the semantic information; and training an initial artificial intelligence (AI) network model based on the heterogeneous graph to obtain a multivariate relationship generation model.

2 . The method of claim 1 , wherein, performing semantic parsing on each knowledge text entry to obtain a plurality of entities and semantic information each knowledge text entry comprises:

performing semantic parsing on a knowledge text entry to obtain the plurality of entities in the knowledge text entry; determining a prime attribute and a secondary attribute from the plurality of entities in the knowledge text entry; determining prime content corresponding to the prime attribute, and secondary content corresponding to the secondary attribute; and taking the prime content and the secondary content as the semantic information of the knowledge text entry.

3 . The method of claim 2 , wherein, constructing a heterogeneous graph based on the plurality of entities and the semantic information comprises:

mapping the plurality of entities, the prime attribute, the secondary attribute, the prime content and the secondary content to nodes; describing parts of speech of the nodes based on parts of speech of the plurality of entities, the prime attribute, the secondary attribute, the prime content and the secondary content; constructing an edge between nodes having a corresponding relationship; forming a part-of-speech type of each edge based on the parts of speech of the nodes connected by the edge; and labeling each edge with the part-of-speech type to construct the heterogeneous graph based on the nodes, the edges, and edge labels.

4 . The method of claim 3 , wherein the initial AI network model comprises an input representation layer, a graph attention layer and a linear prediction layer, wherein, training the initial AI network model based on the heterogeneous graph to obtain a multivariate relationship generation model comprises:

inputting the heterogeneous graph to the input representation layer to obtain a vector representation corresponding to the nodes output by the input representation layer, the vector representation describing a semantic relationship among contents corresponding to the nodes; inputting the heterogeneous graph to the graph attention layer to obtain an adaptation factor corresponding to the part-of-speech types of the edges output by the graph attention layer; inputting the vector representation and the adaptation factor to the linear prediction layer to obtain a predictive multivariate relationship output by the linear prediction layer; and training the initial AI network model based on the predictive multivariate relationship and pre-labeled multivariate relationship to obtain the multivariate relationship generation model.

5 . The method of claim 4 , after inputting the heterogeneous graph to the graph attention layer to obtain the adaptation factor corresponding to the part-of-speech types of the edges output by the graph attention layer, further comprising:

adjusting an attention intensity value of the graph attention layer based on the adaptation factor.

6 . The method of claim 4 , wherein, training the initial AI network model based on the predictive multivariate relationship and the pre-labeled multivariate relationship to obtain the multivariate relationship generation model comprises:

determining a loss value between the predictive multivariate relationship and the pre-labeled multivariate relationship; if the loss value satisfies a preset loss threshold, taking the trained AI network model as the multivariate relationship generation model.

7 . An electronic device, comprising:

at least one processor; and a memory communicatively coupled to the at least one processor; wherein, the memory is stored with instructions executable by the at least one processor, when the instructions are executed by the at least one processor, the at least one processor is caused to execute the method for training a multivariate relationship generation model, comprising: obtaining a plurality of knowledge text entries; performing semantic parsing on each knowledge text entry to obtain a plurality of entities and semantic information of each knowledge text entry; constructing a heterogeneous graph based on the plurality of entities and the semantic information; and training an initial artificial intelligence (AI) network model based on the heterogeneous graph to obtain a multivariate relationship generation model.

8 . The device of claim 7 , wherein, performing semantic parsing on each knowledge text entry to obtain a plurality of entities and semantic information each knowledge text entry comprises:

9 . The device of claim 8 , wherein, constructing a heterogeneous graph based on the plurality of entities and the semantic information comprises:

10 . The device of claim 9 , wherein the initial AI network model comprises an input representation layer, a graph attention layer and a linear prediction layer, wherein, training the initial AI network model based on the heterogeneous graph to obtain a multivariate relationship generation model comprises:

11 . The device of claim 10 , wherein after inputting the heterogeneous graph to the graph attention layer to obtain the adaptation factor corresponding to the part-of-speech types of the edges output by the graph attention layer, the at least one processor is further configured to perform:

adjusting an attention intensity value of the graph attention layer based on the adaptation factor.

12 . The device of claim 10 , wherein, training the initial AI network model based on the predictive multivariate relationship and the pre-labeled multivariate relationship to obtain the multivariate relationship generation model comprises:

13 . A non-transitory computer-readable storage medium having computer instructions stored thereon, wherein the computer instructions are configured to cause a computer to execute the method for training a multivariate relationship generation model, comprising:

14 . The storage medium of claim 13 , wherein, performing semantic parsing on each knowledge text entry to obtain a plurality of entities and semantic information each knowledge text entry comprises:

15 . The storage medium of claim 14 , wherein, constructing a heterogeneous graph based on the plurality of entities and the semantic information comprises:

16 . The storage medium of claim 15 , wherein the initial AI network model comprises an input representation layer, a graph attention layer and a linear prediction layer, wherein, training the initial AI network model based on the heterogeneous graph to obtain a multivariate relationship generation model comprises:

17 . The storage medium of claim 16 , wherein after inputting the heterogeneous graph to the graph attention layer to obtain the adaptation factor corresponding to the part-of-speech types of the edges output by the graph attention layer, the computer instructions are further configured to cause the computer to execute:

adjusting an attention intensity value of the graph attention layer based on the adaptation factor.

18 . The storage medium of claim 16 , wherein, training the initial AI network model based on the predictive multivariate relationship and the pre-labeled multivariate relationship to obtain the multivariate relationship generation model comprises:

US17/480,575

2020-12-21

2021-09-21

Method for training multivariate relationship generation model, electronic device and medium

Pending

US20220004892A1

( en )

Applications Claiming Priority (2)

Application Number

Priority Date

Filing Date

Title

CN202011520491.4A

CN112507040B

( en )

2020-12-21

Training method and device for multivariate relation generation model, electronic equipment and medium

CN202011520491.4

2020-12-21

Publications (1)

Publication Number

Publication Date

US20220004892A1

true

US20220004892A1 ( en )

2022-01-06

Family

ID=74922899

Family Applications (1)

Application Number

Title

Priority Date

Filing Date

US17/480,575

Pending

US20220004892A1

( en )

2020-12-21

2021-09-21

Method for training multivariate relationship generation model, electronic device and medium

Country Status (5)

Country

Link

( 1 )

US20220004892A1

( en )

( 1 )

EP3913543A3

( en )

( 1 )

JP2022028908A

( en )

( 1 )

KR20220005416A

( en )

( 1 )

CN112507040B

( en )

Cited By (10)

* Cited by examiner, â Cited by third party

Publication number

Priority date

Publication date

Assignee

Title

CN114694769A

( en )

2022-03-28

2022-07-01

åäº¬ç¾åº¦ç½è®¯ç§ææéå¬å¸

Molecular representation method, molecular representation model training method and device

CN114896974A

( en )

2022-04-21

2022-08-12

åäº¬ç¾åº¦ç½è®¯ç§ææéå¬å¸

Media information processing method, device, electronic equipment and storage medium

CN114942997A

( en )

2022-04-21

2022-08-26

é¿éå·´å·´ï¼ä¸å½ï¼æéå¬å¸

Data processing method, model training method, risk identification method, equipment and storage medium

US20230281467A1

( en )

2022-03-07

2023-09-07

Jpmorgan Chase Bank, N.A.

Systems and methods for building an architecture knowledge graph

CN116932013A

( en )

2023-09-15

2023-10-24

å¹¿å·å¸æ¡å¤å¨æ¢°è®¾å¤å·¥ç¨æéå¬å¸

Medical gas equipment operation system updating method and system based on artificial intelligence

CN116932737A

( en )

2023-07-25

2023-10-24

ä¸ç§å¤©çæ°æ®ç§æè¡ä»½æéå¬å¸

Character attribute extraction method based on big data

CN117274006A

( en )

2023-11-21

2023-12-22

å±±ä¸å±±å¤§é¸¥çè½¯ä»¶è¡ä»½æéå¬å¸

A proposition method, system, equipment and storage medium

CN118520879A

( en )

2024-05-23

2024-08-20

å¹¿å·å¸é«ç§éä¿¡ææ¯è¡ä»½æéå¬å¸

Semantic integrity judging method and system based on multi-head attention mechanism

US20240330679A1

( en )

2023-03-30

2024-10-03

Microsoft Technology Licensing, Llc

Heterogeneous tree graph neural network for label prediction

CN119358687A

( en )

2024-12-26

2025-01-24

åäº¬é¿äºç§ææéå¬å¸

A method and device for collaborative reasoning of multiple heterogeneous models

Families Citing this family (22)

* Cited by examiner, â Cited by third party

Publication number

Priority date

Publication date

Assignee

Title

CN113704482B

( en )

2021-04-14

2025-08-26

è¾è®¯ç§æï¼æ·±å³ï¼æéå¬å¸

A template determination method and related device for knowledge graph

US12459276B2

( en )

2021-04-21

2025-11-04

Ricoh Company, Ltd.

Contact member, drying apparatus, and printing apparatus for treating liquid-coated substrates

CN113221565A

( en )

2021-05-07

2021-08-06

åäº¬ç¾åº¦ç½è®¯ç§ææéå¬å¸

Entity recognition model training method and device, electronic equipment and storage medium

CN113642319B

( en )

2021-07-29

2022-11-29

åäº¬ç¾åº¦ç½è®¯ç§ææéå¬å¸

Text processing method and device, electronic equipment and storage medium

CN113792155B

( en )

2021-08-30

2022-12-09

åäº¬ç¾åº¦ç½è®¯ç§ææéå¬å¸

Text verification method and device based on knowledge graph, electronic equipment and medium

CN114234995A

( en )

2021-12-21

2022-03-25

æµæ±æ°æºäº¤é¢ç§æè¡ä»½æéå¬å¸

A navigation method, device, electronic device and storage medium

CN114329244B

( en )

2021-12-28

2025-08-01

åäº¬ç¾åº¦ç½è®¯ç§ææéå¬å¸

Map interest point query method, device, equipment, storage medium and program product

CN114595686B

( en )

2022-03-11

2023-02-03

åäº¬ç¾åº¦ç½è®¯ç§ææéå¬å¸

Knowledge extraction method, and training method and device of knowledge extraction model

CN114596574A

( en )

2022-03-22

2022-06-07

åäº¬ç¾åº¦ç½è®¯ç§ææéå¬å¸

Text recognition method, apparatus, electronic device and medium

CN114676694B

( en )

2022-03-25

2024-12-24

ä¸å½å·¥åé¶è¡è¡ä»½æéå¬å¸

Business model generation method, device, equipment, medium and program product

CN115146073B

( en )

2022-07-08

2024-06-21

åä¸å¸èå¤§å¦

A test question knowledge point marking method and its application for cross-spatial semantic knowledge injection

CN115423105A

( en )

2022-09-06

2022-12-02

ä¸å½äººæ°è§£æ¾ååäºç§å¦é¢ç³»ç»å·¥ç¨ç ç©¶é¢

Pre-training language model construction method, system and device

CN116484870B

( en )

2022-09-09

2024-01-05

åäº¬ç¾åº¦ç½è®¯ç§ææéå¬å¸

Method, device, equipment and medium for extracting text information

CN115759101A

( en )

2022-11-28

2023-03-07

ä¸è½¯éå¢è¡ä»½æéå¬å¸

Model training method, data processing device, storage medium and electronic equipment

CN116383442A

( en )

2022-12-28

2023-07-04

ä¸æµ·éèæè´§ä¿¡æ¯ææ¯æéå¬å¸

Intelligent analysis method and system based on multivariate data and atomic algorithm

CN116127319B

( en )

2023-01-16

2024-03-29

åäº¬ç¾åº¦ç½è®¯ç§ææéå¬å¸

Multi-mode negative sample construction and model pre-training method, device, equipment and medium

CN116308172A

( en )

2023-03-21

2023-06-23

ä¸å½åä¸é¶è¡è¡ä»½æéå¬å¸

Method, device, equipment and storage medium for determining system entry

CN116564408B

( en )

2023-04-28

2024-03-01

ä¸æµ·ç§æå¤§å¦

Synthetic lethal gene pair prediction methods, devices, equipment and media based on knowledge graph reasoning

CN116796162A

( en )

2023-05-23

2023-09-22

é¿éå·´å·´ï¼ä¸å½ï¼æéå¬å¸

Entity relationship recognition method and entity relationship analysis model training method

CN116468131B

( en )

2023-06-19

2023-09-01

æé½å¸å¥ç¹è½¯ä»¶æéå¬å¸

Automatic AI (advanced technology attachment) driven project method and system based on staged retraining

CN118821919B

( en )

2024-07-26

2024-12-20

åäº¬æ®å·´å¤§æ°æ®æéå¬å¸

Knowledge management system and method

CN120163062B

( en )

2025-03-21

2025-09-23

æ¹åå¤§å¦

Product design generation method, system, equipment and medium

Citations (2)

* Cited by examiner, â Cited by third party

Publication number

Priority date

Publication date

Assignee

Title

US20150339379A1

( en )

2014-05-26

2015-11-26

International Business Machines Corporation

Method of searching for relevant node, and computer therefor and computer program

US20190244107A1

( en )

2018-02-06

2019-08-08

Hrl Laboratories, Llc

Domain adaption learning system

Family Cites Families (11)

* Cited by examiner, â Cited by third party

Publication number

Priority date

Publication date

Assignee

Title

US20100185568A1

( en )

2009-01-19

2010-07-22

Kibboko, Inc.

Method and System for Document Classification

US20180260389A1

( en )

2017-03-08

2018-09-13

Fujitsu Limited

Electronic document segmentation and relation discovery between elements for natural language processing

CN108304911B

( en )

2018-01-09

2020-03-13

Knowledge extraction method, system and equipment based on memory neural network

US11625620B2

( en )

2018-08-16

2023-04-11

Oracle International Corporation

Techniques for building a knowledge graph in limited knowledge domains

US11210590B2

( en )

2019-01-08

2021-12-28

International Business Machines Corporation

Generating free text representing semantic relationships between linked entities in a knowledge graph

CN112015859B

( en )

2019-05-31

2023-08-18

ç¾åº¦å¨çº¿ç½ç»ææ¯ï¼åäº¬ï¼æéå¬å¸

Knowledge hierarchy extraction method and device for text, computer equipment and readable medium

CN110765235B

( en )

2019-09-09

2023-09-05

Training data generation method, device, terminal and readable medium

CN110674317B

( en )

2019-09-30

2022-04-12

åäº¬é®çµå¤§å¦

A method and device for entity linking based on graph neural network

CN110727806B

( en )

2019-12-17

2020-08-11

åäº¬ç¾åº¦ç½è®¯ç§ææéå¬å¸

Text processing method and device based on natural language and knowledge graph

CN111950291B

( en )

2020-06-22

2024-02-23

åäº¬ç¾åº¦ç½è®¯ç§ææéå¬å¸

Semantic representation model generation method, device, electronic device and storage medium

CN111950293B

( en )

2020-06-24

2023-06-23

åäº¬ç¾åº¦ç½è®¯ç§ææéå¬å¸

Semantic representation model generation method, device, electronic device and storage medium

2020

2020-12-21

CN202011520491.4A

patent/CN112507040B/en

active

Active

2021

2021-09-21

US17/480,575

patent/US20220004892A1/en

active

Pending

2021-09-22

EP21198333.3A

patent/EP3913543A3/en

not_active

Withdrawn

2021-11-29

KR1020210167756A

patent/KR20220005416A/en

active

Pending

2021-12-01

JP2021195422A

patent/JP2022028908A/en

active

Pending

Patent Citations (2)

* Cited by examiner, â Cited by third party

Publication number

Priority date

Publication date

Assignee

Title

US20150339379A1

( en )

2014-05-26

2015-11-26

International Business Machines Corporation

Method of searching for relevant node, and computer therefor and computer program

US20190244107A1

( en )

2018-02-06

2019-08-08

Hrl Laboratories, Llc

Domain adaption learning system

Non-Patent Citations (2)

* Cited by examiner, â Cited by third party

Title

Nathani et al., " Learning Attention-based Embeddings for Relation Prediction in Knowledge Graphs, " arXiv:1906.01195v1 [cs.LG] 4 Jun 2019 (Year: 2019)

Wang et al., " Heterogeneous Graph Attention Network, " WWW â19, May 13â17, 2019, San Francisco, CA, USA (Year: 2019)

Cited By (11)

* Cited by examiner, â Cited by third party

Publication number

Priority date

Publication date

Assignee

Title

US20230281467A1

( en )

2022-03-07

2023-09-07

Jpmorgan Chase Bank, N.A.

Systems and methods for building an architecture knowledge graph

CN114694769A

( en )

2022-03-28

2022-07-01

åäº¬ç¾åº¦ç½è®¯ç§ææéå¬å¸

Molecular representation method, molecular representation model training method and device

CN114896974A

( en )

2022-04-21

2022-08-12

åäº¬ç¾åº¦ç½è®¯ç§ææéå¬å¸

Media information processing method, device, electronic equipment and storage medium

CN114942997A

( en )

2022-04-21

2022-08-26

é¿éå·´å·´ï¼ä¸å½ï¼æéå¬å¸

Data processing method, model training method, risk identification method, equipment and storage medium

US20240330679A1

( en )

2023-03-30

2024-10-03

Microsoft Technology Licensing, Llc

Heterogeneous tree graph neural network for label prediction

US12585946B2

( en )

2023-03-30

2026-03-24

Microsoft Technology Licensing, Llc

Heterogeneous tree graph neural network for label prediction

CN116932737A

( en )

2023-07-25

2023-10-24

Character attribute extraction method based on big data

CN116932013A

( en )

2023-09-15

2023-10-24

å¹¿å·å¸æ¡å¤å¨æ¢°è®¾å¤å·¥ç¨æéå¬å¸

Medical gas equipment operation system updating method and system based on artificial intelligence

CN117274006A

( en )

2023-11-21

2023-12-22

å±±ä¸å±±å¤§é¸¥çè½¯ä»¶è¡ä»½æéå¬å¸

A proposition method, system, equipment and storage medium

CN118520879A

( en )

2024-05-23

2024-08-20

å¹¿å·å¸é«ç§éä¿¡ææ¯è¡ä»½æéå¬å¸

Semantic integrity judging method and system based on multi-head attention mechanism

CN119358687A

( en )

2024-12-26

2025-01-24

åäº¬é¿äºç§ææéå¬å¸

A method and device for collaborative reasoning of multiple heterogeneous models

Also Published As

Publication number

Publication date

KR20220005416A

( en )

2022-01-13

EP3913543A2

( en )

2021-11-24

JP2022028908A

( en )

2022-02-16

CN112507040B

( en )

2023-08-08

CN112507040A

( en )

2021-03-16

EP3913543A3

( en )

2022-04-27

Method for training multivariate relationship generation model, electronic … — Beijing Baidu Netcom Science And Technology Co., Ltd. (US20220004892A1)

Related documents