CN111680109A - Knowledge graph representation learning model training method and device and electronic equipment - Google Patents
Knowledge graph representation learning model training method and device and electronic equipment Download PDFInfo
- Publication number
- CN111680109A CN111680109A CN202010324229.6A CN202010324229A CN111680109A CN 111680109 A CN111680109 A CN 111680109A CN 202010324229 A CN202010324229 A CN 202010324229A CN 111680109 A CN111680109 A CN 111680109A
- Authority
- CN
- China
- Prior art keywords
- training data
- tail
- triple
- entity
- tail entity
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000012549 training Methods 0.000 title claims abstract description 341
- 238000000034 method Methods 0.000 title claims abstract description 63
- 239000013598 vector Substances 0.000 claims abstract description 131
- 230000006870 function Effects 0.000 claims abstract description 114
- 230000008569 process Effects 0.000 claims description 26
- 238000004590 computer program Methods 0.000 claims description 15
- 238000010276 construction Methods 0.000 claims description 7
- 238000003860 storage Methods 0.000 claims description 7
- 239000000284 extract Substances 0.000 claims description 6
- 238000010586 diagram Methods 0.000 description 10
- 230000009471 action Effects 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/28—Databases characterised by their database models, e.g. relational or object models
- G06F16/284—Relational databases
- G06F16/288—Entity relationship models
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Electrically Operated Instructional Devices (AREA)
Abstract
The application provides a knowledge graph representation learning model training method, a knowledge graph representation learning model training device and electronic equipment, wherein the method comprises the following steps: acquiring triple training data in a knowledge graph; constructing quadruple training data according to the triple training data, wherein the quadruple training data comprises the triple training data and weights corresponding to the triple training data; inputting the quadruple training data into a preset model, and training the preset model; and when the loss function value of the loss function of the preset model is smaller than the loss function threshold value, determining the trained preset model as a knowledge graph representation learning model. Therefore, when the training knowledge graph represents the learning model, the importance degree of different triple training data is considered, so that the accuracy of the vectors corresponding to the head entity, the vectors corresponding to the tail entity and the relationship between the head entity and the tail entity extracted by the trained knowledge graph representation learning model is higher.
Description
Technical Field
The application relates to the technical field of knowledge maps and deep learning, in particular to a knowledge map representation learning model training method and device and electronic equipment.
Background
With the rapid development of science and technology, the construction and application of knowledge maps are rapidly developed. The knowledge spectrogram is a multi-relationship graph consisting of nodes and edges of different types, wherein each node represents an entity, and each edge represents a relationship between two nodes (called a head node and a tail node) connected by the edge, namely, the head entity and the tail entity are connected together through a specific relationship. And the triple training data consisting of the head entity, the tail entity and the relationship constitutes a fact. For example, (yaoming, born in shanghai) is a fact.
Although facts composed of triples are effective in representing structured data, in practical applications, facts composed of triples are difficult to manipulate, and for this reason knowledge graphs represent learning applications, and the purpose of knowledge graphs is to convert head entities, tail entities, and relationships into vectors.
The inventor discovers that in the process of implementing the invention: in the related art, a knowledge graph representation learning method treats each fact or entity equally when converting a head entity, a tail entity and a relation into a vector. However, in practical application scenarios, the importance of different facts or entities is different. Therefore, by using the technical scheme of the related technology, the accuracy of the vector corresponding to the head entity, the vector corresponding to the tail entity and the vector corresponding to the relationship obtained by conversion is low.
Disclosure of Invention
In order to solve the technical problems of low accuracy of vectors corresponding to head entities, tail entities and relations obtained through conversion in the related art, the application provides a knowledge graph representation learning model training method and device and electronic equipment.
In a first aspect, an embodiment of the present application provides a method for training a knowledge graph representation learning model, where the method includes:
acquiring triple training data in a knowledge graph, wherein the triple training data comprises a head entity, a tail entity and a relation between the head entity and the tail entity;
constructing quadruplet training data according to the triple training data, wherein the quadruplet training data comprises the triple training data and weights corresponding to the triple training data, and the weights corresponding to the triple training data are used for representing the importance degree of the triple training data;
inputting the four-tuple training data into a preset model, and training the preset model, wherein in the process of training the preset model, the preset model extracts a vector corresponding to the head entity, a vector corresponding to the tail entity and a vector corresponding to the relationship between the head entity and the tail entity;
and when the loss function value of the loss function of the preset model is smaller than the loss function threshold value, determining the trained preset model as a knowledge graph representation learning model.
Optionally, the constructing quadruple training data according to the triple training data includes:
counting the number of tail entities in triple training data in a knowledge graph;
grouping the tail entities according to the quantity interval of the quantity of the tail entities to obtain a tail entity group, wherein the tail entities with the quantity of the tail entities in the same quantity interval are divided into the same tail entity group;
determining the weight of the corresponding tail entity group based on the interval value of the quantity interval, wherein the smaller the interval value of one quantity interval is, the larger the weight of the tail entity group corresponding to the quantity interval is;
for each tail entity group, determining the weight of the tail entity group as the weight corresponding to the triple training data of each tail entity included in the tail entity group;
and determining the triple training data and the weight corresponding to the triple training data as quadruple training data.
Optionally, the loss function of the preset model is determined based on a score function of the triple training data, a weight corresponding to the triple training data, a score function of the negative triple training data, and a weight corresponding to the negative triple training data;
the negative triple training data is obtained by converting a head entity and/or a tail entity in the triple training data, and the weight corresponding to the negative triple training data is the weight of a tail entity group corresponding to a quantity interval where the quantity of the tail entity of the negative triple training data is located.
Optionally, the preset loss function of the preset model is:
wherein h represents a head entity, t represents a tail entity, r represents the relationship between the head entity h and the tail entity t, and omega represents the weight corresponding to the triple training data; gamma is a preset boundary value; (h, r, t) represents triplet training data, f (h, r, t) represents a scoring function of the triplet training data, and S-new represents a quadruplet training data set; ω 'represents the weight corresponding to the negative triplet training data, (h', r, t ') table negative triplet training data, f (h', r, t ') represents the score function of the negative triplet training data, and S' -new represents the negative quadruplet training data set.
Optionally, when f (h, r, t) and f (h ', r, t') are energy functions in the TransE model, the method further comprises:
and in the process of adjusting the loss function value of the loss function of the preset model to be smaller than the loss function threshold, adjusting the vector corresponding to h, the vector corresponding to r and the vector corresponding to t, so that the difference between the sum of the vector corresponding to h and the vector corresponding to r and the vector corresponding to t is smaller than the preset vector.
In a second aspect, an embodiment of the present application provides a knowledge graph representation learning model training apparatus, including:
the three-tuple training data acquisition module is used for acquiring three-tuple training data in the knowledge graph, wherein the three-tuple training data comprises a head entity, a tail entity and a relation between the head entity and the tail entity;
the quadruplet training data construction module is used for constructing quadruplet training data according to the triple training data, the quadruplet training data comprises the triple training data and the weight corresponding to the triple training data, and the weight corresponding to the triple training data is used for representing the importance degree of the triple training data;
the preset model training module is used for inputting the four-tuple training data into a preset model and training the preset model, wherein in the process of training the preset model, the preset model extracts a vector corresponding to the head entity, a vector corresponding to the tail entity and a vector corresponding to the relationship between the head entity and the tail entity;
and the knowledge graph representation learning model determining module is used for determining the trained preset model as the knowledge graph representation learning model when the loss function value of the loss function of the preset model is smaller than the loss function threshold value.
Optionally, the quadruple training data construction module is specifically configured to:
counting the number of tail entities in triple training data in a knowledge graph;
grouping the tail entities according to the quantity interval of the quantity of the tail entities to obtain a tail entity group, wherein the tail entities with the quantity of the tail entities in the same quantity interval are divided into the same tail entity group;
determining the weight of the corresponding tail entity group based on the interval value of the quantity interval, wherein the smaller the interval value of one quantity interval is, the larger the weight of the tail entity group corresponding to the quantity interval is;
for each tail entity group, determining the weight of the tail entity group as the weight corresponding to the triple training data of each tail entity included in the tail entity group;
and determining the triple training data and the weight corresponding to the triple training data as quadruple training data.
Optionally, the loss function of the preset model is determined based on a score function of the triple training data, a weight corresponding to the triple training data, a score function of the negative triple training data, and a weight corresponding to the negative triple training data;
the negative triple training data is obtained by converting a head entity and/or a tail entity in the triple training data, and the weight corresponding to the negative triple training data is the weight of a tail entity group corresponding to a quantity interval where the quantity of the tail entity of the negative triple training data is located.
Optionally, the loss function of the preset model is:
wherein h represents a head entity, t represents a tail entity, r represents the relationship between the head entity h and the tail entity t, and omega represents the weight corresponding to the triple training data; gamma is a preset boundary value; (h, r, t) represents triplet training data, f (h, r, t) represents a score function corresponding to the triplet training data, and S-new represents a quadruplet training data set; ω 'represents the weight corresponding to the negative triplet training data, (h', r, t ') table negative triplet training data, f (h', r, t ') represents the score function of the negative triplet training data, and S' -new represents the negative quadruplet training data set.
Optionally, when f (h, r, t) and f (h ', r, t') are energy functions in the TransE model, the apparatus further comprises:
and the vector adjusting module is used for adjusting the vector corresponding to h, the vector corresponding to r and the vector corresponding to t in the process of adjusting the loss function value of the loss function of the preset model to be smaller than the loss function threshold value, so that the difference value between the sum of the vector corresponding to h and the vector corresponding to r and the vector corresponding to t is smaller than the preset vector.
In a third aspect, an embodiment of the present application provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor implements the steps of the method for training a knowledge graph representation learning model according to the first aspect when executing the program.
In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the steps of the method for training a knowledge graph representation learning model according to the first aspect.
According to the technical scheme provided by the embodiment of the application, when the knowledge graph is trained to represent the learning model, triple training data in the knowledge graph are obtained, quadruple training data are constructed according to the triple training model, and the quadruple training data not only comprise the triple training data, but also comprise weights corresponding to the triple training data; after the quadruple training data are obtained, the quadruple training data can be input into a preset model to train the preset model, and in the process of training the preset model, the preset model can learn a vector corresponding to a head entity, a vector corresponding to a tail entity and a vector corresponding to the relationship between the head entity and the tail entity; and finally, when the loss function value of the loss function of the preset model is smaller than the loss function threshold value, determining the trained preset model as a knowledge graph representation learning model.
Therefore, according to the technical scheme provided by the embodiment of the application, when the knowledge graph is trained to represent the learning model, the importance degree of different triple training data is considered, so that the trained knowledge graph represents the vector corresponding to the head entity, the vector corresponding to the tail entity and the vector corresponding to the relationship between the head entity and the tail entity extracted by the learning model, and the accuracy is high.
Drawings
FIG. 1 is a flow chart of steps of a method for training a knowledge graph representation learning model according to an embodiment of the present application;
FIG. 2 is a flow chart of the steps of one implementation of S120 in the embodiment shown in FIG. 1;
FIG. 3 is a block diagram of a knowledge graph representation learning model training apparatus according to an embodiment of the present disclosure;
fig. 4 is a block diagram of an electronic device according to an embodiment of the present disclosure.
Detailed Description
In order to make the aforementioned objects, features and advantages of the present application more comprehensible, the present application is described in further detail with reference to the accompanying drawings and the detailed description.
In order to solve the technical problem that when a head entity, a tail entity and a relation are converted into vectors by a knowledge graph representation learning method in the related art, the accuracy of the converted vectors corresponding to the head entity, the converted vectors corresponding to the tail entity and the converted vectors corresponding to the relation is low due to the fact that each fact or entity is treated equally, the application provides a knowledge graph representation learning model training method, a knowledge graph representation learning model training device and electronic equipment.
In a first aspect, a detailed description is first given of a knowledge graph representation learning model training method provided in an embodiment of the present application.
As shown in fig. 1, a method for training a knowledge graph representation learning model provided in an embodiment of the present application may specifically include the following steps:
and S110, acquiring triple training data in the knowledge graph.
The triple training data comprises a head entity, a tail entity and a relation between the head entity and the tail entity.
Specifically, the knowledge spectrogram is a multi-relationship graph composed of nodes and edges of different types, wherein each node represents an entity, and each edge represents a relationship between two nodes (called a head node and a tail node) connected by the edge, that is, represents that the head entity and the tail entity are connected together through a specific relationship. And the head entity, the tail entity, and the relationship between the head entity and the tail entity constitute the triplet training data.
For example, in the knowledge-graph, the head entity is yaoming, the tail entity is shanghai, and the relationship between the head entity and the tail entity is birth, then the triplet training data is: (Yaoming, born in Shanghai).
And S120, constructing quadruple training data according to the triple training data.
The quadruple training data comprises triple training data and weights corresponding to the triple training data, and the weights corresponding to the triple training data are used for representing the importance degree of the triple training data.
Specifically, the inventors found out in the process of implementing the invention that: the importance of different facts or entities in the knowledge-graph is different, so that after the triple training data in the knowledge-graph is obtained, the importance degree of the triple training data can be judged.
For example, a merchant has many tags, the merchant is a head entity, each tag is a tail entity, and the relationship between the merchant and the tag is: is labeled. In practical applications, the importance of the multiple tags that the merchant has varies. For example, the labels such as "serve well" and "have large table" are relatively general, that is, many merchants are likely to be marked with the labels such as "serve well" and "have large table", and the general label is not very important for the merchants; the labels such as 'suitable for walking baby' and 'suitable for couples' are relatively distinctive, only a small number of merchants can be marked with the labels such as 'suitable for walking baby' and 'suitable for couples', and the labels are relatively important for the merchants.
Therefore, the triple training data (the merchant is labeled and is suitable for walking babies) and the triple training data (the merchant is labeled and is suitable for couples) have higher importance degree to the merchant, namely the triple training data has higher importance. And (the merchant, the large table and the tag) such triplet training data has low importance for the merchant, that is, the significance of such triplet training data is low.
As can be seen from the above description, for a triplet of training data, the importance of the triplet of training data may be determined by how many times tail entities are present in the knowledge graph. It can be understood that, if a tail entity included in a triplet of training data appears less frequently in the knowledge graph, indicating that the tail entity is representative, the significance level of the triplet of training data is higher, and therefore, the weight corresponding to the triplet of training data is larger. On the contrary, if the tail entity included in a triplet of training data appears more times in the knowledge graph, which indicates that the tail entity is not representative, the significance level of the triplet of training data is lower, and therefore, the weight corresponding to the triplet of training data is smaller.
For clarity of description of the scheme, a detailed description of the specific implementation of constructing quad training data according to triple training data at S120 will be provided in the following examples.
And S130, inputting the quadruple training data into a preset model, and training the preset model.
In the process of training the preset model, the preset model extracts a vector corresponding to a head entity, a vector corresponding to a tail entity and a vector corresponding to the relationship between the head entity and the tail entity.
Specifically, after the quadruple training data is obtained, the quadruple training data can be input into the preset model, and after the quadruple training data is obtained by the preset model, the vector corresponding to the head entity, the vector corresponding to the tail entity, and the vector corresponding to the relationship between the head entity and the tail entity can be extracted, that is, the vector corresponding to the head entity, the vector corresponding to the tail entity, and the vector corresponding to the relationship between the head entity and the tail entity are learned by the preset model.
The preset model may be: any one of a TransE model, a TransH model, a TransR & CTRANSR model, a TransD model, an NTN model, an SSP model and a DKRL model can be understood by a person skilled in the art, and the description of the models is omitted herein, and the preset model is not specifically limited in the embodiments of the present application.
And S140, when the loss function value of the loss function of the preset model is smaller than the loss function threshold value, determining the trained preset model as a knowledge graph representation learning model.
Specifically, in the process of training the preset model, the loss function value of the loss function of the preset model is adjusted, and when the loss function value of the loss function of the preset model is smaller than the loss function threshold, the preset model is trained well, and at this time, the trained preset model can be determined to be a knowledge graph representation learning model. The size of the loss function threshold is not specifically limited in the embodiments of the present application.
For clarity of description of the scheme, the loss function of the preset model will be described in detail in the following embodiments.
According to the technical scheme provided by the embodiment of the application, when the knowledge graph is trained to represent the learning model, triple training data in the knowledge graph are obtained, quadruple training data are constructed according to the triple training model, and the quadruple training data not only comprise the triple training data, but also comprise weights corresponding to the triple training data; after the quadruple training data are obtained, the quadruple training data can be input into a preset model to train the preset model, and in the process of training the preset model, the preset model can learn a vector corresponding to a head entity, a vector corresponding to a tail entity and a vector corresponding to the relationship between the head entity and the tail entity; and finally, when the loss function value of the loss function of the preset model is smaller than the loss function threshold value, determining the trained preset model as a knowledge graph representation learning model.
Therefore, according to the technical scheme provided by the embodiment of the application, when the knowledge graph is trained to represent the learning model, the importance degree of different triple training data is considered, so that the trained knowledge graph represents the vector corresponding to the head entity, the vector corresponding to the tail entity and the vector corresponding to the relationship between the head entity and the tail entity extracted by the learning model, and the accuracy is high.
For clarity of description of the scheme, a detailed description of the specific implementation of constructing quad training data according to triple training data at S120 will be provided in the following examples.
As shown in fig. 2, S120, constructing quadruple training data according to triple training data may include the following steps:
and S121, counting the number of tail entities in the triple training data in the knowledge graph.
Specifically, since the knowledge graph has a large amount of triplet training data, each triplet training data includes a tail entity, and some triplet training data correspond to the same tail entity, for example, if merchant 1 and merchant 2 are labeled "good service", then (merchant 1 is labeled good service), and (merchant 2 is labeled good service), the two triplet training data correspond to the same tail entity. Therefore, the number of tail entities in the triplet training data may be counted, and the number of tail entities may be one or multiple.
And S122, grouping the tail entities according to the number interval of the tail entities to obtain tail entity groups.
And the tail entities with the number of the tail entities in the same number interval are divided into the same tail entity group.
Specifically, after the number of tail entities is obtained through statistics, the tail entities may be grouped according to a number interval in which the number of tail entities is located, so as to obtain a tail entity group. For example, the tail entities with the number of tail entities in the number interval (1,100) can be divided into one group, the tail entities with the number of tail entities in the number interval (100,200) can be divided into one group, and the tail entities with the number of tail entities in the number interval (200,300) can be divided into one group. Of course, this is only one example of grouping the tail entities according to the number interval in which the number of the tail entities is located, and the number interval is not specifically limited in the embodiment of the present application.
And S123, determining the weight of the corresponding tail entity grouping based on the interval value of the number interval.
Wherein, the smaller the interval value of a quantity interval, the larger the weight of the tail entity grouping corresponding to the quantity interval.
As can be seen from the embodiment shown in fig. 1, if a tail entity included in a triplet of training data occurs a small number of times in the knowledge graph, which indicates that the tail entity is representative, the significance level of the triplet of training data is higher, and therefore, the weight corresponding to the triplet of training data is larger. On the contrary, if the tail entity included in a triplet of training data appears more times in the knowledge graph, which indicates that the tail entity is not representative, the significance level of the triplet of training data is lower, and therefore, the weight corresponding to the triplet of training data is smaller. To sum up, the smaller the interval value of a quantity interval, the greater the weight of the tail entity group corresponding to the quantity interval. It should be noted that, in the embodiment of the present application, the size of the weight of each tail entity group is not specifically limited.
And S124, determining the tail entity grouping weight as the weight corresponding to the triple training data of each tail entity included in the tail entity grouping for each tail entity grouping.
After determining the weight corresponding to each tail entity group, the weight corresponding to the tail entity group may be determined as the weight corresponding to each tail entity in the tail entity group. After the weights corresponding to the tail entities are determined, the weights corresponding to the triple training data including the tail entities are obtained.
For example, the weight of a certain tail entity group is weight 1, then the weight corresponding to each tail entity in the tail entity group is weight 1, and the weights corresponding to the triplet training data including each tail entity are weight 1.
And S125, determining the triple training data and the weight corresponding to the triple training data as quadruple training data.
After the triple training data and the weight corresponding to the triple training data are determined, the quadruple training data can be obtained. Specifically, the quadruple training data may be (head entity, relationship, tail entity, weight corresponding to the triple training data).
Therefore, according to the technical scheme provided by the embodiment, the triple training data can be accurately obtained by counting the number of tail entities in the triple training data, so that the quadruple training data is accurately constructed, and the knowledge graph representation learning model is trained by using the quadruple training data in the subsequent steps.
Based on the embodiment shown in fig. 2, the loss function of the preset model is determined based on the score function of the triplet training data, the weight corresponding to the triplet training data, the score function of the negative triplet training data, and the weight corresponding to the negative triplet training data.
The negative triple training data are obtained by converting a head entity and/or a tail entity in the triple training data, and the weight corresponding to the negative triple training data is the weight of a tail entity group corresponding to a quantity interval where the tail entity quantity of the negative triple training data is located.
Specifically, the training data for training the preset model may be divided into positive sample training data and negative sample training data, where the positive sample training data is triplet training data, and may be (yaoming, born from shanghai, for example). The negative sample training data is negative triple training data, and the negative triple training data is obtained by transforming a head entity and/or a tail entity in the triple training data. For example, where the triplet training data is (yaoming, birth, shanghai), the negative triplet training data may be (yaoming, birth, fujian), or (wufang, birth, shanghai), etc. After the negative triple training data is obtained, the number of tail entities in the negative triple training data can be counted, a number interval where the number of the tail entities is located is determined, and the weight of the tail entity group corresponding to the number interval where the number of the tail entities is located is determined as the weight corresponding to the negative triple training data.
Therefore, when the loss function of the preset model is determined, the weight corresponding to the triple training data and the weight corresponding to the negative triple training data are considered, namely the importance information of knowledge in the knowledge spectrogram is considered, so that the determined preset loss function of the preset model is more accurate, and the vector corresponding to the head entity, the vector corresponding to the tail entity and the vector corresponding to the relationship between the head entity and the tail entity learned by the preset model are more accurate.
As an implementation manner of the embodiment of the present application, a loss function of the preset model is:
wherein h represents a head entity, t represents a tail entity, r represents the relationship between the head entity h and the tail entity t, and omega represents the weight corresponding to the triple training data; gamma is a preset boundary value; (h, r, t) represents triplet training data, f (h, r, t) represents a scoring function of the triplet training data, and S-new represents a quadruplet training data set; ω 'represents the weight corresponding to the negative triplet training data, (h', r, t ') table negative triplet training data, f (h', r, t ') represents the score function of the negative triplet training data, and S' -new represents the negative quadruplet training data set.
Specifically, in this implementation, a loss function L _ new based on adaptive weight is proposed, and a vector corresponding to a head entity, a vector corresponding to a tail entity, and a vector of a relationship between the head entity and the tail entity are obtained by learning by minimizing a loss function value of the loss function. Specifically, a random gradient descent method may be adopted to minimize a loss function value of the loss function, and a vector corresponding to a head entity, a vector corresponding to a tail entity, a vector corresponding to a relationship between the head entity and the tail entity, and a mutual relationship between the three vectors in the knowledge map are obtained through learning.
Also, in one embodiment, when f (h, r, t) and f (h ', r, t') are energy functions in the TransE model, the method may further include the steps of:
and in the process of adjusting the loss function value of the loss function of the preset model to be smaller than the loss function threshold, adjusting the vector corresponding to h, the vector corresponding to r and the vector corresponding to t, so that the difference between the sum of the vector corresponding to h and the vector corresponding to r and the vector corresponding to t is smaller than the preset vector.
Specifically, the process of minimizing the loss function value of the loss function is a process of minimizing the score function, and the process of minimizing the loss function value of the loss function is a process of achieving the optimization goal. When f in the score function is an energy function in a TransE model, then, in the process of minimizing the loss function value of the loss function, when the type of the relation r is a simple relation type 1-1 or a complex relation type 1-N, N-1, N-N, the difference between the sum of the vector corresponding to h and the vector corresponding to r and the vector corresponding to t is smaller than a preset vector by continuously keeping the vector corresponding to h, the vector corresponding to r and the vector corresponding to t, namely, the vector corresponding to h and the vector corresponding to r are equal to the vector corresponding to t as much as possible.
According to the technical scheme provided by the embodiment, the vector corresponding to h and the vector corresponding to r are controlled to be equal to the vector corresponding to t as much as possible, and when r and t are the same, a plurality of triple training data can be represented through the relationship among the vector corresponding to h, the vector corresponding to r and the vector corresponding to t. For example, one triplet of training data is (yaoming, birth, shanghai) and the other triplet of training data is (wu, birth, shanghai), which may indicate that yaoming and wu are both born in shanghai by the relationship between a vector corresponding to h, a vector corresponding to r, and a vector corresponding to t.
It is noted that, for simplicity of explanation, the method embodiments are described as a series of acts or combination of acts, but those skilled in the art will appreciate that the present application is not limited by the order of acts, as some steps may, in accordance with the present application, occur in other orders and concurrently. Further, those skilled in the art will also appreciate that the embodiments described in the specification are exemplary and that no action is necessarily required in this application.
In a second aspect, an embodiment of the present application provides a knowledge graph representation learning model training apparatus, as shown in fig. 3, the apparatus includes:
a triplet training data obtaining module 310, configured to obtain triplet training data in a knowledge graph, where the triplet training data includes a head entity, a tail entity, and a relationship between the head entity and the tail entity;
a quadruplet training data construction module 320, configured to construct quadruplet training data according to the triple training data, where the quadruplet training data includes the triple training data and a weight corresponding to the triple training data, and the weight corresponding to the triple training data is used to represent an importance degree of the triple training data;
a preset model training module 330, configured to input the quadruple training data into a preset model, and train the preset model, where in the process of training the preset model, the preset model extracts a vector corresponding to the head entity, a vector corresponding to the tail entity, and a vector corresponding to a relationship between the head entity and the tail entity;
and a knowledge graph representation learning model determining module 340, configured to determine the trained preset model as a knowledge graph representation learning model when the loss function value of the loss function of the preset model is smaller than the loss function threshold.
Therefore, according to the technical scheme provided by the embodiment of the application, when the knowledge graph is trained to represent the learning model, the importance degree of different triple training data is considered, so that the trained knowledge graph represents the vector corresponding to the head entity, the vector corresponding to the tail entity and the vector corresponding to the relationship between the head entity and the tail entity extracted by the learning model, and the accuracy is high.
Optionally, the quadruple training data construction module is specifically configured to:
counting the number of tail entities in triple training data in a knowledge graph;
grouping the tail entities according to the quantity interval of the quantity of the tail entities to obtain a tail entity group, wherein the tail entities with the quantity of the tail entities in the same quantity interval are divided into the same tail entity group;
determining the weight of the corresponding tail entity group based on the interval value of the quantity interval, wherein the smaller the interval value of one quantity interval is, the larger the weight of the tail entity group corresponding to the quantity interval is;
for each tail entity group, determining the weight of the tail entity group as the weight corresponding to the triple training data of each tail entity included in the tail entity group;
and determining the triple training data and the weight corresponding to the triple training data as quadruple training data.
Optionally, the loss function of the preset model is determined based on a score function of the triple training data, a weight corresponding to the triple training data, a score function of the negative triple training data, and a weight corresponding to the negative triple training data;
the negative triple training data is obtained by converting a head entity and/or a tail entity in the triple training data, and the weight corresponding to the negative triple training data is the weight of a tail entity group corresponding to a quantity interval where the quantity of the tail entity of the negative triple training data is located.
Optionally, the loss function of the preset model is:
wherein h represents a head entity, t represents a tail entity, r represents the relationship between the head entity h and the tail entity t, and omega represents the weight corresponding to the triple training data; gamma is a preset boundary value; (h, r, t) represents triplet training data, f (h, r, t) represents a score function corresponding to the triplet training data, and S-new represents a quadruplet training data set; ω 'represents the weight corresponding to the negative triplet training data, (h', r, t ') table negative triplet training data, f (h', r, t ') represents the score function of the negative triplet training data, and S' -new represents the negative quadruplet training data set.
Optionally, when f (h, r, t) and f (h ', r, t') are energy functions in the TransE model, the apparatus further comprises:
and the vector adjusting module is used for adjusting the vector corresponding to h, the vector corresponding to r and the vector corresponding to t in the process of adjusting the loss function value of the loss function of the preset model to be smaller than the loss function threshold value, so that the difference value between the sum of the vector corresponding to h and the vector corresponding to r and the vector corresponding to t is smaller than the preset vector.
In a third aspect, an embodiment of the present application provides an electronic device, as shown in fig. 4, including a memory 410, a processor 420, and a computer program stored on the memory and executable on the processor, where the processor implements the steps of the method for training a knowledge graph representation learning model according to the first aspect when executing the program.
Therefore, according to the technical scheme provided by the embodiment of the application, when the knowledge graph is trained to represent the learning model, the importance degree of different triple training data is considered, so that the trained knowledge graph represents the vector corresponding to the head entity, the vector corresponding to the tail entity and the vector corresponding to the relationship between the head entity and the tail entity extracted by the learning model, and the accuracy is high.
In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the steps of the method for training a knowledge graph representation learning model according to the first aspect.
Therefore, according to the technical scheme provided by the embodiment of the application, when the knowledge graph is trained to represent the learning model, the importance degree of different triple training data is considered, so that the trained knowledge graph represents the vector corresponding to the head entity, the vector corresponding to the tail entity and the vector corresponding to the relationship between the head entity and the tail entity extracted by the learning model, and the accuracy is high.
The embodiments in the present specification are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, apparatus, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, terminal devices (systems), and computer program products according to the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing terminal to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing terminal to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing terminal to cause a series of operational steps to be performed on the computer or other programmable terminal to produce a computer implemented process such that the instructions which execute on the computer or other programmable terminal provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
While the preferred embodiments of the present application have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including preferred embodiments and all alterations and modifications as fall within the scope of the application.
Finally, it should also be noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or terminal. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or terminal that comprises the element.
The knowledge graph representation learning model training method and the knowledge graph representation learning model training device provided by the application are introduced in detail, specific examples are applied in the text to explain the principle and the implementation mode of the application, and the description of the above embodiments is only used for helping to understand the method and the core idea of the application; meanwhile, for a person skilled in the art, according to the idea of the present application, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present application.
Claims (12)
1. A method of knowledge graph representation learning model training, the method comprising:
acquiring triple training data in a knowledge graph, wherein the triple training data comprises a head entity, a tail entity and a relation between the head entity and the tail entity;
constructing quadruplet training data according to the triple training data, wherein the quadruplet training data comprises the triple training data and weights corresponding to the triple training data, and the weights corresponding to the triple training data are used for representing the importance degree of the triple training data;
inputting the four-tuple training data into a preset model, and training the preset model, wherein in the process of training the preset model, the preset model extracts a vector corresponding to the head entity, a vector corresponding to the tail entity and a vector corresponding to the relationship between the head entity and the tail entity;
and when the loss function value of the loss function of the preset model is smaller than the loss function threshold value, determining the trained preset model as a knowledge graph representation learning model.
2. The method of claim 1, wherein constructing quad training data from the triple training data comprises:
counting the number of tail entities in triple training data in a knowledge graph;
grouping the tail entities according to the quantity interval of the quantity of the tail entities to obtain a tail entity group, wherein the tail entities with the quantity of the tail entities in the same quantity interval are divided into the same tail entity group;
determining the weight of the corresponding tail entity group based on the interval value of the quantity interval, wherein the smaller the interval value of one quantity interval is, the larger the weight of the tail entity group corresponding to the quantity interval is;
for each tail entity group, determining the weight of the tail entity group as the weight corresponding to the triple training data of each tail entity included in the tail entity group;
and determining the triple training data and the weight corresponding to the triple training data as quadruple training data.
3. The method of claim 2, wherein the loss function of the predetermined model is determined based on a score function of the triplet training data, weights corresponding to the triplet training data, a score function of the negative triplet training data, and weights corresponding to the negative triplet training data;
the negative triple training data is obtained by converting a head entity and/or a tail entity in the triple training data, and the weight corresponding to the negative triple training data is the weight of a tail entity group corresponding to a quantity interval where the quantity of the tail entity of the negative triple training data is located.
4. The method of claim 3, wherein the loss function of the predetermined model is:
wherein h represents a head entity, t represents a tail entity, r represents the relationship between the head entity h and the tail entity t, and omega represents the weight corresponding to the triple training data; gamma is a preset boundary value; (h, r, t) represents triplet training data, f (h, r, t) represents a scoring function of the triplet training data, and S-new represents a quadruplet training data set; ω 'represents the weight corresponding to the negative triplet training data, (h', r, t ') table negative triplet training data, f (h', r, t ') represents the score function of the negative triplet training data, and S' _ new represents the negative quadruplet training data set.
5. The method of claim 4, wherein when f (h, r, t) and f (h ', r, t') are energy functions in the TransE model, the method further comprises:
and in the process of adjusting the loss function value of the loss function of the preset model to be smaller than the loss function threshold, adjusting the vector corresponding to h, the vector corresponding to r and the vector corresponding to t, so that the difference between the sum of the vector corresponding to h and the vector corresponding to r and the vector corresponding to t is smaller than the preset vector.
6. A knowledge graph representation learning model training apparatus, the apparatus comprising:
the three-tuple training data acquisition module is used for acquiring three-tuple training data in the knowledge graph, wherein the three-tuple training data comprises a head entity, a tail entity and a relation between the head entity and the tail entity;
the quadruplet training data construction module is used for constructing quadruplet training data according to the triple training data, the quadruplet training data comprises the triple training data and the weight corresponding to the triple training data, and the weight corresponding to the triple training data is used for representing the importance degree of the triple training data;
the preset model training module is used for inputting the four-tuple training data into a preset model and training the preset model, wherein in the process of training the preset model, the preset model extracts a vector corresponding to the head entity, a vector corresponding to the tail entity and a vector corresponding to the relationship between the head entity and the tail entity;
and the knowledge graph representation learning model determining module is used for determining the trained preset model as the knowledge graph representation learning model when the loss function value of the loss function of the preset model is smaller than the loss function threshold value.
7. The apparatus of claim 6, wherein the quadruple training data construction module is specifically configured to:
counting the number of tail entities in triple training data in a knowledge graph;
grouping the tail entities according to the quantity interval of the quantity of the tail entities to obtain a tail entity group, wherein the tail entities with the quantity of the tail entities in the same quantity interval are divided into the same tail entity group;
determining the weight of the corresponding tail entity group based on the interval value of the quantity interval, wherein the smaller the interval value of one quantity interval is, the larger the weight of the tail entity group corresponding to the quantity interval is;
for each tail entity group, determining the weight of the tail entity group as the weight corresponding to the triple training data of each tail entity included in the tail entity group;
and determining the triple training data and the weight corresponding to the triple training data as quadruple training data.
8. The apparatus of claim 7, wherein the loss function of the predetermined model is determined based on a score function of the triplet training data, a weight corresponding to the triplet training data, a score function of the negative triplet training data, and a weight corresponding to the negative triplet training data;
the negative triple training data is obtained by converting a head entity and/or a tail entity in the triple training data, and the weight corresponding to the negative triple training data is the weight of a tail entity group corresponding to a quantity interval where the quantity of the tail entity of the negative triple training data is located.
9. The apparatus of claim 8, wherein the loss function of the predetermined model is:
wherein h represents a head entity, t represents a tail entity, r represents the relationship between the head entity h and the tail entity t, and omega represents the weight corresponding to the triple training data; gamma is a preset boundary value; (h, r, t) represents triplet training data, f (h, r, t) represents a score function corresponding to the triplet training data, and S-new represents a quadruplet training data set; ω 'represents the weight corresponding to the negative triplet training data, (h', r, t ') table negative triplet training data, f (h', r, t ') represents the score function of the negative triplet training data, and S' _ new represents the negative quadruplet training data set.
10. The apparatus of claim 9, wherein when f (h, r, t) and f (h ', r, t') are energy functions in a TransE model, the apparatus further comprises:
and the vector adjusting module is used for adjusting the vector corresponding to h, the vector corresponding to r and the vector corresponding to t in the process of adjusting the loss function value of the loss function of the preset model to be smaller than the loss function threshold value, so that the difference value between the sum of the vector corresponding to h and the vector corresponding to r and the vector corresponding to t is smaller than the preset vector.
11. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps of the knowledge graph representation learning model training method as claimed in any one of claims 1 to 5 when executing the program.
12. A computer-readable storage medium, characterized in that the computer-readable storage medium has stored thereon a computer program which, when being executed by a processor, carries out the steps of the knowledge graph representation learning model training method according to any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010324229.6A CN111680109B (en) | 2020-04-22 | 2020-04-22 | Knowledge graph representation learning model training method and device and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010324229.6A CN111680109B (en) | 2020-04-22 | 2020-04-22 | Knowledge graph representation learning model training method and device and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111680109A true CN111680109A (en) | 2020-09-18 |
| CN111680109B CN111680109B (en) | 2024-03-29 |
Family
ID=72433691
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010324229.6A Active CN111680109B (en) | 2020-04-22 | 2020-04-22 | Knowledge graph representation learning model training method and device and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111680109B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111930968A (en) * | 2020-10-13 | 2020-11-13 | 支付宝(杭州)信息技术有限公司 | Method and device for updating business model |
| CN112434812A (en) * | 2020-11-26 | 2021-03-02 | 中山大学 | Knowledge graph link prediction method and system based on dual quaternion |
| CN112732940A (en) * | 2021-01-15 | 2021-04-30 | 医渡云（北京）技术有限公司 | Model-based medical knowledge graph reasoning method, device, equipment and medium |
| CN112732941A (en) * | 2021-01-15 | 2021-04-30 | 医渡云（北京）技术有限公司 | Model-based medical knowledge graph construction method, device, equipment and medium |
| CN113742488A (en) * | 2021-07-30 | 2021-12-03 | 清华大学 | Embedded knowledge graph completion method and device based on multitask learning |
| CN113780002A (en) * | 2021-08-13 | 2021-12-10 | 北京信息科技大学 | Knowledge reasoning method and device based on graph representation learning and deep reinforcement learning |
| CN113792344A (en) * | 2021-09-17 | 2021-12-14 | 平安国际智慧城市科技股份有限公司 | Data desensitization processing method, device, equipment and storage medium |
| CN113946695A (en) * | 2021-12-20 | 2022-01-18 | 山东新希望六和集团有限公司 | Method and device for generating animal pedigree and computer equipment |
| CN115033717A (en) * | 2022-08-12 | 2022-09-09 | 杭州恒生聚源信息技术有限公司 | Triple extraction model training method, triple extraction method, device and equipment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106934042A (en) * | 2017-03-16 | 2017-07-07 | 中国人民解放军国防科学技术大学 | A kind of knowledge mapping represents model and its method |
| US20180060733A1 (en) * | 2016-08-31 | 2018-03-01 | International Business Machines Corporation | Techniques for assigning confidence scores to relationship entries in a knowledge graph |
| CN109033129A (en) * | 2018-06-04 | 2018-12-18 | 桂林电子科技大学 | Multi-source Information Fusion knowledge mapping based on adaptive weighting indicates learning method |
-
2020
- 2020-04-22 CN CN202010324229.6A patent/CN111680109B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180060733A1 (en) * | 2016-08-31 | 2018-03-01 | International Business Machines Corporation | Techniques for assigning confidence scores to relationship entries in a knowledge graph |
| CN106934042A (en) * | 2017-03-16 | 2017-07-07 | 中国人民解放军国防科学技术大学 | A kind of knowledge mapping represents model and its method |
| CN109033129A (en) * | 2018-06-04 | 2018-12-18 | 桂林电子科技大学 | Multi-source Information Fusion knowledge mapping based on adaptive weighting indicates learning method |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111930968A (en) * | 2020-10-13 | 2020-11-13 | 支付宝(杭州)信息技术有限公司 | Method and device for updating business model |
| CN112434812A (en) * | 2020-11-26 | 2021-03-02 | 中山大学 | Knowledge graph link prediction method and system based on dual quaternion |
| CN112434812B (en) * | 2020-11-26 | 2023-09-05 | 中山大学 | Knowledge graph link prediction method and system based on dual quaternion |
| WO2022151812A1 (en) * | 2021-01-15 | 2022-07-21 | 医渡云（北京）技术有限公司 | Medical knowledge graph construction method and apparatus based on model, device, and medium |
| CN112732940A (en) * | 2021-01-15 | 2021-04-30 | 医渡云（北京）技术有限公司 | Model-based medical knowledge graph reasoning method, device, equipment and medium |
| CN112732941A (en) * | 2021-01-15 | 2021-04-30 | 医渡云（北京）技术有限公司 | Model-based medical knowledge graph construction method, device, equipment and medium |
| CN112732940B (en) * | 2021-01-15 | 2023-07-14 | 医渡云（北京）技术有限公司 | Reasoning method, device, equipment and medium of medical knowledge graph based on model |
| CN113742488B (en) * | 2021-07-30 | 2022-12-02 | 清华大学 | Embedded knowledge graph completion method and device based on multitask learning |
| CN113742488A (en) * | 2021-07-30 | 2021-12-03 | 清华大学 | Embedded knowledge graph completion method and device based on multitask learning |
| CN113780002A (en) * | 2021-08-13 | 2021-12-10 | 北京信息科技大学 | Knowledge reasoning method and device based on graph representation learning and deep reinforcement learning |
| CN113780002B (en) * | 2021-08-13 | 2023-11-28 | 北京信息科技大学 | Knowledge reasoning method and device based on graph representation learning and deep reinforcement learning |
| CN113792344A (en) * | 2021-09-17 | 2021-12-14 | 平安国际智慧城市科技股份有限公司 | Data desensitization processing method, device, equipment and storage medium |
| CN113792344B (en) * | 2021-09-17 | 2024-06-04 | 平安国际智慧城市科技股份有限公司 | Data desensitization processing method, device, equipment and storage medium |
| CN113946695B (en) * | 2021-12-20 | 2022-04-22 | 山东新希望六和集团有限公司 | Method and device for generating animal pedigree and computer equipment |
| CN113946695A (en) * | 2021-12-20 | 2022-01-18 | 山东新希望六和集团有限公司 | Method and device for generating animal pedigree and computer equipment |
| CN115033717A (en) * | 2022-08-12 | 2022-09-09 | 杭州恒生聚源信息技术有限公司 | Triple extraction model training method, triple extraction method, device and equipment |
| CN115033717B (en) * | 2022-08-12 | 2022-11-08 | 杭州恒生聚源信息技术有限公司 | Triple extraction model training method, triple extraction method, device and equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111680109B (en) | 2024-03-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111680109A (en) | Knowledge graph representation learning model training method and device and electronic equipment | |
| CN109299284B (en) | Knowledge graph representation learning method based on structural information and text description | |
| CN111639679B (en) | Small sample learning method based on multi-scale metric learning | |
| RU2586864C2 (en) | Method and arrangement for local rule of training competition, which leads to a sparse connectivity | |
| CN110046698A (en) | Heterogeneous figure neural network generation method, device, electronic equipment and storage medium | |
| CN105354307A (en) | Image content identification method and apparatus | |
| CN110276442A (en) | A kind of searching method and device of neural network framework | |
| CN109697289A (en) | It is a kind of improved for naming the Active Learning Method of Entity recognition | |
| CN109448795A (en) | The recognition methods of circRNA a kind of and device | |
| CN109726400A (en) | Entity word recognition result evaluation method, apparatus, equipment and entity word extraction system | |
| CN107392310A (en) | neural network model training method and device | |
| CN117158923A (en) | Remote home-care monitoring method based on meta universe | |
| CN115169227A (en) | Design concept generation network construction method and concept scheme automatic generation method | |
| CN110197213A (en) | Image matching method, device and equipment neural network based | |
| CN110222737A (en) | A kind of search engine user satisfaction assessment method based on long memory network in short-term | |
| CN112132690B (en) | Method and device for pushing foreign exchange product information, computer equipment and storage medium | |
| CN114332550A (en) | Model training method, system, storage medium and terminal equipment | |
| EP4227850A1 (en) | Program, learning method, and information processing apparatus | |
| CN108021985A (en) | A kind of model parameter training method and device | |
| CN109740163A (en) | Semantic representation resource generation method and device applied to deep learning model | |
| CN115080748A (en) | Weak supervision text classification method and device based on noisy label learning | |
| CN111159397B (en) | Text classification method and device and server | |
| CN111666770B (en) | Semantic matching method and device | |
| CN108460453A (en) | It is a kind of to be used for data processing method, the apparatus and system that CTC is trained | |
| JP6993250B2 (en) | Content feature extractor, method, and program |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |