CN110619053A - Training method of entity relation extraction model and method for extracting entity relation - Google Patents
Training method of entity relation extraction model and method for extracting entity relation Download PDFInfo
- Publication number
- CN110619053A CN110619053A CN201910883330.2A CN201910883330A CN110619053A CN 110619053 A CN110619053 A CN 110619053A CN 201910883330 A CN201910883330 A CN 201910883330A CN 110619053 A CN110619053 A CN 110619053A
- Authority
- CN
- China
- Prior art keywords
- entity
- training
- relation
- target
- relationship
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Animal Behavior & Ethology (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Machine Translation (AREA)
Abstract
The application discloses a training method of an entity relationship extraction model and a method for extracting entity relationships, and relates to the field of big data. The specific implementation scheme is as follows: matching a training text according to entity relationship triples in a known database, wherein the entity relationship triples comprise known entity pairs and known relationships corresponding to the known entity pairs, and the training text comprises the known entity pairs; predicting the relation of the known entity pair in the training text based on the initial model to obtain a predicted relation; and adjusting parameters of the initial model according to the difference between the known relation and the prediction relation corresponding to the training texts to obtain an entity relation extraction model, wherein the entity relation extraction model is used for predicting the relation of the target entity in the target text. According to the technical scheme, the entity relation in the training text does not need to be manually marked, so that the entity relation can be rapidly extracted.
Description
Technical Field
The present application relates to the field of data processing, and more particularly, to the field of big data.
Background
In knowledge graph construction, knowledge extraction is a core foundation, and entity relations are important components of knowledge extraction. Currently, there is a large amount of free text, but these unstructured information are cluttered and cannot be fully utilized by users. How to extract structured knowledge, especially entity relations, from the free texts has very important significance.
Disclosure of Invention
The embodiment of the application provides a training method of an entity relationship extraction model and a method for extracting an entity relationship, so as to solve one or more technical problems in the prior art.
In a first aspect, an embodiment of the present application provides a method for training an entity relationship extraction model, including:
matching a training text according to entity relationship triples in a known database, wherein the entity relationship triples comprise known entity pairs and known relationships corresponding to the known entity pairs, and the training text comprises the known entity pairs;
predicting the relation of the known entity pair in the training text based on the initial model to obtain a predicted relation;
and adjusting parameters of the initial model according to the difference between the known relation and the prediction relation corresponding to the training texts to obtain an entity relation extraction model, wherein the entity relation extraction model is used for predicting the relation of the target entity in the target text.
According to the technical scheme, the entity relationship triplets are constructed for the known database, and then the corresponding training texts are matched from the mass free texts, so that the entity relationship extraction model is trained. Because the entity pairs and the entity relations in the entity relation triplets are known, the entity relations in the training texts do not need to be manually labeled, and the training efficiency can be improved.
In one embodiment, the initial model is a bi-directional gated loop unit-attention model.
The technical scheme can learn the context information of each word in the training text based on the bidirectional gating circulation unit, and the attention mechanism can learn the weight of the word related to the context so as to better represent the training text and enable the trained model to have better performance.
In one embodiment, predicting relationships of known entity pairs in training text based on an initial model includes:
determining a plurality of training characters from the training text, wherein the training characters comprise training reference characters matched with known entity pairs;
obtaining entity position characteristics of the training characters based on the position relation between the training characters and the training reference characters;
inputting the character features and the entity position features of each training character into an initial model, and predicting the relation of the known entity pairs in the training text.
According to the technical scheme, the training text is represented as the character feature and the entity position feature, so that multi-dimensional vector representation can be established to better represent the training text.
In a second aspect, an embodiment of the present application provides a method for extracting entity relationships, including:
predicting the relation of a target entity pair in a target text based on an entity relation extraction model to obtain an extraction result to be selected, wherein the entity relation extraction model is obtained by training according to any one of the training methods;
judging whether the extraction result to be selected meets a preset condition or not;
and under the condition that the extraction result to be selected is judged not to meet the preset condition, performing syntactic dependency analysis on the target text to determine the relation of the target entity pair in the target text.
According to the technical scheme, by introducing syntax dependency relationship extraction, a new relationship type can be effectively mined, relationships except a database relationship set are supplemented, and the relationship set is enriched.
In one embodiment, predicting the relationship of the target entity pair in the target text based on the entity relationship extraction model comprises:
determining a plurality of target characters from the target text, wherein the target characters comprise target reference characters matched with the target entity pairs;
obtaining entity position characteristics of the target character based on the position relation between the target character and the target reference character;
inputting the character features and entity position features of each target character into an entity relationship extraction model, and predicting the relationship of the target entity pair in a target text.
In a third aspect, an embodiment of the present application provides a training apparatus for an entity relationship extraction model, including:
the matching module is used for matching a training text according to an entity relationship triple in a known database, wherein the entity relationship triple comprises a known entity pair and a known relationship corresponding to the known entity pair, and the training text comprises the known entity pair;
the prediction relation determining module is used for predicting the relation of the known entity pair in the training text based on the initial model to obtain a prediction relation;
and the adjusting module is used for adjusting the parameters of the initial model according to the difference between the known relations and the predicted relations corresponding to the training texts to obtain an entity relation extraction model, and the entity relation extraction model is used for predicting the relation of the target entity in the target text.
In one embodiment, the initial model is a bi-directional gated loop unit-attention model.
In a fourth aspect, an embodiment of the present application provides an apparatus for extracting entity relationships, including:
the model prediction module is used for predicting the relation of a target entity pair in a target text based on an entity relation extraction model to obtain an extraction result to be selected, and the entity relation extraction model is obtained by training according to any one of the training methods;
the judging module is used for judging whether the extraction result to be selected meets the preset condition or not;
and the syntactic dependency analysis module is used for carrying out syntactic dependency analysis on the target text under the condition that the extraction result to be selected is judged not to meet the preset condition so as to determine the relation of the target entity pair in the target text.
Other effects of the above-described alternative will be described below with reference to specific embodiments.
Drawings
The drawings are included to provide a better understanding of the present solution and are not intended to limit the present application. Wherein:
FIG. 1 is a flowchart of a method for training an entity relationship extraction model according to an embodiment of the present application;
FIG. 2 is an exemplary diagram of composing a training text according to one implementation of an embodiment of the present application;
FIG. 3 is a schematic diagram of a Skip-gram structure of a word2vec model according to an embodiment of the present application;
FIG. 4 is an exemplary diagram of an initial model structure according to an embodiment of the present application;
FIG. 5 is a flowchart of a method for extracting entity relationships according to an embodiment of the present application;
FIG. 6 is a diagram of an application example of a method for extracting entity relationships according to an embodiment of the present application;
FIG. 7 is a block diagram of a training apparatus for an entity relationship extraction model according to an embodiment of the present application;
FIG. 8 is a block diagram of an apparatus for extracting entity relationships according to an embodiment of the present application;
FIG. 9 is a block diagram of an electronic device of a method for training an entity relationship extraction model according to an embodiment of the present application;
fig. 10 is a block diagram of an electronic device of a method for extracting entity relationships according to an embodiment of the present application.
Detailed Description
The following description of the exemplary embodiments of the present application, taken in conjunction with the accompanying drawings, includes various details of the embodiments of the application for the understanding of the same, which are to be considered exemplary only. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present application. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.
In the related technology, strategies such as template rules, statistics, clustering and the like are adopted to extract the relation of the target entity pair from the target text. Because the template rule needs to depend on expert definition, the method has poor expandability, and the accuracy of relation extraction is difficult to ensure based on statistics and clustering strategies. In the related technology, a self-expanding (BootStraping) algorithm is adopted for extracting the relationship, namely, for the relationship to be extracted, a plurality of seed examples are manually set, and then a relationship template corresponding to the relationship and more examples are iteratively extracted from data. The number of relation templates expanded by the method is limited, and the accuracy rate is reduced as the number of the templates is increased. The related art regards the relationship extraction task as a classification problem. Effective features are designed according to training data so as to learn various classification models, and then the trained classifier is used for predicting the relation. The method needs to rely on a large amount of manual labeling corpora, the manual labeling is time-consuming and labor-consuming, model training cannot be performed if no data is labeled, and meanwhile, the model training is greatly limited by the data scale. In contrast, in the embodiment of the application, the entity relationship triple is constructed for the known database, and then the corresponding training text is matched from the massive free text, so that the entity relationship extraction model is trained. Because the entity pairs and entity relationships in the entity relationship triplets are known, there is no need to manually label the entity relationships in the training text.
In one implementation, fig. 1 shows a flowchart of a training method of the entity-relationship extraction model of the embodiment. As shown in fig. 1, the training method includes:
step S101: and matching a training text according to the entity relationship triples in the known database, wherein the entity relationship triples comprise known entity pairs and known relationships corresponding to the known entity pairs, and the training text comprises the known entity pairs.
The known database may include existing database Knowledge, i.e., a Knowledge Base (KB), such as a Structured Query Language (SQL) database. In one example, entity-relationship triplets are constructed from a known database in a remote supervised manner, such as Subject-predicate-Object (SPO) triplets, i.e., entity-relationship triplets in KB are used to align free text (e.g., encyclopedia, internet news, etc.) to achieve automatic annotation. Thus, this approach can serve as remote supervision of unmarked data relationship extraction. It can be assumed that if there is some relationship between two entities in the KB, then an unstructured sentence containing both entities can show this relationship. For example, if the entity "Steve Jobs" and the entity "Apple" have a relationship of "fountain" in KB, the unstructured text "Steve Jobs was the co-fountain and CEO of Apple and for hydroPixar." containing the two entities can be used as a training text (e.g., a training example) to train the initial model.
In one example, as shown in fig. 2, the triple library includes a plurality of entity relationship triples constructed from known databases, such as < china, capital, beijing >, < apple, color, red >, < apple, founder, Steve Jobs >, etc. The sentence 1 corresponding to the entity relationship triple < china, capital, beijing > is associated (matched) from the free text, and the capital of china is beijing. "and sentence 2" Beijing is a beautiful city, in the middle of China. "further, the sentence 1 and the sentence 2 can be used as training texts, and the entity relationship triple < china, capital, beijing > is used as the labeled data of the sentence 1 and the sentence 2.
The method of the embodiment fully establishes the relationship bridge between the entity relationship triple and the mass free text in the known database through remote supervision, can conveniently and quickly construct the training text, and can save a large amount of marking time.
Step S102: and predicting the relation of the known entity pair in the training text based on the initial model to obtain a predicted relation.
The initial model may be a deep learning model, such as a Convolutional Neural Network (CNN) or a bidirectional Recurrent Neural Network (RNN). In this embodiment, the initial model may adopt a Bi-directional Gated-loop-Unit-attention (Bi-Gated-current-Unit-attention) model. The Bigru can learn the context information of each word in the training text, and the attention mechanism can learn the weight of the word related to the context, so that the training text is better represented, and the trained model has better performance.
In one embodiment, step S102 may include: determining a plurality of training characters from the training text, wherein the training characters comprise training reference characters matched with known entity pairs; obtaining entity position characteristics of the training characters based on the position relation between the training characters and the training reference characters; inputting the character features and the entity position features of each training character into an initial model, and predicting the relation of the known entity pairs in the training text.
Firstly, the training text may be subjected to a character-dividing operation, that is, the training text is divided into a plurality of characters (training characters), so as to obtain character features corresponding to each training character. The character features can be obtained by extracting features of character (character) granularity from the training text based on a word vector (word 2vec) model. In one example, character features of the training text may be determined based on the Skip-gram structure of the word2vec model. FIG. 3 shows a schematic diagram of the Skip-gram structure of the word2vec model. As shown in FIG. 3, based on a certain character feature w (t) of the input, character features w (t) corresponding to the context of the character, such as w (t-2), w (t-1), w (t +1) and w (t +2), can be learned.
The entity location features are used to characterize the distance of each training character relative to each entity in a known pair of entities. Since the known entity pair includes two known entities (e.g., known entity 1 and known entity 2), the entity location features of each training character are two, one of which is the distance of the training character from the known entity 1 and the other of which is the distance of the training character from the known entity 2.
Wherein the training characters matched with the known entity pairs are training reference characters. If training character 1 matched with known entity 1 is used as the first training reference character, the first entity position features of training character 1 and known entity 1 can be represented by "0", the training character and the first entity position feature of known entity 1 positioned at the left side of training character 1 can be represented by corresponding negative numbers, and the training character and the first entity position feature of known entity 1 positioned at the right side of training character 1 can be represented by corresponding positive numbers. In a similar way, the second entity location characteristic of each training character and the known entity 2 can be obtained. And then two groups of entity position characteristics corresponding to the training texts can be obtained. In one example, each physical location feature may be translated the same distance such that each physical location feature may be represented by a positive number.
In one example, as shown in FIG. 4, the character cutting result for the training text, i.e. each training character x determined from the training text, is input at the input layer (input layer) of the initial model1～xT. In the character embedding layer (character embedding layer) of the initial model, a training character x is learned1～xTRespectively corresponding model features e1～eT. For example: training character x1Corresponding model features e1Including a 100-dimensional character feature, a 5-dimensional first entity location feature, and a 5-dimensional second entity location feature. Model features e of training text1～eTCharacterizing the Bigru layer of the initial model as a bidirectional eigenvector, splicing the bidirectional eigenvector to obtain a corresponding eigenvector h1～hT. Feature vector h1～hTThe analysis effects on the entity relations are different in size, and the weights w of different feature vectors can be learned based on the attention layer of the initial starting model1～wTEach feature vector and its corresponding weight are weight-summed (i.e., w)1×h1+w2×h2+w3×h3+……+wT×hT) And then output to the output layer (output layer) of the initial model. The output Layer comprises a Fully Connected Layer (full Connected Layer) and a softmax network, and is used for outputting the prediction relation y according to the weight summation result.
That is to say, in this embodiment, the constructed training text may be subjected to character feature and entity location feature representation, and is input into the multilayer Bigru network, then input into the attention network, and finally the prediction relationship of the training text is obtained through the softmax network.
Step S103: and adjusting parameters of the initial model according to the difference between the known relation and the prediction relation corresponding to the training texts to obtain an entity relation extraction model, wherein the entity relation extraction model is used for predicting the relation of the target entity in the target text.
And each training text corresponds to a prediction relation and a known relation, wherein the prediction relation is obtained based on initial model prediction, and the known relation is a known relation in an entity relation triple corresponding to the training text. In the training process of the initial model, a loss function is established based on the prediction relation and the known relation of each training text, and the parameters of the initial model are continuously adjusted until the loss function value reaches a set value, so that a trained model, namely an entity relation extraction model, is obtained.
In one implementation, fig. 5 is a flowchart illustrating a method for extracting entity relationships according to the embodiment. As shown in fig. 5, the method includes:
step S501: predicting the relation of the target entity pair in the target text based on an entity relation extraction model to obtain an extraction result to be selected, wherein the entity relation extraction model can be obtained by training according to the training method in the embodiment;
step S502: judging whether the extraction result to be selected meets a preset condition or not;
step S503: and under the condition that the extraction result to be selected is judged not to meet the preset condition, performing syntactic dependency analysis on the target text to determine the relation of the target entity pair in the target text.
That is to say, in this embodiment, the entity relationship extraction model is preferentially used to extract the relationship of the target entity pair in the target text, so as to obtain the extraction result to be selected. And if the extraction result to be selected is not empty, the extraction result to be selected meets the preset condition, and the extraction result to be selected can be used as the relation of the target entity pair in the target text. If the extraction result to be selected is an unknown type or the relation confidence coefficient of the extraction result to be selected is lower than a set value, the extraction result to be selected does not meet the preset condition, and then the syntactic dependency analysis can be performed on the target text to obtain the relation of the target entity in the target text. And if the syntactic dependency analysis is carried out on the target text, and the obtained result is null, returning an unknown relation, namely that the relation of the target entity in the target text is unknown.
In one embodiment, step S501 may include: determining a plurality of target characters from the target text, wherein the target characters comprise target reference characters matched with the target entity pairs; obtaining entity position characteristics of the target character based on the position relation between the target character and the target reference character; inputting the character features and entity position features of each target character into an entity relationship extraction model, and predicting the relationship of the target entity pair in a target text.
Firstly, character splitting operation can be carried out on a target text, and then character features of the target text are determined based on a Skip-gram structure of a word2vec model. Similar to the training method described above, since the target entity pair includes two target entities, the entity position feature of each target character is two. Furthermore, the target text corresponds to two sets of entity location features. And based on the character characteristics and the entity position characteristics of the target text, predicting the relation of the target entity pair in the target text by using an entity relation extraction model.
And performing syntactic dependency analysis on the target text, namely determining the syntactic structure of the target text or the dependency relationship among vocabularies in the target text. This may include two aspects, one is to determine the grammar system of the target text, i.e. to give formal definition to the grammar structure of the legal sentences in the target text; another aspect is syntactic analysis techniques, i.e. the automatic derivation of the syntactic structure of the target text, according to a given syntactic hierarchy, the analysis of the syntactic units contained in the target text and the relations between these syntactic units.
In one example, syntactic dependency analysis typically satisfies the following conditions: 1) only one component in a sentence is independent; 2) other components of the sentence are subordinate to a certain component; 3) neither component can depend on two or more components; 4) if component a is directly dependent on component B, and component C is located between a and B in the sentence, then component C is either dependent on a, or dependent on B, or dependent on some component between a and B; 5) the other components on the left and right sides of the central component are not related to each other.
In one example, a predicate-centric fact triple relationship may be extracted from the target text, including: the relationships of a main subject guest, a fixed language post-positioned guest, a moving guest relationship, a mediating guest relationship and the like.
The following table shows an example of syntactic dependency analysis on the target text "J bought K company".
Wherein, SBV (subject-verb) represents the main-meaning relationship, i.e. the relationship between nouns and actions; HED is a virtualized head node; MT (mood-tense) represents a morphological structure, such as temporal help words (e.g., right, left, over) and mood help words (e.g., bar, o, wool) for expressing sentences; ATT (attribute) represents a centering relationship, namely a relationship between a fixed language and a central word, such as the fixed language has a modification or restriction effect on the central word; the VOB (verb-object) represents the verb relationship, i.e. the relationship of the non-predicate verb and its object, i.e. the two words in the VOB constitute the verb phrase and serve as the other modifier of the sentence. The dependencies of column 6 represent the character's dependencies with the head node.
In this embodiment, the entity relationship extraction model may be preferentially used to extract the relationship, and the syntactic dependency relationship is used to extract the relationship when the obtained relationship is of an unknown type or has a low relationship confidence, so that a new relationship type may be effectively mined, and a relationship set may be enriched.
Fig. 6 shows an application example diagram of extracting entity relationships in the present embodiment. As shown in fig. 6, in the present example, corresponding training texts may be matched from a mass of free texts according to entity relationship triples in a known database; training a Bigru-attention model based on the training texts, and performing model learning to obtain a trained model, namely an entity relationship extraction model; inputting a given target entity pair and a target text into an entity relation extraction model to obtain an extraction result to be selected; under the condition that the extraction result to be selected is not empty, outputting the extraction result to be selected, namely taking the output result of the entity relationship extraction model as the relationship of the target entity pair in the target text; and under the condition that the extraction result to be selected is empty or the relation confidence coefficient is low, carrying out syntactic analysis learning on the target text by utilizing syntactic dependency to obtain an output result.
In one implementation, fig. 7 is a block diagram of a training apparatus for an entity relationship extraction model according to an embodiment of the present application. As shown in fig. 7, the training apparatus includes:
a matching module 701, configured to match a training text according to an entity relationship triple in a known database, where the entity relationship triple includes a known entity pair and a known relationship corresponding to the known entity pair, and the training text includes the known entity pair;
a prediction relationship determining module 702, configured to predict, based on the initial model, a relationship of a known entity pair in the training text, so as to obtain a prediction relationship;
an adjusting module 703 is configured to adjust parameters of the initial model according to a difference between a known relationship and a predicted relationship corresponding to the plurality of training texts, so as to obtain an entity relationship extraction model, where the entity relationship extraction model is used to predict a relationship of the target entity pair in the target text.
In one embodiment, the initial model is a bi-directional gated loop unit-attention model.
In one embodiment, the predictive relationship determination module 702 includes: the training character determining unit is used for determining a plurality of training characters from the training text, wherein the training characters comprise training reference characters matched with the known entity pairs; the first entity position characteristic determining unit is used for obtaining entity position characteristics of training characters based on the position relation between the training characters and the training reference characters; and the first input unit is used for inputting the character characteristics and the entity position characteristics of each training character into the initial model and predicting the relation of the known entity pair in the training text.
In one implementation, fig. 8 illustrates an apparatus for extracting entity relationships according to an embodiment of the present application. As shown in fig. 8, the apparatus includes:
the model prediction module 801 is used for predicting the relation of a target entity pair in a target text based on an entity relation extraction model to obtain an extraction result to be selected, wherein the entity relation extraction model is obtained by training according to any one of the above training methods;
the judging module 802 is configured to judge whether the to-be-selected extraction result meets a preset condition;
and the syntactic dependency analysis module 803 is configured to, when it is determined that the to-be-selected extraction result does not satisfy the preset condition, perform syntactic dependency analysis on the target text to determine a relationship of the target entity pair in the target text.
In one embodiment, the model prediction module 801 includes: the target character determining unit is used for determining a plurality of target characters from the target text, wherein the target characters comprise target reference characters matched with the target entity pairs; the second entity position characteristic determining unit is used for obtaining the entity position characteristic of the target character based on the position relation between the target character and the target reference character; and the second input unit is used for inputting the character characteristics and the entity position characteristics of each target character into the entity relationship extraction model and predicting the relationship of the target entity pair in the target text.
Fig. 9 is a block diagram of an electronic device according to an embodiment of the present application, illustrating a method for training an entity relationship extraction model. The electronic device is intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the present application that are described and/or claimed herein.
As shown in fig. 9, the electronic apparatus includes: one or more first processors 901, a first memory 902, and interfaces for connecting the respective components, including a high-speed interface and a low-speed interface. The various components are interconnected using different buses and may be mounted on a common motherboard or in other manners as desired. The first processor may process instructions for execution within the electronic device, including instructions stored in or on the first memory to display Graphical information for a Graphical User Interface (GUI) on an external input/first output device, such as a display device coupled to the Interface. In other embodiments, multiple first processors and/or multiple buses may be used, along with multiple first memories and multiple first memories, if desired. Also, multiple electronic devices may be connected, with each device providing portions of the necessary operations (e.g., as a server array, a group of blade servers, or multiple first processor systems). In fig. 9, a first processor 901 is taken as an example.
The first memory 902 is a non-transitory computer readable storage medium as provided herein. The first memory stores instructions executable by at least one first processor to cause the at least one first processor to perform the method for training the entity relationship extraction model provided herein. The non-transitory computer-readable storage medium of the present application stores computer instructions for causing a computer to perform the navigation method provided herein.
The first memory 902, which is a non-transitory computer readable storage medium, may be used to store non-transitory software programs, non-transitory computer executable programs, and modules, such as program instructions/modules corresponding to the training method of the entity relationship extraction model in the embodiment of the present application (for example, the matching module 701, the predicted relationship determination module 702, and the adjustment module 703 shown in fig. 7). The first processor 901 executes various functional applications of the server and data processing, i.e., implements the training method of the entity relationship extraction model in the above embodiments, by running non-transitory software programs, instructions, and modules stored in the first memory 902.
The first memory 902 may include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function; the storage data area may store data created by use of the electronic device according to the training method of the entity relationship extraction model, and the like. Further, the first memory 902 may include a high speed random access first memory and may also include a non-transitory first memory, such as at least one magnetic disk first storage device, flash memory device, or other non-transitory solid state first storage device. In some embodiments, the first memory 902 may optionally include a first memory remotely disposed from the first processor 901, and these remote first memories may be connected to an electronic device of the training method of the entity-relationship extraction model through a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
The electronic device of the training method of the entity relationship extraction model may further include: a first input device 903 and a first output device 904. The first processor 901, the first memory 902, the first input device 903 and the first output device 904 may be connected by a bus or other means, and fig. 9 illustrates an example of connection by a bus.
The first input device 903 may receive input numeric or character information and generate key signal inputs related to user settings and function control of the electronic device of the training method of the entity-relationship extraction model, such as a first input device of a touch screen, a keypad, a mouse, a track pad, a touch pad, a pointing stick, one or more mouse buttons, a track ball, a joystick, or the like. The first output device 904 may include a display device, an auxiliary lighting device (e.g., an LED), a haptic feedback device (e.g., a vibration motor), and the like. The display device may include, but is not limited to, a Liquid Crystal Display (LCD) such as a Liquid crystal Cr9 star display 9, a Light Emitting Diode (LED) display, and a plasma display. In some implementations, the display device can be a touch screen.
Fig. 10 is a block diagram of an electronic device according to an embodiment of the present application, illustrating a method for extracting entity relationships. The electronic device is intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the present application that are described and/or claimed herein.
As shown in fig. 10, the electronic apparatus includes: one or more second processors 1001, a second memory 1002, and interfaces for connecting the components, including a high-speed interface and a low-speed interface. The various components are interconnected using different buses and may be mounted on a common motherboard or in other manners as desired. The second processor may process instructions for execution within the electronic device, including instructions stored in or on the second memory to display Graphical information for a Graphical User Interface (GUI) on an external input/second output device, such as a display device coupled to the interface. In other embodiments, multiple second processors and/or multiple buses may be used, along with multiple second memories and multiple second memories, if desired. Also, multiple electronic devices may be connected, with each device providing portions of the necessary operations (e.g., as a server array, a group of blade servers, or multiple secondary processor systems). Fig. 10 illustrates an example of a second processor 1001.
The second memory 1002 is a non-transitory computer readable storage medium provided herein. The second memory stores instructions executable by at least one second processor to cause the at least one second processor to perform the method for extracting entity relationships provided herein. The non-transitory computer-readable storage medium of the present application stores computer instructions for causing a computer to perform the navigation method provided herein.
The second memory 1002, which is a non-transitory computer readable storage medium, may be used for storing non-transitory software programs, non-transitory computer executable programs, and modules, such as program instructions/modules corresponding to the method for extracting entity relationships in the embodiments of the present application (for example, the model prediction module 801, the judgment module 802, and the syntactic dependency analysis module 803 shown in fig. 8). The second processor 1001 executes various functional applications of the server and data processing, i.e., implements the method of extracting entity relationships in the above-described embodiments, by running non-transitory software programs, instructions, and modules stored in the second memory 1002.
The second memory 1002 may include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function; the storage data area may store data created according to use of the electronic device of the method of extracting entity relationships, and the like. In addition, the second memory 1002 can include a high speed random access second memory, and can also include a non-transitory second memory, such as at least one magnetic disk second storage device, flash memory device, or other non-transitory solid state second storage device. In some embodiments, the second memory 1002 may optionally include a second memory remotely located from the second processor 1001, and these remote second memories may be connected via a network to an electronic device of the method of extracting entity relationships. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
The electronic device of the method of extracting entity relationships may further include: a second input device 1003 and a second output device 1004. The second processor 1001, the second memory 1002, the second input device 1003, and the second output device 1004 may be connected by a bus or other means, and the bus connection is exemplified in fig. 10.
The second input means 1003 may receive input numeric or character information and generate key signal inputs related to user settings and function control of the electronic device of the method of extracting entity relationships, such as a touch screen, a keypad, a mouse, a track pad, a touch pad, a pointer, one or more mouse buttons, a track ball, a joystick, etc. The second output device 1004 may include a display apparatus, an auxiliary lighting device (e.g., an LED), a haptic feedback device (e.g., a vibration motor), and the like. The display device may include, but is not limited to, a Liquid Crystal Display (LCD) such as a Cr10sta display 10, a Light Emitting Diode (LED) display, and a plasma display. In some implementations, the display device can be a touch screen.
Various implementations of the systems and techniques described here can be realized in digital electronic circuitry, Integrated circuitry, Application Specific Integrated Circuits (ASICs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs that are executable and/or interpretable on a programmable system including at least one programmable processor, which may be special or general purpose, receiving data and instructions from, and transmitting data and instructions to, a storage system, at least one input device, and at least one output device.
These computer programs (also known as programs, software applications, or code) include machine instructions for a programmable processor, and may be implemented using high-level procedural and/or object-oriented programming languages, and/or assembly/machine languages. As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, apparatus, and/or device (e.g., magnetic discs, optical disks, memory, Programmable Logic Devices (PLDs)) used to provide machine instructions and/or data to a programmable processor, including a machine-readable medium that receives machine instructions as a machine-readable signal. The term "machine-readable signal" refers to any signal used to provide machine instructions and/or data to a programmable processor.
To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having: a display device (e.g., a CRT (Cathode Ray Tube) or LCD (liquid crystal display) monitor) for displaying information to a user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which a user can provide input to the computer. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form, including acoustic, speech, or tactile input.
The systems and techniques described here can be implemented in a computing system that includes a back-end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), Wide Area Networks (WANs), and the internet.
The computer system may include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other.
According to the technical scheme of the embodiment of the application, the entity relationship extraction model is trained by constructing the entity relationship triples for the known database and further matching the corresponding training texts from the mass free texts. Because the entity pairs and the entity relations in the entity relation triplets are known, the entity relations in the training texts do not need to be manually marked, and the model can be conveniently and quickly trained so as to improve the efficiency of entity relation extraction. Furthermore, by introducing syntactic dependency relationship extraction, new relationship types can be effectively mined, relationships outside the database relationship set are supplemented, and the relationship set is enriched.
It should be understood that various forms of the flows shown above may be used, with steps reordered, added, or deleted. For example, the steps described in the present application may be executed in parallel, sequentially, or in different orders, and the present invention is not limited thereto as long as the desired results of the technical solutions disclosed in the present application can be achieved.
The above-described embodiments should not be construed as limiting the scope of the present application. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may be made in accordance with design requirements and other factors. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.
Claims (11)
1. A method for training an entity relationship extraction model is characterized by comprising the following steps:
matching a training text according to an entity relationship triple in a known database, wherein the entity relationship triple comprises a known entity pair and a known relationship corresponding to the known entity pair, and the training text comprises the known entity pair;
predicting the relation of the known entity pair in the training text based on an initial model to obtain a predicted relation;
and adjusting parameters of the initial model according to the difference between the known relations and the predicted relations corresponding to the training texts to obtain an entity relation extraction model, wherein the entity relation extraction model is used for predicting the relation of a target entity in the target text.
2. The training method of claim 1, wherein the initial model is a bi-directional gated loop unit-attention model.
3. The training method of claim 1, wherein predicting the relationship of the known entity pair in the training text based on an initial model comprises:
determining a plurality of training characters from the training text, wherein the training characters comprise training reference characters matched with the known entity pairs;
obtaining entity position characteristics of the training characters based on the position relation between the training characters and the training reference characters;
inputting the character features and entity position features of the training characters into the initial model, and predicting the relation of the known entity pairs in the training text.
4. A method for extracting entity relationships, comprising:
predicting the relation of a target entity pair in a target text based on an entity relation extraction model to obtain an extraction result to be selected, wherein the entity relation extraction model is obtained by training according to the training method of any one of claims 1 to 3;
judging whether the extraction result to be selected meets a preset condition or not;
and under the condition that the extraction result to be selected is judged not to meet the preset condition, performing syntactic dependency analysis on the target text to determine the relation of the target entity pair in the target text.
5. The method of claim 4, wherein predicting the relationship of the target entity pair in the target text based on the entity relationship extraction model comprises:
determining a plurality of target characters from the target text, wherein the target characters comprise target reference characters matched with the target entity pairs;
obtaining entity position characteristics of the target character based on the position relation between the target character and the target reference character;
inputting the character features and the entity position features of the target characters into the entity relationship extraction model, and predicting the relationship of the target entities in the target text.
6. An apparatus for training an entity-relationship extraction model, comprising:
the matching module is used for matching a training text according to an entity relationship triple in a known database, wherein the entity relationship triple comprises a known entity pair and a known relationship corresponding to the known entity pair, and the training text comprises the known entity pair;
the prediction relation determining module is used for predicting the relation of the known entity pair in the training text based on an initial model to obtain a prediction relation;
and the adjusting module is used for adjusting the parameters of the initial model according to the difference between the known relations and the predicted relations corresponding to the training texts to obtain an entity relation extraction model, and the entity relation extraction model is used for predicting the relation of the target entity pair in the target text.
7. The training apparatus of claim 6, wherein the initial model is a bi-directional gated loop unit-attention model.
8. An apparatus for extracting entity relationships, comprising:
the model prediction module is used for predicting the relation of a target entity pair in a target text based on an entity relation extraction model to obtain an extraction result to be selected, wherein the entity relation extraction model is obtained by training according to the training method of any one of claims 1 to 3;
the judging module is used for judging whether the extraction result to be selected meets a preset condition or not;
and the syntactic dependency analysis module is used for performing syntactic dependency analysis on the target text to determine the relation of the target entity pair in the target text under the condition that the to-be-selected extraction result is judged not to meet the preset condition.
9. An electronic device, comprising:
at least one processor; and
a memory communicatively coupled to the at least one processor; wherein the content of the first and second substances,
the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-3.
10. An electronic device, comprising:
at least one processor; and
a memory communicatively coupled to the at least one processor; wherein the content of the first and second substances,
the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method recited in claim 4 or 5.
11. A non-transitory computer readable storage medium having stored thereon computer instructions for causing the computer to perform the method of any one of claims 1-5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910883330.2A CN110619053A (en) | 2019-09-18 | 2019-09-18 | Training method of entity relation extraction model and method for extracting entity relation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910883330.2A CN110619053A (en) | 2019-09-18 | 2019-09-18 | Training method of entity relation extraction model and method for extracting entity relation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110619053A true CN110619053A (en) | 2019-12-27 |
Family
ID=68923411
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910883330.2A Pending CN110619053A (en) | 2019-09-18 | 2019-09-18 | Training method of entity relation extraction model and method for extracting entity relation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110619053A (en) |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111274391A (en) * | 2020-01-15 | 2020-06-12 | 北京百度网讯科技有限公司 | SPO extraction method and device, electronic equipment and storage medium |
| CN111274407A (en) * | 2020-01-15 | 2020-06-12 | 北京百度网讯科技有限公司 | Triple confidence degree calculation method and device in knowledge graph |
| CN111339314A (en) * | 2020-02-19 | 2020-06-26 | 北京百度网讯科技有限公司 | Method and device for generating triple-group data and electronic equipment |
| CN111400451A (en) * | 2020-03-16 | 2020-07-10 | 北京百度网讯科技有限公司 | Information extraction method, information extraction device and electronic equipment |
| CN111414482A (en) * | 2020-03-20 | 2020-07-14 | 北京百度网讯科技有限公司 | Event argument extraction method and device and electronic equipment |
| CN111444349A (en) * | 2020-03-06 | 2020-07-24 | 深圳追一科技有限公司 | Information extraction method and device, computer equipment and storage medium |
| CN111539209A (en) * | 2020-04-15 | 2020-08-14 | 北京百度网讯科技有限公司 | Method and apparatus for entity classification |
| CN111694967A (en) * | 2020-06-11 | 2020-09-22 | 腾讯科技（深圳）有限公司 | Attribute extraction method and device, electronic equipment and medium |
| CN111814476A (en) * | 2020-06-09 | 2020-10-23 | 北京捷通华声科技股份有限公司 | Method and device for extracting entity relationship |
| CN112464637A (en) * | 2020-11-27 | 2021-03-09 | 平安科技（深圳）有限公司 | Label-based optimization model training method, device, equipment and storage medium |
| CN112487206A (en) * | 2020-12-09 | 2021-03-12 | 中国电子科技集团公司第三十研究所 | Entity relationship extraction method for automatically constructing data set |
| CN112836064A (en) * | 2021-02-24 | 2021-05-25 | 吉林大学 | Knowledge graph complementing method and device, storage medium and electronic equipment |
| CN113051356A (en) * | 2021-04-21 | 2021-06-29 | 深圳壹账通智能科技有限公司 | Open relationship extraction method and device, electronic equipment and storage medium |
| CN113190686A (en) * | 2021-04-25 | 2021-07-30 | 浙江大学软件学院（宁波）管理中心（宁波软件教育中心） | Knowledge graph relation completion method for equipment maintenance of numerical control machine tool |
| CN113836316A (en) * | 2021-09-23 | 2021-12-24 | 北京百度网讯科技有限公司 | Processing method, training method, device, equipment and medium for ternary group data |
| CN114386411A (en) * | 2020-10-16 | 2022-04-22 | 北京金山数字娱乐科技有限公司 | Relationship extraction method and device |
| CN114595686A (en) * | 2022-03-11 | 2022-06-07 | 北京百度网讯科技有限公司 | Knowledge extraction method, and training method and device of knowledge extraction model |
| CN116069956A (en) * | 2023-03-29 | 2023-05-05 | 之江实验室 | Drug knowledge graph entity alignment method and device based on mixed attention mechanism |
| CN116522935A (en) * | 2023-03-29 | 2023-08-01 | 北京德风新征程科技股份有限公司 | Text data processing method, processing device and electronic equipment |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180157643A1 (en) * | 2016-12-06 | 2018-06-07 | Siemens Aktiengesellschaft | Device and method for natural language processing |
| WO2019085328A1 (en) * | 2017-11-02 | 2019-05-09 | 平安科技（深圳）有限公司 | Enterprise relationship extraction method and device, and storage medium |
| CN109902171A (en) * | 2019-01-30 | 2019-06-18 | 中国地质大学（武汉） | Text Relation extraction method and system based on layering knowledge mapping attention model |
| CN110134772A (en) * | 2019-04-18 | 2019-08-16 | 五邑大学 | Medical text Relation extraction method based on pre-training model and fine tuning technology |
-
2019
- 2019-09-18 CN CN201910883330.2A patent/CN110619053A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180157643A1 (en) * | 2016-12-06 | 2018-06-07 | Siemens Aktiengesellschaft | Device and method for natural language processing |
| WO2019085328A1 (en) * | 2017-11-02 | 2019-05-09 | 平安科技（深圳）有限公司 | Enterprise relationship extraction method and device, and storage medium |
| CN109902171A (en) * | 2019-01-30 | 2019-06-18 | 中国地质大学（武汉） | Text Relation extraction method and system based on layering knowledge mapping attention model |
| CN110134772A (en) * | 2019-04-18 | 2019-08-16 | 五邑大学 | Medical text Relation extraction method based on pre-training model and fine tuning technology |
Non-Patent Citations (1)
| Title |
|---|
| 佘恒: "《基于深度学习的中文文本实体关系抽取研究与实现》", 《中国优秀博硕士学位论文全文数据库(硕士)信息科技辑》 * |
Cited By (38)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20210216819A1 (en) * | 2020-01-15 | 2021-07-15 | Beijing Baidu Netcom Science And Technology Co., Ltd. | Method, electronic device, and storage medium for extracting spo triples |
| CN111274407A (en) * | 2020-01-15 | 2020-06-12 | 北京百度网讯科技有限公司 | Triple confidence degree calculation method and device in knowledge graph |
| CN111274391B (en) * | 2020-01-15 | 2023-09-01 | 北京百度网讯科技有限公司 | SPO extraction method and device, electronic equipment and storage medium |
| CN111274391A (en) * | 2020-01-15 | 2020-06-12 | 北京百度网讯科技有限公司 | SPO extraction method and device, electronic equipment and storage medium |
| CN111339314A (en) * | 2020-02-19 | 2020-06-26 | 北京百度网讯科技有限公司 | Method and device for generating triple-group data and electronic equipment |
| CN111339314B (en) * | 2020-02-19 | 2024-02-13 | 北京百度网讯科技有限公司 | Ternary group data generation method and device and electronic equipment |
| CN111444349A (en) * | 2020-03-06 | 2020-07-24 | 深圳追一科技有限公司 | Information extraction method and device, computer equipment and storage medium |
| CN111444349B (en) * | 2020-03-06 | 2023-09-12 | 深圳追一科技有限公司 | Information extraction method, information extraction device, computer equipment and storage medium |
| CN111400451A (en) * | 2020-03-16 | 2020-07-10 | 北京百度网讯科技有限公司 | Information extraction method, information extraction device and electronic equipment |
| CN111400451B (en) * | 2020-03-16 | 2023-05-09 | 北京百度网讯科技有限公司 | Information extraction method, information extraction device and electronic equipment |
| US11880397B2 (en) | 2020-03-20 | 2024-01-23 | Beijing Baidu Netcom Science Technology Co., Ltd. | Event argument extraction method, event argument extraction apparatus and electronic device |
| CN111414482B (en) * | 2020-03-20 | 2024-02-20 | 北京百度网讯科技有限公司 | Event argument extraction method and device and electronic equipment |
| CN111414482A (en) * | 2020-03-20 | 2020-07-14 | 北京百度网讯科技有限公司 | Event argument extraction method and device and electronic equipment |
| CN111539209B (en) * | 2020-04-15 | 2023-09-15 | 北京百度网讯科技有限公司 | Method and apparatus for entity classification |
| CN111539209A (en) * | 2020-04-15 | 2020-08-14 | 北京百度网讯科技有限公司 | Method and apparatus for entity classification |
| CN111814476B (en) * | 2020-06-09 | 2024-04-16 | 北京捷通华声科技股份有限公司 | Entity relation extraction method and device |
| CN111814476A (en) * | 2020-06-09 | 2020-10-23 | 北京捷通华声科技股份有限公司 | Method and device for extracting entity relationship |
| CN111694967B (en) * | 2020-06-11 | 2023-10-20 | 腾讯科技（深圳）有限公司 | Attribute extraction method, attribute extraction device, electronic equipment and medium |
| CN111694967A (en) * | 2020-06-11 | 2020-09-22 | 腾讯科技（深圳）有限公司 | Attribute extraction method and device, electronic equipment and medium |
| CN114386411A (en) * | 2020-10-16 | 2022-04-22 | 北京金山数字娱乐科技有限公司 | Relationship extraction method and device |
| CN112464637B (en) * | 2020-11-27 | 2024-02-27 | 平安科技（深圳）有限公司 | Label-based optimization model training method, device, equipment and storage medium |
| CN112464637A (en) * | 2020-11-27 | 2021-03-09 | 平安科技（深圳）有限公司 | Label-based optimization model training method, device, equipment and storage medium |
| CN112487206B (en) * | 2020-12-09 | 2022-09-20 | 中国电子科技集团公司第三十研究所 | Entity relationship extraction method for automatically constructing data set |
| CN112487206A (en) * | 2020-12-09 | 2021-03-12 | 中国电子科技集团公司第三十研究所 | Entity relationship extraction method for automatically constructing data set |
| CN112836064B (en) * | 2021-02-24 | 2023-05-16 | 吉林大学 | Knowledge graph completion method and device, storage medium and electronic equipment |
| CN112836064A (en) * | 2021-02-24 | 2021-05-25 | 吉林大学 | Knowledge graph complementing method and device, storage medium and electronic equipment |
| WO2022222300A1 (en) * | 2021-04-21 | 2022-10-27 | 深圳壹账通智能科技有限公司 | Open relationship extraction method and apparatus, electronic device, and storage medium |
| CN113051356B (en) * | 2021-04-21 | 2023-05-30 | 深圳壹账通智能科技有限公司 | Open relation extraction method and device, electronic equipment and storage medium |
| CN113051356A (en) * | 2021-04-21 | 2021-06-29 | 深圳壹账通智能科技有限公司 | Open relationship extraction method and device, electronic equipment and storage medium |
| CN113190686A (en) * | 2021-04-25 | 2021-07-30 | 浙江大学软件学院（宁波）管理中心（宁波软件教育中心） | Knowledge graph relation completion method for equipment maintenance of numerical control machine tool |
| CN113836316A (en) * | 2021-09-23 | 2021-12-24 | 北京百度网讯科技有限公司 | Processing method, training method, device, equipment and medium for ternary group data |
| CN113836316B (en) * | 2021-09-23 | 2023-01-03 | 北京百度网讯科技有限公司 | Processing method, training method, device, equipment and medium for ternary group data |
| CN114595686A (en) * | 2022-03-11 | 2022-06-07 | 北京百度网讯科技有限公司 | Knowledge extraction method, and training method and device of knowledge extraction model |
| CN114595686B (en) * | 2022-03-11 | 2023-02-03 | 北京百度网讯科技有限公司 | Knowledge extraction method, and training method and device of knowledge extraction model |
| CN116069956A (en) * | 2023-03-29 | 2023-05-05 | 之江实验室 | Drug knowledge graph entity alignment method and device based on mixed attention mechanism |
| CN116522935A (en) * | 2023-03-29 | 2023-08-01 | 北京德风新征程科技股份有限公司 | Text data processing method, processing device and electronic equipment |
| CN116522935B (en) * | 2023-03-29 | 2024-03-29 | 北京德风新征程科技股份有限公司 | Text data processing method, processing device and electronic equipment |
| CN116069956B (en) * | 2023-03-29 | 2023-07-04 | 之江实验室 | Drug knowledge graph entity alignment method and device based on mixed attention mechanism |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110619053A (en) | Training method of entity relation extraction model and method for extracting entity relation | |
| CN111523326B (en) | Entity chain finger method, device, equipment and storage medium | |
| CN111428507B (en) | Entity chain finger method, device, equipment and storage medium | |
| EP3852001A1 (en) | Method and apparatus for generating temporal knowledge graph, device, and medium | |
| US20210312139A1 (en) | Method and apparatus of generating semantic feature, method and apparatus of training model, electronic device, and storage medium | |
| JP7108675B2 (en) | Semantic matching method, device, electronic device, storage medium and computer program | |
| US11507751B2 (en) | Comment information processing method and apparatus, and medium | |
| US11537792B2 (en) | Pre-training method for sentiment analysis model, and electronic device | |
| US20210200813A1 (en) | Human-machine interaction method, electronic device, and storage medium | |
| CN112507101B (en) | Method and device for establishing pre-training language model | |
| CN111079945B (en) | End-to-end model training method and device | |
| CN111709234A (en) | Training method and device of text processing model and electronic equipment | |
| US20210191938A1 (en) | Summarized logical forms based on abstract meaning representation and discourse trees | |
| CN112528001B (en) | Information query method and device and electronic equipment | |
| US11704326B2 (en) | Generalization processing method, apparatus, device and computer storage medium | |
| CN111737954A (en) | Text similarity determination method, device, equipment and medium | |
| US11321370B2 (en) | Method for generating question answering robot and computer device | |
| CN113657100A (en) | Entity identification method and device, electronic equipment and storage medium | |
| CN111708800A (en) | Query method and device and electronic equipment | |
| US11468236B2 (en) | Method and apparatus for performing word segmentation on text, device, and medium | |
| CN112541359A (en) | Document content identification method and device, electronic equipment and medium | |
| CN111640511A (en) | Medical fact verification method and device, electronic equipment and storage medium | |
| CN111539209A (en) | Method and apparatus for entity classification | |
| CN111966781A (en) | Data query interaction method and device, electronic equipment and storage medium | |
| CN114444462B (en) | Model training method and man-machine interaction method and device |
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 |