CN111881683A - Method and device for generating relation triples, storage medium and electronic equipment - Google Patents

Abstract

The present application provides a method, device, storage medium and electronic device for generating relation triples, wherein the method includes: acquiring a representation code corresponding to input text; identifying candidate subjects from the representation code; judging the representation Whether there is an object that has a target semantic relationship with the candidate subject in the encoding; if there is, a relationship triplet is generated according to the candidate subject, the target semantic relationship and the object. By adopting the solutions of the embodiments of the present application, the relation triples in the input text can be recognized comprehensively and accurately.

Description

Method, device, storage medium and electronic device for generating relation triples

technical field

The present application relates to the technical field of information processing, and in particular, to a method, apparatus, storage medium and electronic device for generating relation triples.

Background technique

The basic components of knowledge graphs are relational facts, which take the form of (subject, relation, object) to connect two entities through semantic relations, called relation triples. Extracting relational triples from natural language texts is a key step in building large-scale knowledge graphs. However, the current methods for generating relational triples are not comprehensive and accurate enough.

SUMMARY OF THE INVENTION

In order to solve the above problems, the embodiments of the present application provide a method, device, storage medium and electronic device for generating a relation triple. The technical solution is as follows:

In a first aspect, an embodiment of the present application provides a method for generating a relation triple, including the following steps:

Get the representation code corresponding to the input text;

identifying candidate subjects from the representation encoding;

Judging whether there is an object having a target semantic relationship with the candidate subject in the representation encoding;

If there is, a relation triple is generated according to the candidate subject, the semantic relation and the object.

Optionally, the obtaining the representation code corresponding to the input text includes:

get input text;

The input text is encoded by the BERT encoder to generate a representation code corresponding to the input text.

Optionally, the judging whether the representation encoding exists after the object with the semantic relationship of the candidate subject further includes:

If not, it is determined that the candidate subject cannot form a relation triple based on the target semantic relation.

Optionally, the identifying a candidate subject from the representation encoding includes:

identifying a plurality of candidate subjects from the representation encoding using a subject tagger;

The judging whether there is an object having a target semantic relationship with the candidate subject in the representation encoding includes:

judging whether there are objects that have a semantic relationship with each candidate subject in the representation code;

If there is, according to the candidate subject, the semantic relationship and the object, a relationship triple is generated, including:

If there is, at least one relation triple is generated from each of the candidate subjects, the semantic relation and each of the objects.

Optionally, the judging whether there is an object having a target semantic relationship with the candidate subject in the representation encoding includes:

Using the object tagger corresponding to the target semantic relationship, it is judged whether there is an object having the target semantic relationship with the candidate subject in the representation encoding.

Optionally, the use of the object tagger corresponding to the target semantic relationship to determine whether there is an object having the target semantic relationship with the candidate subject in the representation encoding includes:

Using a plurality of object taggers, it is judged in parallel whether there is an object having a target semantic relationship with the candidate subject in the representation encoding, and each object tagger in the plurality of object taggers corresponds to a different target semantic relationship.

Optionally, the use of the object tagger corresponding to the target semantic relationship to determine whether there is an object having the target semantic relationship with the candidate subject in the representation encoding includes:

Using the object tagger corresponding to the target semantic relationship, based on the encoded representation of each target word in the representation encoding and the encoded representation of the candidate subject, calculate the probability that each target word corresponds to the starting position of the object;

Using the object tagger corresponding to the target semantic relationship, based on the encoded representation of each target word in the representation encoding and the encoded representation of the candidate subject, calculate the probability that each target word corresponds to the end position of the object;

According to the probability that each target word corresponds to the starting position of the object and the probability that each target word corresponds to the ending position of the object, it is determined whether there is an object having the target semantic relationship with the candidate subject in the representation encoding.

In a second aspect, an embodiment of the present application provides an apparatus for generating a relation triple, including:

an encoding acquisition unit, used to acquire the representation encoding corresponding to the input text;

a subject identification unit for identifying candidate subjects from the representation encoding;

an object judgment unit, used for judging whether there is an object having a target semantic relationship with the candidate subject in the representation code;

A relation generating unit, configured to generate a relation triplet according to the candidate subject, the semantic relation and the object if there is one.

In a third aspect, an embodiment of the present application provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, implements the steps of any one of the foregoing methods.

In a fourth aspect, an embodiment of the present application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, where the processor implements any of the above when executing the program steps of the method.

In the embodiment of the present application, the representation code corresponding to the input text is obtained; the candidate subject is identified from the representation code; it is judged whether there is an object that has a target semantic relationship with the candidate subject in the representation code; The candidate subject, the target semantic relation and the object generate relation triples. Based on the candidate subject, it is determined whether there is an object that has a target semantic relationship with the candidate subject in the representation encoding. By adopting the solutions of the embodiments of the present application, the relation triples in the input text can be recognized comprehensively and accurately.

Description of drawings

Fig. 1 provides different relation triple overlapping modes provided by the embodiment of the present application;

2 is a schematic flowchart of a method for generating a relation triple according to an embodiment of the present application;

3 is a schematic diagram of the overall framework structure of another method for generating relation triples provided by an embodiment of the present application;

4 is a schematic flowchart of another method for generating a relation triplet provided by an embodiment of the present application;

5 is a schematic structural diagram of an apparatus for generating a relation triplet provided by an embodiment of the present application;

FIG. 6 is a schematic structural diagram of an electronic device involved in an embodiment of the present application.

Detailed ways

The present application will be further introduced below with reference to the accompanying drawings and embodiments.

In the following introduction, the terms "first" and "second" are used for descriptive purposes only, and should not be construed as indicating or implying relative importance. The following description provides multiple embodiments of the present application, and different embodiments may be substituted or combined, so the present application may also be considered to include all possible combinations of the same and/or different embodiments described. Thus, if one embodiment includes features A, B, C and another embodiment includes features B, D, the application should also be considered to include all other possible combinations of one or more of A, B, C, D example, although this example may not be explicitly described in the following content.

The following description provides examples, and does not limit the scope, applicability, or examples set forth in the claims. Changes may be made in the function and arrangement of elements described without departing from the scope of the present disclosure. Various examples may omit, substitute or add various procedures or components as appropriate. For example, the methods described may be performed in an order different from that described, and various steps may be added, omitted, or combined. Furthermore, features described with respect to some examples may be combined in other examples.

Relational triples can be generated by the pipeline method and the joint method. The pipelined method divides the relation triple extraction task into two independent steps. First, all entities are extracted from the sentence, and the identified entities are grouped into entity pairs, and then the relation classification method is used to determine the relationship between the entities. We end up with relational triples. Different from pipeline methods, joint methods aim to propose a joint model to simultaneously extract entities and their semantic relations, including feature-based joint methods and neural network-based joint methods. At present, the neural network-based joint method is a relatively advanced method in the field of relation triple extraction, and the method provided by the embodiments of the present application belongs to the neural network-based joint method.

The pipelined method divides relation triple extraction into two independent steps, ignoring the correlation between the two steps of entity extraction and relation classification, and there will inevitably be an error propagation problem, that is, the previous step (entity extraction). ) can affect the performance of the latter step (relation classification). The feature-based joint method relies heavily on third-party natural language processing tools, requires tedious feature engineering, and consumes a lot of manpower and resources. Existing joint methods based on neural networks use neural networks for feature extraction without the need to manually design features, and achieve a certain improvement in relational triplet extraction tasks. With the continuous development of the field of natural language processing, the extraction of relational triples in simple contexts (for example, a sentence contains only one relational triplet) has been able to achieve good results. However, existing methods, whether pipelined or federated, typically model relationships as discrete labels between pairs of entities. In complex contexts (a sentence contains multiple relation triples, sometimes as many as five or more), especially when multiple triples overlap, relation classification becomes a The extremely difficult unbalanced multi-classification problem leads to the fact that the final extracted relation triples are not comprehensive and accurate. Therefore, none of the existing methods can deal well with the problem of relation triple extraction in complex contexts.

The goal of relation triple extraction task is to identify all possible (subject, relation, object) triples from a sentence, some of which may share some subject or object, i.e. triples There may be overlapping issues between them. According to the different overlapping forms between triples, it is divided into three types: normal (Normal), single entity overlap (SEO, Single Entity Overlap) and entity pair overlap (EPO, Entity Pair Overlap). FIG. 1 provides different overlapping manners of relation triples according to an embodiment of the present application. As shown in Figure 1, from top to bottom, they are normal type, entity pair overlap type and single entity overlap type.

Based on the above analysis, the embodiments of the present application provide a novel cascaded binary labeling method and device for relation triplet extraction. Different from traditional methods that model relationships as discrete labels between pairs of entities, the embodiments of the present application propose to model relationships as functions that map from subject to object, which provides a brand-new knowledge graph research community. From the perspective of this paper, we re-examine the classic relational triplet extraction task, and on this basis, we implement a novel cascaded binary annotation framework that is not troubled by the overlapping triplet problem for relational triplet extraction.

Referring to FIG. 2, FIG. 2 is a schematic flowchart of a method for generating a relation triple according to an embodiment of the present application, and the method includes:

S201. Obtain a representation code corresponding to the input text.

Indicates that the encoding is an encoding generated from the input text. A corresponding representation code can be generated from the input text through a preset method, and subsequent processing processes are all performed based on the representation code.

Optionally, S201 may include:

get input text;

The input text is encoded by the BERT encoder to generate a representation code corresponding to the input text.

The encoder module aims to extract rich semantic feature information from the input sentence, and the embodiment of the present application uses a pre-trained BERT model to encode the context information. Among them, the BERT model is a language representation model based on a multi-layer bidirectional Transformer structure, which can simultaneously consider the left and right context information of each word to learn the deep representation features of the text. After the input text is encoded by the BERT encoder, the resulting N-th layer encoded representation h _N will be input to the cascaded decoder, and finally the relation triples contained in the sentence are decoded.

S202. Identify a candidate subject from the representation code.

The candidate subject may be identified by decoding the representation code obtained in S201. At least one candidate subject is identified. Candidate subjects can be nouns or pronouns representing people, things, things, places, or abstract concepts.

S203. Determine whether there is an object having a target semantic relationship with the candidate subject in the representation encoding.

The target semantic relation is used to describe the relational fact between the subject and the object. Multiple semantic relationships can exist in the same representation encoding. Objects can be nouns or pronouns that represent people, things, things, places, or abstract concepts.

Different from the existing method, the method of the embodiment of the present application uses a function of mapping a subject to an object to determine whether there is an object having a target semantic relationship with the candidate subject in the representation encoding. Identify possible objects in the token encoding given the target semantic relation and target subject.

S204. If there is, generate a relation triplet according to the candidate subject, the semantic relation and the object.

Relation triples can be used to represent the relationship facts between subjects and objects, and relation triples can take the form of (subject, relation, object), abbreviated as (s, r, o) To represent.

Optionally, after S203, it also includes:

If not, it is determined that the candidate subject cannot form a relation triple based on the target semantic relation.

If no possible object can be identified in the identification code given the target semantic relationship and the target subject, it can be judged that the candidate subject is not a real subject and cannot form a relation triple.

Optionally, S203 may include:

Using the object tagger corresponding to the target semantic relationship, it is judged whether there is an object having the target semantic relationship with the candidate subject in the representation encoding.

Multiple corresponding object taggers can be set in the system according to different target semantic relations. Each object tagger is only used to identify specific object semantic relations. The method of the embodiment of the present application does not need to learn the traditional relation classifier f(s,o)→r, but learns relation-specific object taggers f _r (s)→o, each tagger will All possible objects are identified under the condition of the subject and the subject.

Optionally, the use of the object tagger corresponding to the target semantic relationship to determine whether there is an object having the target semantic relationship with the candidate subject in the representation encoding includes:

Using a plurality of object taggers, it is judged in parallel whether there is an object having a target semantic relationship with the candidate subject in the representation encoding, and each object tagger in the plurality of object taggers corresponds to a different target semantic relationship.

Multiple object taggers are set in the system, and each object tagger only needs to identify the corresponding target semantic relationship. The system identifies objects with different semantic relationships with candidate subjects through multiple object taggers. It should be noted that the same candidate subject and the same semantic relationship may correspond to multiple objects, and the same candidate subject and different semantic relationships may also correspond to different objects.

Optionally, the use of the object tagger corresponding to the target semantic relationship to determine whether there is an object having the target semantic relationship with the candidate subject in the representation encoding includes:

Using the object tagger corresponding to the target semantic relationship, based on the encoded representation of each target word in the representation encoding and the encoded representation of the candidate subject, calculate the probability that each target word corresponds to the starting position of the object;

Using the object tagger corresponding to the target semantic relationship, based on the encoded representation of each target word in the representation encoding and the encoded representation of the candidate subject, calculate the probability that each target word corresponds to the end position of the object;

According to the probability that each target word corresponds to the starting position of the object and the probability that each target word corresponds to the ending position of the object, it is determined whether there is an object having the target semantic relationship with the candidate subject in the representation encoding.

The probability of the starting position of the object and the probability of the ending position of the object can be calculated by the following formulas:

和

分别表示输入文本中的第i个单词对应于一个客体起始和结束位置的概率，如果概率超过某一阈值，则该位置的单词将被赋予标签“1”。否则，该位置的单词将被赋予标签“0”。标签“1”代表第i个单词为一个客体起始位置或结束位置。标签“0”代表第i个单词不为一个客体起始位置或结束位置。x_i是输入句子中第i个单词的编码表示，W_(·)表示可训练权重参数，b_(·)表示偏置参数，σ表示sigmoid激活函数，

是识别出的第k个候选主体的编码表示。对于主体标注器中识别出的每一个主体，将迭代式地进行相同的解码过程。通过上述解码过程，在给定候选主体的情况下，能够同时识别出客体和对应的关系。in,

and

respectively represent the probability that the i-th word in the input text corresponds to the starting and ending positions of an object. If the probability exceeds a certain threshold, the word at this position will be assigned the label "1". Otherwise, the word at that position will be given the label "0". The label "1" indicates that the ith word is an object start position or end position. The label "0" indicates that the ith word is not an object start position or end position. x _i is the encoded representation of the ith word in the input sentence, W _{( )} represents the trainable weight parameter, b _{( )} represents the bias parameter, σ represents the sigmoid activation function,

is the encoded representation of the k-th candidate subject identified. The same decoding process is performed iteratively for each subject identified in the subject tagger. Through the above decoding process, given a candidate subject, the object and the corresponding relationship can be simultaneously identified.

In the method for generating a relation triplet provided by the embodiment of the present application, based on the candidate subject, it is judged whether there is an object that has a target semantic relationship with the candidate subject in the encoding. By adopting the solutions of the embodiments of the present application, the relation triples in the input text can be recognized comprehensively and accurately.

Referring to FIG. 3, FIG. 3 is a schematic flowchart of a method for generating a relation triplet provided by an embodiment of the present application. The method includes:

S301. Obtain a representation code corresponding to the input text.

S302. Use a subject tagger to identify multiple candidate subjects from the representation encoding.

S303. Determine whether each object having a semantic relationship with each candidate subject exists in the representation encoding.

S304. If there is, generate at least one relation triplet according to each of the candidate subjects, the semantic relation and each of the objects.

In representation encoding, a body tagger is used to identify multiple candidate bodies. For each candidate subject among the plurality of candidate subjects, each object having a semantic relationship with each candidate subject is sequentially determined in the representation encoding. With the method of this embodiment of the present application, each object that has a semantic relationship with each candidate subject can be found in sequence.

It should be noted that two processes are involved in the method of the embodiment of the present application. The first process is for multiple candidate subjects, which are identified in the representation encoding. The second process is for the judgment process in step S303. There may be multiple object taggers in the system, each object tagger corresponds to a specific semantic relationship, and different object taggers correspond to different semantic relationships. For each candidate subject, multiple object taggers are used to determine whether there is an object corresponding to the candidate subject in the representation encoding, so as to fully and accurately identify the relation triples in the input text.

In addition, in the process of using the object tagger to determine whether there is an object corresponding to the candidate subject in the representation encoding, multiple relation-specific object taggers can work simultaneously without interfering with each other, and each identify the object corresponding to the candidate subject. It is also possible to set the execution sequence of multiple object taggers in the system, and control the multiple object taggers to perform recognition in sequence according to the above execution sequence. This embodiment of the present application does not limit how to control the object tagger to perform identification.

In order to make the solutions provided by the embodiments of the present application easier to understand, a specific implementation method is provided below. Instead of learning a traditional relation classifier f(s,o)→r, the method learns relation-specific object annotators f _r (s)→o, each of which will be conditioned on a given relation and subject All possible objects are identified below. Under this framework, the relation triple extraction problem is decomposed into the following two-step cascade process: first, all possible subjects in the sentence are identified; then for each subject, a relation-specific tagger is used to simultaneously identify List all possible relationships and corresponding objects.

FIG. 4 is a schematic diagram of an overall framework structure of another method for generating relation triples provided by an embodiment of the present application. As shown in FIG. 4 , the framework of the embodiment of the present application adopts an "encoder-decoder" structure, which are respectively a BERT-based encoder (BERT Encoder) and a cascaded decoder (Cascade Decoder). Among them, the two most critical modules are the subject tagger (Subject Tagger) module and the relation-specific object tagger (Relation-Specific Object Taggers) module in the cascade decoder. It will be implemented as a cascaded binary tagger built on top of the encoder structure. The technical details are as follows:

1. BERT encoder

The encoder module aims to extract rich semantic feature information from the input sentence, and the present invention uses a pre-trained BERT model to encode the context information.

2. Cascade decoder

The cascaded decoder in the embodiment of the present application consists of two parts, which are a subject tagger module and a relation-specific object tagger module. The decoding process consists of the following cascaded two steps, first using a subject tagger to identify all possible subjects from the input sentence, and then, for each candidate subject, using a relation-specific object tagger to identify whether there is a sentence with A given candidate subject has a semantically related object. If it exists, mark all the objects, and finally generate a relationship triplet according to the corresponding subject, relationship and object. Otherwise, the candidate subject is not a real subject and cannot form a relation triple.

2.1 Principal Annotator

The subject tagger identifies all subjects by directly decoding the encoded representation produced by the BERT encoder. Specifically, the subject tagger adopts two binary classifiers with the same structure, and determines whether the current word corresponds to the starting or ending position of a subject by assigning a 0/1 binary label to each word in the sentence, thereby determining The starting and ending range of the subject. The body tagger operates on each word as follows:

和

分别表示输入句子中的第i个单词对应于一个主体起始和结束位置的概率，如果概率超过某一阈值，则该位置的单词将被赋予标签“1”。否则，该位置的单词将被赋予标签“0”。x_i是输入句子中第i个单词的编码表示，W_(·)表示可训练权重参数，b_(·)表示偏置参数，σ表示sigmoid激活函数。in,

and

respectively represent the probability that the i-th word in the input sentence corresponds to the starting and ending positions of a subject. If the probability exceeds a certain threshold, the word at this position will be assigned the label "1". Otherwise, the word at that position will be given the label "0". x _i is the encoded representation of the ith word in the input sentence, W _{( )} represents the trainable weight parameter, b _{( )} represents the bias parameter, and σ represents the sigmoid activation function.

2.2 Relation-Specific Object Annotators

Based on the entities extracted by the subject tagger, the relation-specific object tagger aims to simultaneously identify subject-related objects and their semantic relations. As shown in Figure 4, the relation-specific object tagger consists of a series of binary classifiers, each group corresponding to a specific relation, and its structure is the same as the binary classifier in the subject tagger. Unlike the subject tagger, which directly decodes the encoded representation produced by the BERT encoder, the object tagger also considers the feature v _sub of the candidate subject while decoding, and operates on each word as follows:

和

分别表示输入句子中的第i个单词对应于一个客体起始和结束位置的概率，如果概率超过某一阈值，则该位置的单词将被赋予标签“1”。否则，该位置的单词将被赋予标签“0”，

是主体标注器识别出的第k个主体的编码表示。对于主体标注器中识别出的每一个主体，本申请实施例将迭代式地进行相同的解码过程。通过上述解码过程，在给定候选主体的情况下，本申请实施例所提出的关系特定的客体标注器能够同时识别出客体和对应的关系。in,

and

respectively represent the probability that the ith word in the input sentence corresponds to the starting and ending positions of an object. If the probability exceeds a certain threshold, the word at this position will be assigned the label "1". Otherwise, the word at that position will be given the label "0",

is the encoded representation of the k-th subject identified by the subject tagger. For each subject identified in the subject tagger, the embodiment of the present application will perform the same decoding process iteratively. Through the above decoding process, given a candidate subject, the relationship-specific object tagger proposed in the embodiment of the present application can identify the object and the corresponding relationship at the same time.

In the research related to knowledge graph, the solution of the embodiment of this application proposes for the first time that the relationship is modeled as a function that maps subjects to objects, and based on this, a new type of cascaded binary annotation framework is implemented to perform relationship triples. Extract. In the relation triple extraction task, relations are modeled as functions that map subjects to objects, and a cascaded binary annotation framework is used for extraction.

By modeling the relationship as a function, the novel cascaded binary annotation framework proposed by the solution of the embodiment of the present application will not suffer from the problem of overlapping triples, and can more efficiently extract information from complex contexts compared to existing methods. Extract multiple relational triples simultaneously. Experiments show that this method greatly improves the performance of relational triplet extraction in multiple scenarios, especially on the overlapping triplet problem, which is difficult to solve by existing methods. On the NYT and WebNLG datasets The F1 value of the current best level is increased from 72.1% to 89.6%, and from 61.6% to 91.8%, which can effectively ensure the high-precision automatic construction of knowledge graphs in complex contexts.

Please refer to FIG. 5. FIG. 5 is a schematic structural diagram of an apparatus for generating a relation triple according to an embodiment of the present application. As shown in Figure 5, the generating device of the shown relation triple includes:

an encoding acquisition unit 501, used for acquiring the representation encoding corresponding to the input text;

a subject identification unit 502, configured to identify candidate subjects from the representation code;

an object judgment unit 503, configured to judge whether there is an object having a target semantic relationship with the candidate subject in the representation encoding;

The relationship generating unit 504 is configured to generate a relationship triplet according to the candidate subject, the semantic relationship and the object, if it exists.

Optionally, the code obtaining unit 501 is specifically configured to:

get input text;

The input text is encoded by the BERT encoder to generate a representation code corresponding to the input text.

Optionally, the relationship generating unit 504 is further configured to:

If not, it is determined that the candidate subject cannot form a relation triple based on the target semantic relation.

Optionally, the subject identification unit 502 is specifically configured to:

identifying a plurality of candidate subjects from the representation encoding using a subject tagger;

The object judgment unit 503 is specifically used for:

judging whether there are objects that have a semantic relationship with each candidate subject in the representation code;

The relationship generating unit 504 is specifically used for:

If there is, at least one relation triple is generated from each of the candidate subjects, the semantic relation and each of the objects.

Optionally, the object judgment unit 503 is specifically configured to:

Using the object tagger corresponding to the target semantic relationship, it is judged whether there is an object having the target semantic relationship with the candidate subject in the representation encoding.

Optionally, the object judgment unit 503 is specifically configured to:

Using a plurality of object taggers, it is determined whether there is an object having a target semantic relationship with the candidate subject in the representation encoding, and each object tagger in the plurality of object taggers corresponds to a different target semantic relationship.

Optionally, the object judgment unit 503 is specifically configured to:

Using the object tagger corresponding to the target semantic relationship, based on the encoded representation of each target word in the representation encoding and the encoded representation of the candidate subject, calculate the probability that each target word corresponds to the starting position of the object;

Using the object tagger corresponding to the target semantic relationship, based on the encoded representation of each target word in the representation encoding and the encoded representation of the candidate subject, calculate the probability that each target word corresponds to the end position of the object;

According to the probability that each target word corresponds to the starting position of the object and the probability that each target word corresponds to the ending position of the object, it is determined whether there is an object having the target semantic relationship with the candidate subject in the representation encoding.

Those skilled in the art can clearly understand that the technical solutions of the embodiments of the present application can be implemented by means of software and/or hardware. The "unit" and "module" in this specification refer to software and/or hardware that can perform a specific function independently or in cooperation with other components, wherein the hardware can be, for example, an FPGA (Field-Programmable Gate Array, Field Programmable Gate Array). , IC (Integrated Circuit, integrated circuit) and so on.

Each processing unit and/or module in the embodiments of the present application may be implemented by analog circuits that implement the functions described in the embodiments of the present application, or may be implemented by software that executes the functions described in the embodiments of the present application.

Embodiments of the present application further provide a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, implements the steps of the foregoing method for generating a relation triple. Among them, the computer-readable storage medium may include, but is not limited to, any type of disk, including floppy disks, optical disks, DVDs, CD-ROMs, micro-drives, and magneto-optical disks, ROM, RAM, EPROM, EEPROM, DRAM, VRAM, flash memory devices , magnetic or optical cards, nanosystems (including molecular memory ICs), or any type of medium or device suitable for storing instructions and/or data.

Referring to FIG. 6 , it shows a schematic structural diagram of an electronic device involved in an embodiment of the present application, and the electronic device can be used to implement the method for generating relation triples provided in the foregoing embodiments. Specifically:

The memory 1020 may be used to store software programs and modules, and the processor 1080 executes various functional applications and data processing by running the software programs and modules stored in the memory 1020 . The memory 1020 may mainly include a stored program area and a stored data area, wherein the stored program area may store an operating system, an application program required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; Data (such as audio data, phone book, etc.) created by the use of the terminal device, etc. Additionally, memory 1020 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other volatile solid state storage device. Accordingly, the memory 1020 may further include a memory controller to provide access to the memory 1020 by the processor 1080 and the input unit 1030 .

The input unit 1030 may be used to receive input numerical or character information, and generate keyboard, mouse, joystick, optical or trackball signal input related to user settings and function control. Specifically, the input unit 1030 may include a touch-sensitive surface 1031 (eg, a touch screen, a touch pad, or a touch frame). Touch-sensitive surface 1031, also known as a touch display or trackpad, collects touch operations by a user on or near it (such as a user using a finger, stylus, etc., any suitable object or accessory on or on touch-sensitive surface 1031). operation near the touch-sensitive surface 1031), and drive the corresponding connection device according to the preset program. Optionally, the touch-sensitive surface 1031 may include two parts, a touch detection device and a touch controller. Among them, the touch detection device detects the user's touch orientation, detects the signal brought by the touch operation, and transmits the signal to the touch controller; the touch controller receives the touch information from the touch detection device, converts it into contact coordinates, and then sends it to the touch controller. To the processor 1080, and can receive the command sent by the processor 1080 and execute it. In addition, the touch-sensitive surface 1031 may be implemented using resistive, capacitive, infrared, and surface acoustic wave types.

The display unit 1040 may be used to display information input by the user or information provided to the user and various graphical user interfaces of the terminal device, which may be composed of graphics, text, icons, videos and any combination thereof. The display unit 1040 may include a display panel 1041. Optionally, the display panel 1041 may be configured in the form of an LCD (Liquid Crystal Display, liquid crystal display), an OLED (Organic Light-Emitting Diode, organic light emitting diode) and the like. Further, the touch-sensitive surface 1031 can cover the display panel 1041, and when the touch-sensitive surface 1031 detects a touch operation on or near it, it transmits it to the processor 1080 to determine the type of the touch event, and then the processor 1080 determines the type of the touch event according to the touch event. Type provides corresponding visual output on display panel 1041 . Although touch-sensitive surface 1031 and display panel 1041 may be implemented as two separate components to implement input and input functions, in some embodiments, touch-sensitive surface 1031 and display panel 1041 may be integrated to implement input and output functions.

The processor 1080 is the control center of the terminal device, using various interfaces and lines to connect various parts of the entire terminal device, by running or executing the software programs and/or modules stored in the memory 1020, and calling the data stored in the memory 1020. , perform various functions of the terminal equipment and process data, so as to monitor the terminal equipment as a whole. Optionally, the processor 1080 may include one or more processing cores; wherein, the processor 1080 may integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface, and application programs, etc. The modem processor mainly handles wireless communication. It can be understood that, the above-mentioned modulation and demodulation processor may not be integrated into the processor 1080.

Specifically in this embodiment, the display unit of the terminal device is a touch screen display, the terminal device further includes a memory, and one or more programs, wherein one or more programs are stored in the memory and configured to be displayed by one or more programs. The execution of the one or more programs by the above processor includes the steps of implementing the above method for generating relation triples.

In the several embodiments provided in this application, it should be understood that the disclosed apparatus and method may be implemented in other manners. The apparatus embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components may be combined, or Can be integrated into another system, or some features can be ignored, or not implemented. In addition, the coupling, or direct coupling, or communication connection between the components shown or discussed may be through some interfaces, and the indirect coupling or communication connection of devices or units may be electrical, mechanical or other forms. of.

Each functional unit in each embodiment of the present application may be all integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units It can be implemented in the form of hardware or in the form of hardware plus software functional units.

The above descriptions are only preferred embodiments of the present application, and are not intended to limit the present application. For those skilled in the art, the present application may have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this application shall be included within the protection scope of this application.

Claims (10)

1. a generation method of relation triple, is characterized in that, described method comprises: Get the representation code corresponding to the input text; identifying candidate subjects from the representation encoding; Judging whether there is an object having a target semantic relationship with the candidate subject in the representation encoding; If there is, a relation triplet is generated according to the candidate subject, the target semantic relation and the object. 2. The method according to claim 1, wherein the obtaining the corresponding representation code of the input text comprises: get input text; The input text is encoded by the BERT encoder to generate a representation code corresponding to the input text. 3. The method according to claim 1, wherein the judging whether the representation encoding exists after the object with the target semantic relationship of the candidate subject further comprises: If not, it is determined that the candidate subject cannot form a relation triple based on the target semantic relation. 4. The method of claim 1, wherein the identifying candidate subjects from the representation encoding comprises: identifying a plurality of candidate subjects from the representation encoding using a subject tagger; The judging whether there is an object having a target semantic relationship with the candidate subject in the representation encoding includes: judging whether there are objects that have a semantic relationship with each candidate subject in the representation code; If there is, according to the candidate subject, the semantic relationship and the object, a relationship triple is generated, including: If there is, at least one relation triple is generated from each of the candidate subjects, the semantic relation and each of the objects. 5. The method according to claim 1, wherein the judging whether there is an object having a target semantic relationship with the candidate subject in the representation encoding comprises: Using the object tagger corresponding to the target semantic relationship, it is judged whether there is an object having the target semantic relationship with the candidate subject in the representation encoding. 6 . The method according to claim 5 , wherein the use of the object tagger corresponding to the target semantic relationship to determine whether there is an object having the target semantic relationship with the candidate subject in the representation encoding includes: 6 . : Using a plurality of object taggers, it is judged in parallel whether there is an object having a target semantic relationship with the candidate subject in the representation encoding, and each object tagger in the plurality of object taggers corresponds to a different target semantic relationship. 7 . The method according to claim 5 , wherein, by using an object tagger corresponding to the target semantic relationship, judging whether there is an object having the target semantic relationship with the candidate subject in the representation encoding, comprising: 8 . : Using the object tagger corresponding to the target semantic relationship, based on the encoded representation of each target word in the representation encoding and the encoded representation of the candidate subject, calculate the probability that each target word corresponds to the starting position of the object; Using the object tagger corresponding to the target semantic relationship, based on the encoded representation of each target word in the representation encoding and the encoded representation of the candidate subject, calculate the probability that each target word corresponds to the end position of the object; According to the probability that each target word corresponds to the starting position of the object and the probability that each target word corresponds to the ending position of the object, it is determined whether there is an object having the target semantic relationship with the candidate subject in the representation encoding. 8. A device for generating relation triples, wherein the device comprises: an encoding acquisition unit, used to acquire the representation encoding corresponding to the input text; a subject identification unit for identifying candidate subjects from the representation encoding; an object judgment unit, used for judging whether there is an object having a target semantic relationship with the candidate subject in the representation code; A relation generating unit, configured to generate a relation triplet according to the candidate subject, the semantic relation and the object if there is one. 9. A computer-readable storage medium on which a computer program is stored, characterized in that, when the program is executed by a processor, the steps of the method according to any one of the claims 1-7 are implemented. 10. An electronic device, comprising a memory, a processor and a computer program stored in the memory and running on the processor, characterized in that, when the processor executes the program, the implementation of the claims 1-7 The steps of any one of the methods.

Images