WO2020232882A1

WO2020232882A1 - Named entity recognition method and apparatus, device, and computer readable storage medium

Info

Publication number: WO2020232882A1
Application number: PCT/CN2019/103141
Authority: WO
Inventors: 邓悦; 金戈; 徐亮
Original assignee: 平安科技（深圳）有限公司
Priority date: 2019-05-20
Filing date: 2019-08-28
Publication date: 2020-11-26
Also published as: CN110298019A; CN110298019B

Abstract

Provided in the present application are a named entity recognition method and apparatus, a device, and a computer readable storage medium, the method comprising: by means of a word encoding layer, obtaining a first word vector respectively corresponding to each word; by means of a character encoding layer and a bidirectional long short-term memory network layer, obtaining a second word vector respectively corresponding to each word; and inputting the first word vector and the second word vector respectively corresponding to each word into a named entity recognition layer to obtain a named entity. The present application can increase the precision of named entity recognition.

Description

Named entity recognition method, device, equipment and computer readable storage medium

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office, the application number is 201910420794.X, and the invention title is "Named Entity Recognition Method, Apparatus, Equipment, and Computer-readable Storage Medium" on May 20, 2019. The entire content is incorporated into this application by reference.

Technical field

This application relates to the technical field of semantic parsing, and in particular to a named entity recognition method, device, equipment, and computer-readable storage medium.

Background technique

Named Entity Recognition (NER) is a basic task in natural language processing, which refers to identifying named referents from text, paving the way for tasks such as relation extraction. In a narrow sense, proper nouns such as names of persons, places and organizations are recognized. In the smart interview scenario, it is necessary to analyze the interviewer’s answer text, and identify named entities from the answer text, such as person’s name, place name, and organization’s name, so that the interviewer’s information can be structured automatically, such as the interview The name of the person, the school of graduation, and the location of the school of graduation are identified from the answer text and stored in the database.

For Chinese named entity recognition, it currently includes word-based named entity recognition and word-based named entity recognition. However, word-based named entity recognition and word-based named entity recognition both have semantic information missing, and semantic information is missing. This will lead to the problem that the recognition accuracy of the named entity is not high. Therefore, how to improve the recognition accuracy of the named entity is a problem to be solved urgently.

Summary of the invention

The main purpose of this application is to provide a named entity recognition method, device, equipment and computer-readable storage medium, aiming to improve the accuracy of named entity recognition.

In the first aspect, this application provides a named entity recognition method, which includes the following steps:

When a named entity recognition request is monitored, the target sentence to be recognized is determined according to the named entity recognition request, and a named entity recognition model is obtained, wherein the named entity recognition model includes at least a word encoding layer, a word encoding layer, and a bidirectional Long and short-term memory network layer and named entity recognition layer;

Inputting the target sentence into the word coding layer to obtain a first word vector corresponding to each word in the target sentence;

Input the target sentence into the word encoding layer to obtain a target word vector corresponding to each word in the target sentence;

Using word as a unit, input the target word vector of each word in each word into the bidirectional long-term short-term memory network layer in turn to obtain a second word vector corresponding to each word;

The first word vector and the second word vector corresponding to each word are input into the named entity recognition layer to obtain the named entity in the target sentence.

In the second aspect, the present application also provides a named entity recognition device, the named entity recognition device includes:

The determining module is used to determine the target sentence to be recognized according to the named entity recognition request when the named entity recognition request is monitored;

An acquisition module for acquiring a named entity recognition model, where the named entity recognition model includes at least a word encoding layer, a word encoding layer, a two-way long and short-term memory network layer, and a named entity recognition layer;

The first word vector determining module is configured to input the target sentence into the word encoding layer to obtain a first word vector corresponding to each word in the target sentence;

A word vector determining module, configured to input the target sentence into the word encoding layer to obtain a target word vector corresponding to each word in the target sentence;

The second word vector determining module is used to input the target word vector of each word in each word into the two-way long-term short-term memory network layer in turn, to obtain a second word vector corresponding to each word respectively;

The named entity recognition module is used to input the first word vector and the second word vector corresponding to each word into the named entity recognition layer to obtain the named entity in the target sentence.

In a third aspect, the present application also provides a computer device that includes a processor, a memory, and a computer program that is stored on the memory and can be executed by the processor, wherein the computer program is When the processor is executed, the steps of the above-mentioned named entity recognition method are realized.

In a fourth aspect, the present application also provides a computer-readable storage medium having a computer program stored on the computer-readable storage medium, and when the computer program is executed by a processor, the steps of the aforementioned named entity identification method are implemented .

This application provides a named entity recognition method, device, equipment, and computer-readable storage medium. This application uses the word encoding layer of the named entity recognition model to obtain the vector representation of each word in the target sentence at the word granularity, and Through the word encoding layer of the named entity recognition model and the two-way long and short-term memory network layer, the vector representation of each word in the target sentence at the word granularity can be obtained, which can reduce the loss of information at the word granularity, and then combine each word in the word The vector representation and the named entity recognition model under the granularity and word granularity can accurately identify the named entity in the sentence and effectively improve the accuracy of the recognition of the named entity.

Description of the drawings

In order to explain the technical solutions of the embodiments of the present application more clearly, the following will briefly introduce the drawings needed in the description of the embodiments. Obviously, the drawings in the following description are some embodiments of the present application. Ordinary technicians can obtain other drawings based on these drawings without creative work.

FIG. 1 is a schematic flowchart of a named entity identification method provided by an embodiment of the application;

Figure 2 is a hierarchical schematic diagram of a named entity recognition model provided by an embodiment of the application;

FIG. 3 is a schematic flowchart of sub-steps of the named entity recognition method in FIG. 1;

4 is a schematic flowchart of another named entity identification method provided by an embodiment of the application;

FIG. 5 is a schematic block diagram of a named entity recognition device provided by an embodiment of this application;

6 is a schematic block diagram of sub-modules of the named entity recognition device in FIG. 5;

FIG. 7 is a schematic block diagram of another named entity recognition device provided by an embodiment of this application;

FIG. 8 is a schematic block diagram of the structure of a computer device related to an embodiment of the application.

The realization, functional characteristics, and advantages of the purpose of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings.

Detailed ways

The technical solutions in the embodiments of the present application will be described clearly and completely in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of them. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of this application.

The flowchart shown in the drawings is merely an illustration, and does not necessarily include all contents and operations/steps, nor does it have to be executed in the described order. For example, some operations/steps can also be decomposed, combined or partially combined, so the actual execution order may be changed according to actual conditions.

The embodiments of the present application provide a named entity recognition method, device, computer equipment, and computer-readable storage medium. Wherein, the named entity identification method can be applied to a server, and the server can be a single server or a server cluster.

Hereinafter, some embodiments of the present application will be described in detail with reference to the accompanying drawings. In the case of no conflict, the following embodiments and features in the embodiments can be combined with each other.

Please refer to FIG. 1. FIG. 1 is a schematic flowchart of a named entity recognition method according to an embodiment of the application.

As shown in FIG. 1, the named entity identification method is used to accurately recommend wealth management products to users, wherein the named entity identification method includes steps S101 to S105.

Step S101: When a named entity recognition request is monitored, a target sentence to be recognized is determined according to the named entity recognition request, and a named entity recognition model is obtained.

Among them, the named entity recognition model is obtained through training. Specifically, because the named entity recognition is a supervised problem, the sample data set is labeled for this reason, and the labeled sample data set is [X, Y], then The input is X=x_1,x_2,x_3,……,x_n, the output is Y=y_1,y_2,y_3,……,y_n, x_1 represents the first word in the sentence sequence, X represents a sentence composed of words, y_1 Represents the annotation corresponding to x_1, Y represents the sequence of annotations. After the labeled sample data set [X, Y] is obtained, the named entity recognition model to be trained is trained based on the labeled sample data set until the name to be trained The entity recognition model converges to obtain a named entity recognition model.

Among them, the named entity recognition model includes at least a word encoding layer, a word encoding layer, a two-way long short-term memory network layer, and a named entity recognition layer, and the named entity recognition layer includes a one-way long short-term memory network layer and a conditional random field (Conditional Random Field). algorithm, CRF) layer. It should be noted that the loss function of the named entity recognition model to be trained can be selected as:

And

Among them, e ^s(X,y) is the sequence score of sentence X,

Is the sum of the sequence scores of all sentences, Z _i,x is the hidden layer output of the i-th word in sentence X in the hidden layer of the LSTM model layer, y _i is the label corresponding to the i-th word in sentence X, and y _i-1 is The label corresponding to the i-1th word in sentence X, n is the number of words in sentence X, matrices W and b represent the transition probability between entity tags, the elements in W are vectors, and the elements in b are Numerical value. In the loss function, the relationship between the hidden layer output of the one-way long and short-term memory network layer and the probability transition matrix of the CRF layer is multiplication, which can increase the hypothesis space of the model and further improve the recognition of the named entity recognition model Accuracy.

Fig. 2 is a hierarchical schematic diagram of a named entity recognition model provided by an embodiment of the application. As shown in Fig. 2, the named entity recognition model includes a word encoding layer, a word encoding layer, a two-way long short-term memory network layer, and a named entity recognition layer. And the target sentence is input into the word coding layer and the word coding layer respectively.

The interviewer’s voice data is collected through the terminal device, and a named entity recognition request carrying the voice data is generated, and the named entity recognition request is sent to the server. When the server detects the named entity recognition request, it will be recognized according to the named entity Request, determine the target sentence to be recognized, and obtain the named entity recognition model. The method for determining the target sentence to be recognized is specifically: acquiring voice data from the named entity request, and performing voice recognition on the voice data to convert the voice data into a text sentence, and then determining the text sentence as a pending sentence. The identified target sentence.

Step S102: Input the target sentence into the word encoding layer to obtain a first word vector corresponding to each word in the target sentence.

After determining the target sentence, the server inputs the target sentence to the word encoding layer of the named entity recognition model, and obtains the first word vector corresponding to each word in the target sentence, specifically: a word vector stored in the word encoding layer Matrix, when the target sentence is input to the word coding layer, the target sentence is split into several words, and according to the word vector matrix, each word in the several words is represented as a corresponding The first word vector is to obtain a word from several words in turn, and record it as the target word, then obtain the word vector corresponding to the target word from the word vector matrix, and determine the word vector as the first target word The word vector is obtained once for each word in several words, so that the first word vector corresponding to each word in the target sentence can be obtained.

It should be noted that one row of the word vector matrix represents the word vector of a word, and the word vector matrix can be set based on actual conditions, which is not specifically limited in this solution.

For example, the word vector matrix is:

Among them, the word corresponding to the first row of the word vector matrix is "apple", the word corresponding to the second row is "phone", and the word corresponding to the last row is "model". Therefore, the first word vector of the word "apple" is [0.1,0.34,......,0.89], the first word vector of the word "mobile phone" is [0.98,0.3,......,0.76], and the second word vector of the word "model" is [0.77,0.3,......,0.22].

Step S103: Input the target sentence into the word encoding layer to obtain a target word vector corresponding to each word in the target sentence.

While inputting the target sentence into the word encoding layer, the target sentence is input into the word encoding layer of the named entity recognition model to obtain the target word vector corresponding to each word in the target sentence, which is specifically pre-stored in the word encoding layer There is a word vector matrix. After the target sentence is input into the word coding layer, the target sentence is split into several single words, and the target word vector corresponding to each word in the target sentence is determined according to the word vector matrix, that is, from Obtain a word from several words one by one and record it as the target word, then obtain the word vector corresponding to the target word from the word vector matrix, and determine the word vector as the target word vector of the target word until the word vector in the word vector matrix Each single word is obtained once, so that the target word vector corresponding to each word in the target sentence can be obtained. It should be noted that a row of the word vector matrix represents a word vector of a single word, and the word vector matrix can be set based on actual conditions, and this solution does not specifically limit this.

Step S104: Using word as a unit, input the target word vector of each word in each word to the bidirectional long-term short-term memory network layer in turn to obtain a second word vector corresponding to each word.

After obtaining the target word vector of each word in the target sentence, the server sequentially inputs the target word vector of each word in each word into the bidirectional long-term and short-term memory network layer of the named entity recognition model in units of words. Get the second word vector corresponding to each word.

In one embodiment, in order to avoid the loss of semantic information at word granularity, it is necessary to characterize the semantic information at word granularity through a bidirectional long-term short-term memory network layer. Specifically, referring to FIG. 3, step S104 includes: sub-step S1041 to sub-step S1042 .

Sub-step S1041, in units of words, input the target word vector of each word in each word to the bidirectional long-term short-term memory network layer in turn to obtain the forward hidden layer output and reverse hidden layer of each word in each word Containing layer output.

After obtaining the target word vector of each word in the target sentence, the server uses the word as a unit to input the target word vector of each word in each word into the bidirectional long-term short-term memory network layer in turn to obtain each word in each word. The forward hidden layer output and the reverse hidden layer output of the word. Among them, the bidirectional long and short-term memory network layer is composed of a forward recurrent neural network (Recurrent Neural Network, RNN) and a reverse RNN. The bidirectional long and short-term memory network is an extension of the traditional long and short-term memory network, which can improve the model of sequence classification problems. performance.

Sub-step S1042, according to the forward hidden layer output and the reverse hidden layer output of each word in each word, determine the second word vector corresponding to each word.

After obtaining the forward hidden layer output and the reverse hidden layer output of each word in each word, each word is determined according to the forward hidden layer output and the reverse hidden layer output of each word in each word The corresponding second word vector is specifically: get the reverse hidden layer output corresponding to the initial word of each word and the forward hidden layer output corresponding to the ending word, and the reverse direction corresponding to the initial word of each word The output of the hidden layer is spliced with the output of the forward hidden layer corresponding to the ending word to obtain the second word vector corresponding to each word.

It should be noted that the splicing method of the reverse hidden layer output corresponding to the initial word and the forward hidden layer output corresponding to the ending word of each word is sequential splicing, for example, the reverse hidden layer corresponding to the initial word of a word The output of the containing layer is [0.2,0.3,……,0.9], and the output of the positive hidden layer corresponding to the ending word is [0.8,0.7,……,0.4], then the second word vector obtained by splicing is [0.2,0.3 ,……,0.9,0.8,0.7,……,0.4].

Step S105: Input the first word vector and the second word vector corresponding to each word to the named entity recognition layer to obtain the named entity in the target sentence.

After obtaining the first word vector and the second word vector corresponding to each word, the server inputs the first word vector and the second word vector corresponding to each word into the named entity recognition layer of the named entity recognition model, Obtain the named entity in the target sentence, that is, take the word as a unit, input the first word vector and the second word vector corresponding to each word into the unidirectional long-term short-term network in the named entity recognition layer, and get each word corresponding The hidden layer output of each word is input to the CRF network in the named entity recognition layer to obtain the entity tag of each word, thereby completing the recognition of the named entity in the sentence to be recognized.

In an embodiment, the first word vector is a representation of semantic information at word granularity, and the second word vector is a representation of semantic information at word granularity. In order to improve the accuracy of named entities, it is necessary to merge the semantic information at word granularity and word granularity. Information representation, specifically, in units of words, the first word vector and second word vector corresponding to each word are sequentially input to the vector splicing sublayer in the named entity recognition layer to obtain the spliced word vector corresponding to each word , And then input the spliced word vector corresponding to each word to the named entity recognition sub-layer in the named entity recognition layer to obtain the named entity in the target sentence, that is, take the word as a unit, the spliced word vector corresponding to each word Input to the unidirectional long- and short-term network in the named entity recognition sublayer to obtain the hidden layer output corresponding to each word, and input the hidden layer output corresponding to each word to the CRF network in the named entity recognition sublayer, Obtain the entity label of each word, thereby completing the recognition of the named entity in the sentence to be recognized. Among them, the named entity recognition layer includes a vector splicing sublayer and a named entity recognition sublayer, and the named entity recognition sublayer is composed of a unidirectional long-term short-term network and a CRF network.

In the named entity recognition method provided by the foregoing embodiments, the word encoding layer of the named entity recognition model can obtain the vector representation of each word in the target sentence at the word granularity, and the word encoding layer and bidirectional length of the named entity recognition model The short-term memory network layer can obtain the vector representation of each word in the target sentence at the word granularity, which can reduce the loss of information at the word granularity, and then combine the vector representation of each word at the word granularity and the word granularity and named entity recognition The model can accurately identify the named entity in the sentence, which effectively improves the accuracy of the named entity recognition.

Please refer to FIG. 4, which is a schematic flowchart of another named entity recognition method provided by an embodiment of the application.

As shown in FIG. 4, the named entity recognition method includes steps S201 to 205.

Step S201: When a named entity recognition request is monitored, determine the target sentence to be recognized according to the named entity recognition request, and obtain a named entity recognition model.

Wherein, the named entity recognition model includes at least a word encoding layer, a word encoding layer, a two-way long short-term memory network layer, and a named entity recognition layer, and the named entity recognition layer includes a one-way long short-term memory network layer and a CRF layer.

Step S202: Input the target sentence into the word encoding layer to obtain a first word vector corresponding to each word in the target sentence.

Step S203: Input the target sentence into the word encoding layer to obtain a target word vector corresponding to each word in the target sentence.

When the target sentence is input into the word encoding layer, the target sentence is input into the word encoding layer of the named entity recognition model, and the target word vector corresponding to each word in the target sentence is obtained. In order to more accurately represent the semantic information under the word granularity, the target word vector corresponding to each word can be obtained by fusing the word vector and the pinyin vector. Specifically, referring to FIG. 4, step S203 includes sub-steps S2031 to S2033.

In sub-step S2031, the target sentence is input to the word vector coding sublayer in the word coding layer to obtain a word vector corresponding to each word in the target sentence.

Among them, the word coding layer includes a word vector coding sublayer, a pinyin vector coding sublayer and a vector stitching sublayer. Input the target sentence to the word vector coding sublayer in the word coding layer to obtain the word vector corresponding to each word in the target sentence. Specifically, a word vector matrix is preset in the word vector coding sublayer, and After the target sentence is input to the word vector encoding sublayer in the word encoding layer, the word vector matrix in the word vector encoding sublayer is used to obtain the word vector corresponding to each word in the target sentence, which is to split the target sentence into several One word, and obtain a word from several words one by one, record it as the target word, and then obtain the word vector corresponding to the target word from the word vector matrix, until each word in the several words is obtained once, so that the The word vector corresponding to each word in the target sentence. It should be noted that the aforementioned word vector matrix can be set based on actual conditions, which is not specifically limited in this embodiment.

In sub-step S2032, the target sentence is input to the pinyin vector coding sublayer in the word coding layer to obtain a pinyin vector corresponding to each word in the target sentence.

While inputting the target sentence into the word vector coding sublayer, input the target sentence into the pinyin vector coding sublayer in the word coding layer to obtain the pinyin vector corresponding to each word in the target sentence, specifically, The target sentence is input to the pinyin vector coding sublayer in the word coding layer, and the character vector matrix corresponding to each pinyin character contained in each word in the target sentence is obtained through the character vector matrix in the pinyin vector coding sublayer, and then According to the natural sequence of the pinyin characters, the character vectors corresponding to the pinyin characters contained in each word are spliced to obtain the pinyin vectors corresponding to each word in the target sentence.

For example, the pinyin characters contained in the word "lang" are "l", "a", "n" and "g", and the natural order is lang. Let the pinyin characters "l", "a", "n" and " The character vectors of "g" are [0.1,0.36,......,0.89], [0.9,0.3,......,0.76], [0.88,0.4,......,0.46 ] And [0.6,0.3,......,0.36], then after the character vectors of the pinyin characters "l", "a", "n" and "g" are spliced, the resulting pinyin vector is [0.1, 0.36,......,0.89,0.9,0.3,......,0.76,0.88,0.4,......,0.46,0.6,0.3,......,0.36] .

Sub-step S2033, in units of words, input the word vector and the pinyin vector corresponding to each word in the target sentence into the vector splicing sub-layer of the word encoding layer in turn to obtain each word in the target sentence The corresponding target word vector respectively.

After determining the word vector and pinyin vector of each word, take the word as a unit, input the word vector and pinyin vector corresponding to each word in the target sentence into the vector splicing sub-layer in the word encoding layer in turn to obtain the target sentence Each word in the corresponding target word vector. Among them, the splicing method includes the word vector splicing before the pinyin vector and the word vector splicing after the pinyin vector. For example, the character vector and pinyin vector of "国" are [0.2,0.36,......,0.86] and [0.3,0.56,......,0.89], then the target character of "国" The vector can be [0.2,0.36,...,0.86,0.3,0.56,...,0.89], and the target word vector for "country" can also be [0.3,0.56,... ..,0.89,0.2,0.36,......,0.86].

Step S204: Using word as a unit, input the target word vector of each word in each word into the two-way long and short-term memory network layer in turn to obtain a second word vector corresponding to each word.

Step S205: Input the first word vector and the second word vector corresponding to each word into the named entity recognition layer to obtain the named entity in the target sentence.

In the named entity recognition method provided by the foregoing embodiment, through the word encoding layer of the named entity recognition model, the vector representation of each word in the target sentence at the word granularity can be obtained, through the word vector, pinyin vector and two-way long and short-term memory The network layer can further accurately characterize the semantic information under the word granularity and reduce the loss of information under the word granularity. Then combined with the vector representation of each word in the word granularity and the word granularity and the named entity recognition model, it can accurately identify The named entity in the sentence effectively improves the recognition accuracy of the named entity.

Please refer to FIG. 5, which is a schematic block diagram of a named entity recognition apparatus provided by an embodiment of the application.

As shown in FIG. 5, the named entity recognition device 300 includes: a determination module 301, an acquisition module 302, a first word vector determination module 303, a word vector determination module 304, a second word vector determination module 305, and a named entity recognition module 306 .

The determining module 301 is configured to determine the target sentence to be recognized according to the named entity recognition request when a named entity recognition request is monitored.

The obtaining module 302 is configured to obtain a named entity recognition model, where the named entity recognition model includes at least a word encoding layer, a word encoding layer, a two-way long and short-term memory network layer, and a named entity recognition layer;

The first word vector determining module 303 is configured to input the target sentence into the word encoding layer to obtain a first word vector corresponding to each word in the target sentence.

The word vector determining module 304 is configured to input the target sentence into the word encoding layer to obtain a target word vector corresponding to each word in the target sentence.

The second word vector determining module 305 is configured to input the target word vector of each word in each word into the two-way long-term short-term memory network layer in turn to obtain the second word vector corresponding to each word in the unit of word .

In an embodiment, as shown in FIG. 6, the second word vector determining module 305 includes:

The hidden layer output determination sub-module 3051 is used to input the target word vector of each word in each word to the bidirectional long-term short-term memory network layer in turn in the unit of word, to obtain the forward direction of each word in each word Hidden layer output and reverse hidden layer output;

The word vector determination sub-module 3052 is used to determine the second word vector corresponding to each word according to the forward hidden layer output and the reverse hidden layer output of each word in each word.

In one embodiment, the word vector determining submodule 3052 is also used to obtain the reverse hidden layer output corresponding to the initial word of each word and the forward hidden layer output corresponding to the ending word; The output of the reverse hidden layer corresponding to the initial word is spliced with the output of the forward hidden layer corresponding to the ending word to obtain the second word vector corresponding to each word.

The named entity recognition module 306 is configured to input the first word vector and the second word vector corresponding to each word into the named entity recognition layer to obtain the named entity in the target sentence.

In one embodiment, the named entity recognition module 306 is further configured to input the first word vector and the second word vector corresponding to each word into the vector splicing in the named entity recognition layer in order in units of words. The sublayer obtains the spliced word vector corresponding to each word; inputs the spliced word vector corresponding to each word to the named entity recognition sublayer in the named entity recognition layer to obtain the named entity in the target sentence.

Please refer to FIG. 7. FIG. 7 is a schematic block diagram of another named entity recognition apparatus provided by an embodiment of the application.

As shown in FIG. 7, the named entity recognition device 400 includes: a determination module 401, an acquisition module 402, a first word vector determination module 403, a word vector determination module 404, a second word vector determination module 405, and a named entity recognition module 406 .

The determining module 401 is configured to determine the target sentence to be recognized according to the named entity recognition request when a named entity recognition request is monitored.

The acquiring module 402 is configured to acquire a named entity recognition model, where the named entity recognition model includes at least a word encoding layer, a word encoding layer, a two-way long and short-term memory network layer, and a named entity recognition layer;

The first word vector determining module 403 is configured to input the target sentence into the word encoding layer to obtain a first word vector corresponding to each word in the target sentence.

The word vector determining module 404 is configured to input the target sentence into the word encoding layer to obtain a target word vector corresponding to each word in the target sentence.

In an embodiment, as shown in FIG. 7, the word vector determining module 404 includes:

The word vector determining submodule 4041 is configured to input the target sentence into the word vector encoding sublayer in the word encoding layer to obtain a word vector corresponding to each word in the target sentence;

The pinyin vector determining submodule 4042 is used to input the target sentence into the pinyin vector coding sublayer in the word coding layer to obtain the pinyin vector corresponding to each word in the target sentence;

The target word vector determination sub-module 4043 is used to input the word vector and the pinyin vector corresponding to each word in the target sentence into the vector splicing sub-layer of the word encoding layer in order in units of words to obtain the Each word in the target sentence corresponds to the target word vector.

In one embodiment, the word vector determining sub-module 4041 is further configured to input the target sentence into the word vector coding sublayer in the word coding layer; the word vector in the word vector coding sublayer Matrix to obtain the word vector corresponding to each word in the target sentence.

In one embodiment, the pinyin vector determining sub-module 4042 is further configured to input the target sentence into the pinyin vector coding sublayer in the character coding layer; the character vector in the pinyin vector coding sublayer is coded by the pinyin vector Matrix to obtain the character vector corresponding to each pinyin character contained in each word in the target sentence; splicing the character vector corresponding to each pinyin character contained in each word to obtain the corresponding character of each word in the target sentence Pinyin vector.

The second word vector determining module 405 is configured to input the target word vector of each word in each word into the two-way long-term short-term memory network layer in turn to obtain a second word vector corresponding to each word in word units .

The named entity recognition module 406 is configured to input the first word vector and the second word vector corresponding to each word into the named entity recognition layer to obtain the named entity in the target sentence.

In one embodiment, the named entity recognition module 406 is further configured to input the first word vector and the second word vector corresponding to each word into the vector splicing in the named entity recognition layer in order in units of words. The sublayer obtains the spliced word vector corresponding to each word; inputs the spliced word vector corresponding to each word to the named entity recognition sublayer in the named entity recognition layer to obtain the named entity in the target sentence.

It should be noted that those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working process of the above described device and each module and unit can refer to the corresponding process in the aforementioned named entity recognition method embodiment , I won’t repeat it here.

The apparatus provided in the foregoing embodiment may be implemented in the form of a computer program, and the computer program may run on the computer device as shown in FIG. 8.

Please refer to FIG. 8. FIG. 8 is a schematic block diagram of a structure of a computer device according to an embodiment of the application. The computer device may be a server.

As shown in FIG. 8, the computer device includes a processor, a memory, and a network interface connected through a system bus, where the memory may include a non-volatile storage medium and an internal memory.

The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions. When the program instructions are executed, the processor can execute any named entity recognition method.

The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

The internal memory provides an environment for the running of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute any named entity identification method.

The network interface is used for network communication, such as sending assigned tasks. Those skilled in the art can understand that the structure shown in FIG. 8 is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may Including more or fewer parts than shown in the figure, or combining some parts, or having a different arrangement of parts.

It should be understood that the processor may be a central processing unit (Central Processing Unit, CPU), the processor may also be other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), and application specific integrated circuits (Application Specific Integrated Circuits). Circuit, ASIC), Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor.

Wherein, in an embodiment, the processor is used to run a computer program stored in a memory to implement the following steps:

In one embodiment, the processor realizes that the target word vector of each word in each word is sequentially input to the bidirectional long-term and short-term memory network layer in the unit of words to obtain the second corresponding to each word. When using word vectors, it is used to achieve:

Using word as a unit, input the target word vector of each word in each word to the bidirectional long-short-term memory network layer in turn to obtain the forward hidden layer output and the reverse hidden layer output of each word in each word;

According to the forward hidden layer output and the reverse hidden layer output of each word in each word, the second word vector corresponding to each word is determined.

In one embodiment, when the processor determines the second word vector corresponding to each word according to the forward hidden layer output and the reverse hidden layer output of each word in each word, it is used to achieve:

Obtain the reverse hidden layer output corresponding to the initial word of each word and the forward hidden layer output corresponding to the ending word;

The reverse hidden layer output corresponding to the initial word of each word is spliced with the forward hidden layer output corresponding to the ending word to obtain the second word vector corresponding to each word.

In one embodiment, when the processor realizes that the first word vector and the second word vector corresponding to each word are input into the named entity recognition layer to obtain the named entity in the target sentence, it is used for achieve:

In units of words, input the first word vector and the second word vector corresponding to each word to the vector splicing sub-layer in the named entity recognition layer in order to obtain the spliced word vector corresponding to each word;

The spliced word vector corresponding to each word is input to the named entity recognition sub-layer in the named entity recognition layer to obtain the named entity in the target sentence.

Wherein, in another embodiment, the processor is configured to run a computer program stored in a memory, so as to input the target sentence into the word encoding layer to obtain a corresponding corresponding to each word in the target sentence. The steps of the target word vector include:

Input the target sentence to the word vector coding sublayer in the word coding layer to obtain a word vector corresponding to each word in the target sentence;

Inputting the target sentence into the pinyin vector coding sublayer in the word coding layer to obtain a pinyin vector corresponding to each word in the target sentence;

In units of words, the word vector and pinyin vector corresponding to each word in the target sentence are sequentially input into the vector splicing sublayer in the word encoding layer to obtain the target corresponding to each word in the target sentence. Word vector.

In one embodiment, the processor is configured to input the target sentence into the word vector encoding sublayer in the word encoding layer to obtain the word vector corresponding to each word in the target sentence. achieve:

Inputting the target sentence into the word vector coding sublayer in the word coding layer;

Through the word vector matrix in the word vector encoding sublayer, the word vector corresponding to each word in the target sentence is obtained.

In an embodiment, the processor is configured to input the target sentence to the pinyin vector coding sublayer in the word coding layer to obtain the pinyin vector corresponding to each word in the target sentence achieve:

Inputting the target sentence into the pinyin vector coding sublayer in the word coding layer;

Obtaining a character vector corresponding to each pinyin character contained in each word in the target sentence through the character vector matrix in the pinyin vector encoding sublayer;

The character vectors corresponding to the pinyin characters contained in each word are spliced to obtain the pinyin vectors corresponding to each word in the target sentence.

The embodiments of the present application also provide a computer-readable storage medium, the computer-readable storage medium stores a computer program, the computer program includes program instructions, and the method implemented when the program instructions are executed can refer to this Various embodiments of the named entity recognition method are applied.

The computer-readable storage medium may be the internal storage unit of the computer device described in the foregoing embodiment, such as the hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (SMC), or a secure digital (Secure Digital, SD) equipped on the computer device. ) Card, Flash Card, etc.

It should be noted that in this article, the terms "include", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements not only includes those elements, It also includes other elements that are not explicitly listed, or elements inherent to the process, method, article, or system. Without more restrictions, the element defined by the sentence "including a..." does not exclude the existence of other identical elements in the process, method, article or system that includes the element.

The serial numbers of the foregoing embodiments of the present application are only for description, and do not represent the advantages and disadvantages of the embodiments. The above are only specific implementations of this application, but the protection scope of this application is not limited to this. Anyone familiar with the technical field can easily think of various equivalents within the technical scope disclosed in this application. Modifications or replacements, these modifications or replacements shall be covered within the protection scope of this application. Therefore, the protection scope of this application shall be subject to the protection scope of the claims.

Claims

A named entity recognition method, including:

When a named entity recognition request is monitored, determine the target sentence to be recognized according to the named entity recognition request;

Obtain a named entity recognition model, where the named entity recognition model includes at least a word encoding layer, a word encoding layer, a bidirectional long and short-term memory network layer, and a named entity recognition layer, and the bidirectional long and short-term memory network layer includes a forward loop neural network And reverse loop neural network;

Inputting the target sentence into the word coding layer to obtain a first word vector corresponding to each word in the target sentence;

Input the target sentence into the word encoding layer to obtain a target word vector corresponding to each word in the target sentence;

In units of words, the target word vectors of each word in each word are sequentially input to the bidirectional long- and short-term memory network layer to obtain the forward hidden layer output and the reverse hidden layer output of each word in each word, And concatenate the output of the reverse hidden layer corresponding to the initial word of each word and the output of the forward hidden layer corresponding to the ending word to obtain a second word vector corresponding to each word;

The first word vector and the second word vector corresponding to each word are input into the named entity recognition layer to obtain the named entity in the target sentence.
The method for recognizing a named entity according to claim 1, wherein the first word vector and the second word vector corresponding to each word are input into the named entity recognition layer to obtain the named entity in the target sentence The steps include:

In units of words, input the first word vector and the second word vector corresponding to each word to the vector splicing sub-layer in the named entity recognition layer in order to obtain the spliced word vector corresponding to each word;

The spliced word vector corresponding to each word is input to the named entity recognition sub-layer in the named entity recognition layer to obtain the named entity in the target sentence.
5. The named entity recognition method according to claim 1, wherein the step of inputting the target sentence into the word encoding layer to obtain the target word vector corresponding to each word in the target sentence comprises:

Input the target sentence to the word vector coding sublayer in the word coding layer to obtain a word vector corresponding to each word in the target sentence;

Inputting the target sentence into the pinyin vector coding sublayer in the word coding layer to obtain a pinyin vector corresponding to each word in the target sentence;

In units of words, the word vector and pinyin vector corresponding to each word in the target sentence are sequentially input into the vector splicing sublayer in the word encoding layer to obtain the target corresponding to each word in the target sentence. Word vector.
The named entity recognition method according to claim 3, wherein the input of the target sentence into the word vector encoding sublayer in the word encoding layer obtains the word vector corresponding to each word in the target sentence The steps include:

Inputting the target sentence into the word vector coding sublayer in the word coding layer;

Through the word vector matrix in the word vector encoding sublayer, the word vector corresponding to each word in the target sentence is obtained.
The named entity recognition method according to claim 3, wherein said inputting said target sentence into a pinyin vector coding sublayer in said character coding layer to obtain a pinyin vector corresponding to each word in said target sentence The steps include:

Inputting the target sentence into the pinyin vector coding sublayer in the word coding layer;

Obtaining a character vector corresponding to each pinyin character contained in each word in the target sentence through the character vector matrix in the pinyin vector encoding sublayer;

The character vectors corresponding to the pinyin characters contained in each word are spliced to obtain the pinyin vectors corresponding to each word in the target sentence.
A named entity recognition device, wherein the named entity recognition device includes:

The determining module is used to determine the target sentence to be recognized according to the named entity recognition request when the named entity recognition request is monitored;

An acquisition module for acquiring a named entity recognition model, where the named entity recognition model includes at least a word encoding layer, a word encoding layer, a two-way long and short-term memory network layer, and a named entity recognition layer. The two-way long and short-term memory network layer includes Forward loop neural network and reverse loop neural network;

The first word vector determining module is configured to input the target sentence into the word encoding layer to obtain a first word vector corresponding to each word in the target sentence;

A word vector determining module, configured to input the target sentence into the word encoding layer to obtain a target word vector corresponding to each word in the target sentence;

The second word vector determination module is used to input the target word vector of each word in each word into the two-way long and short-term memory network layer in turn to obtain the positive implied meaning of each word in each word. Layer output and reverse hidden layer output, and the reverse hidden layer output corresponding to the initial word of each word and the forward hidden layer output corresponding to the ending word are spliced to obtain the corresponding corresponding to each word Second word vector

The named entity recognition module is used to input the first word vector and the second word vector corresponding to each word into the named entity recognition layer to obtain the named entity in the target sentence.
The named entity recognition device according to claim 6, wherein the named entity recognition module is further used for:

In units of words, input the first word vector and the second word vector corresponding to each word to the vector splicing sub-layer in the named entity recognition layer in order to obtain the spliced word vector corresponding to each word;

The spliced word vector corresponding to each word is input to the named entity recognition sub-layer in the named entity recognition layer to obtain the named entity in the target sentence.
7. The named entity recognition device of claim 6, wherein the word vector determining module comprises:

A word vector determining sub-module for inputting the target sentence into the word vector coding sublayer in the word coding layer to obtain a word vector corresponding to each word in the target sentence;

The pinyin vector determining submodule is used to input the target sentence into the pinyin vector coding sublayer in the word coding layer to obtain the pinyin vector corresponding to each word in the target sentence;

The target word vector determination sub-module is used to input the word vector and the pinyin vector corresponding to each word in the target sentence into the vector splicing sublayer in the word encoding layer in order in units of words to obtain the target Each word in the sentence corresponds to the target word vector.
8. The named entity recognition device of claim 8, wherein the word vector determining sub-module is further configured to:

Inputting the target sentence into the word vector coding sublayer in the word coding layer;

Through the word vector matrix in the word vector encoding sublayer, the word vector corresponding to each word in the target sentence is obtained.
8. The named entity recognition device according to claim 8, wherein the pinyin vector determining sub-module is further used for:

Inputting the target sentence into the pinyin vector coding sublayer in the word coding layer;

Obtaining a character vector corresponding to each pinyin character contained in each word in the target sentence through the character vector matrix in the pinyin vector encoding sublayer;

The character vectors corresponding to the pinyin characters contained in each word are spliced to obtain the pinyin vectors corresponding to each word in the target sentence.
A computer device, wherein the computer device includes a processor, a memory, and a computer program stored on the memory and executable by the processor, and when the computer program is executed by the processor, The following steps:

When a named entity recognition request is monitored, determine the target sentence to be recognized according to the named entity recognition request;

Obtain a named entity recognition model, where the named entity recognition model includes at least a word encoding layer, a word encoding layer, a bidirectional long and short-term memory network layer, and a named entity recognition layer, and the bidirectional long and short-term memory network layer includes a forward loop neural network And reverse loop neural network;

Inputting the target sentence into the word coding layer to obtain a first word vector corresponding to each word in the target sentence;

Input the target sentence into the word encoding layer to obtain a target word vector corresponding to each word in the target sentence;

In units of words, the target word vectors of each word in each word are sequentially input to the bidirectional long- and short-term memory network layer to obtain the forward hidden layer output and the reverse hidden layer output of each word in each word, And concatenate the output of the reverse hidden layer corresponding to the initial word of each word and the output of the forward hidden layer corresponding to the ending word to obtain a second word vector corresponding to each word;

The first word vector and the second word vector corresponding to each word are input into the named entity recognition layer to obtain the named entity in the target sentence.
The computer device according to claim 11, wherein the processor realizes that the first word vector and the second word vector corresponding to each word are input into the named entity recognition layer to obtain the target sentence When naming an entity, it is used to achieve:

In units of words, input the first word vector and the second word vector corresponding to each word to the vector splicing sub-layer in the named entity recognition layer in order to obtain the spliced word vector corresponding to each word;

The spliced word vector corresponding to each word is input to the named entity recognition sub-layer in the named entity recognition layer to obtain the named entity in the target sentence.
The computer device according to claim 11, wherein the processor is configured to input the target sentence into the word encoding layer to obtain the target word vector corresponding to each word in the target sentence. achieve:

Input the target sentence to the word vector coding sublayer in the word coding layer to obtain a word vector corresponding to each word in the target sentence;

Inputting the target sentence into the pinyin vector coding sublayer in the word coding layer to obtain a pinyin vector corresponding to each word in the target sentence;

In units of words, the word vector and pinyin vector corresponding to each word in the target sentence are sequentially input into the vector splicing sublayer in the word encoding layer to obtain the target corresponding to each word in the target sentence. Word vector.
The computer device according to claim 13, wherein the processor implements the input of the target sentence into the word vector coding sublayer in the word coding layer to obtain the corresponding word in the target sentence. When a word vector is used, it is used to achieve:

Inputting the target sentence into the word vector coding sublayer in the word coding layer;

Through the word vector matrix in the word vector encoding sublayer, the word vector corresponding to each word in the target sentence is obtained.
The computer device according to claim 13, wherein the processor implements the input of the target sentence into the pinyin vector coding sublayer in the word coding layer to obtain the corresponding corresponding to each word in the target sentence. When pinyin vectors are used to achieve:

Inputting the target sentence into the pinyin vector coding sublayer in the word coding layer;

Obtaining a character vector corresponding to each pinyin character contained in each word in the target sentence through the character vector matrix in the pinyin vector encoding sublayer;

The character vectors corresponding to the pinyin characters contained in each word are spliced to obtain the pinyin vectors corresponding to each word in the target sentence.
A computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the following steps are implemented:

When a named entity recognition request is monitored, determine the target sentence to be recognized according to the named entity recognition request;

Obtain a named entity recognition model, where the named entity recognition model includes at least a word encoding layer, a word encoding layer, a bidirectional long and short-term memory network layer, and a named entity recognition layer, and the bidirectional long and short-term memory network layer includes a forward loop neural network And reverse loop neural network;

Inputting the target sentence into the word coding layer to obtain a first word vector corresponding to each word in the target sentence;

Input the target sentence into the word encoding layer to obtain a target word vector corresponding to each word in the target sentence;

In units of words, the target word vectors of each word in each word are sequentially input to the bidirectional long- and short-term memory network layer to obtain the forward hidden layer output and the reverse hidden layer output of each word in each word, And concatenate the output of the reverse hidden layer corresponding to the initial word of each word and the output of the forward hidden layer corresponding to the ending word to obtain a second word vector corresponding to each word;

The first word vector and the second word vector corresponding to each word are input into the named entity recognition layer to obtain the named entity in the target sentence.
The computer-readable storage medium according to claim 16, wherein the processor realizes that the first word vector and the second word vector corresponding to each word are input into the named entity recognition layer to obtain the target The named entity in the statement is used to achieve:

In units of words, input the first word vector and the second word vector corresponding to each word to the vector splicing sub-layer in the named entity recognition layer in order to obtain the spliced word vector corresponding to each word;

The spliced word vector corresponding to each word is input to the named entity recognition sub-layer in the named entity recognition layer to obtain the named entity in the target sentence.
The computer-readable storage medium of claim 16, wherein the processor implements inputting the target sentence into the word encoding layer to obtain a target word vector corresponding to each word in the target sentence To achieve:

Input the target sentence to the word vector coding sublayer in the word coding layer to obtain a word vector corresponding to each word in the target sentence;

Inputting the target sentence into the pinyin vector coding sublayer in the word coding layer to obtain a pinyin vector corresponding to each word in the target sentence;

In units of words, the word vector and pinyin vector corresponding to each word in the target sentence are sequentially input into the vector splicing sublayer in the word encoding layer to obtain the target corresponding to each word in the target sentence. Word vector.
The computer-readable storage medium according to claim 18, wherein the processor implements the input of the target sentence into the word vector coding sublayer in the word coding layer to obtain each word in the target sentence When corresponding to the word vector, it is used to realize:

Inputting the target sentence into the word vector coding sublayer in the word coding layer;

Through the word vector matrix in the word vector encoding sublayer, the word vector corresponding to each word in the target sentence is obtained.
The computer-readable storage medium according to claim 18, wherein the processor realizes that the target sentence is input to the pinyin vector coding sublayer in the word coding layer to obtain each word in the target sentence When the corresponding pinyin vectors are respectively used to achieve:

Inputting the target sentence into the pinyin vector coding sublayer in the word coding layer;

Obtaining a character vector corresponding to each pinyin character contained in each word in the target sentence through the character vector matrix in the pinyin vector encoding sublayer;

The character vectors corresponding to the pinyin characters contained in each word are spliced to obtain the pinyin vectors corresponding to each word in the target sentence.