WO2020252919A1

WO2020252919A1 - Resume identification method and apparatus, and computer device and storage medium

Info

Publication number: WO2020252919A1
Application number: PCT/CN2019/103268
Authority: WO
Inventors: 石明川; 姚飞
Original assignee: 平安科技（深圳）有限公司
Priority date: 2019-06-20
Filing date: 2019-08-29
Publication date: 2020-12-24
Also published as: CN110442841A; CN110442841B

Abstract

A resume identification method and apparatus, and a computer device and a storage medium. The method comprises: receiving a target resume to be identified (S202); inputting said target resume into a deep neural language programming (DNLP) system, wherein the DNLP system is obtained by training using a bidirectional long-short-term memory recurrent neural network (BI-LSTM-CRF) model (S204); determining a resume template used in said target resume by using the DNLP system (S206); and extracting feature information in said target resume according to the resume template (S208). According to the method, the technical problem in the prior art of low resume identification rate is solved.

Description

Method and device for recognizing resume, computer equipment and storage medium

【Technical Field】

This application relates to the field of computers, in particular to a method and device for identifying resumes, computer equipment, and storage media.

【Background technique】

Resume recognition is a kind of semi-structured text recognition. Because it does not have the natural word order concept of traditional unstructured text, it is difficult to recognize.

The resume recognition system in the prior art is a recognition system based on keywords. For example, "person's name", "mobile phone number", "work history", etc., but if these keywords do not exist in the semi-structured text, the traditional resume recognition system cannot recognize the corresponding corpus. When performing resume recognition in the prior art, based on keyword recognition, regular expressions are usually used. The period contains various resume formats that bring identification difficulties. For example: the person name keyword is followed by the person's name in the resume, but there are also a series of problems such as the number of words, Chinese and English, and spaces in the person's name. The resume may include multiple names, multiple time periods, etc., often with work experience and project experience The problem of confusion in the recognition of the middle, because this part of the resume does not have a unified format, which leads to a very low recognition rate of the resume, and manual screening is required.

For the above-mentioned problems existing in related technologies, no effective solution has been found yet.

[Content of the invention]

In view of this, embodiments of the present application provide a method and device for identifying resumes, computer equipment, and storage media.

On the one hand, an embodiment of the present application provides a method for recognizing a resume, the method comprising: receiving a target resume to be recognized; and inputting the target resume into a deep neural language programming DNLP system, where the DNLP system is It is obtained by training with a bidirectional long and short-term memory loop neural network BI-LSTM-CRF model; using the DNLP system to determine the resume template used by the target resume; and extracting feature information in the target resume according to the resume template.

Optionally, before inputting the target resume into the deep neural linguistic programming DNLP system, the method further includes: determining a plurality of resume samples; using the plurality of resume samples to train the initial nerve of the BI-LSTM-CRF model Network to obtain the DNLP system.

Optionally, using the multiple resume samples to train the initial neural network of the BI-LSTM-CRF model includes: using a supervised classification method to segment the resume text of each resume sample to obtain multiple text blocks that can correspond to artificial labels , Wherein each text block corresponds to a category attribute in the resume; word segmentation is performed on the text block, and feature words of each text block are extracted; the BI-LSTM- is trained using the text block and the corresponding feature words The initial neural network of the CRF model.

Optionally, dividing the resume text of each resume sample by means of supervised classification includes: dividing the following resume text in each resume sample: self-introduction, education experience, work experience, learning experience, project experience; use The label information marks the resume text.

Optionally, extracting the characteristic words of each text block includes: extracting the characteristic words of each text block using the word frequency-reverse file frequency TF-IDF algorithm; where tfidf=tf*idf, and each text block takes the top n of tfidf As a characteristic word, n is a positive integer greater than 1; among them,

n _{i, j} is the number of occurrences of the current word in the text block d _j , the denominator is the sum of the number of occurrences of all words in d _j , and k is any value of i;

|D| is the total number of files in the resume sample, and |{j:t _i |∈d _j }| is the number of files containing the word t _i .

Optionally, training the initial neural network of the BI-LSTM-CRF model by using the text block and the corresponding feature words includes: in the BI layer of the BI-LSTM-CRF model, using pre-trained or randomly initialized The embedding matrix maps each word in the sentence of the text block from a one-hot vector to a low-dimensional dense word vector. Before inputting to the next layer, set the detachment to alleviate overfitting; in the BI-LSTM- In the LSTM layer of the CRF model, extract sentence features, take each feature word sequence of a sentence as the input of each time step of the bidirectional LSTM, and then combine the hidden state sequence output by the forward LSTM and the hidden state output by the reverse LSTM at each position Perform location-wise splicing to obtain a complete hidden state sequence, and output pi, where pi is the probability of belonging to the i tag; in the CRF layer of the BI-LSTM-CRF model, perform sentence-level sequence labeling to obtain linear CRF, Wherein, in the linear CRF calculation formula, the score for the tag of sentence x equal to y is:

Among them, a tag sequence y=(y1,y2,...,yn) whose length is equal to the length of the sentence; the normalized probability obtained by Softmax is:

y'is any value of all labels.

Optionally, when training the initial neural network of the BI-LSTM-CRF model, in the CRF layer of the BI-LSTM-CRF model, the following maximum log likelihood function is used to process the sample data:

^{logP (y x | x) =} score (x, y x) -log (Σ y 'exp (score (x, y'))); where, ^(x, y x) of training samples.

On the other hand, an embodiment of the present application provides a device for recognizing resumes. The device includes: a receiving module for receiving a target resume to be recognized; and an input module for inputting the target resume into a deep neural language program Learn a DNLP system, wherein the DNLP system is obtained by training with a bidirectional long and short-term memory loop neural network BI-LSTM-CRF model; a determining module is used to determine the resume template used by the target resume using the DNLP system; extract The module is used to extract feature information in the target resume according to the resume template.

Optionally, the device further includes: a determination module, configured to determine a plurality of resume samples before the input module inputs the target resume into the deep neural linguistic programming DNLP system; a training module, configured to use the Multiple resume samples train the initial neural network of the BI-LSTM-CRF model to obtain the DNLP system.

Optionally, the training module includes: a segmentation unit for segmenting the resume text of each resume sample in a supervised classification manner to obtain multiple text blocks that can correspond to manual labels, wherein each text block corresponds to a resume An extraction unit, used to segment the text block, and extract the feature words of each text block; a training unit, used to train the BI-LSTM using the text block and corresponding feature words -The initial neural network of the CRF model.

Optionally, the segmentation unit includes: a segmentation subunit for segmenting the following resume text in each resume sample: self-introduction, education experience, work experience, learning experience, project experience; labeling the information with label information Resume text.

Optionally, the extracting unit includes: an extracting subunit for extracting the characteristic words of each text block using the word frequency-inverse document frequency TF-IDF algorithm; where tfidf=tf*idf, each text block takes tfidf top n is a feature word, n is a positive integer greater than 1; among them,

Optionally, the training module includes: a first processing unit configured to use a pre-trained or randomly initialized embedding matrix to convert each sentence in the text block in the BI layer of the BI-LSTM-CRF model Words are mapped from one-hot vectors to low-dimensional dense word vectors. Before inputting the next layer, set detachment to relieve overfitting; the second processing unit is used in the LSTM layer of the BI-LSTM-CRF model In extracting sentence features, each feature word sequence of a sentence is used as the input of each time step of the bidirectional LSTM, and then the hidden state sequence output by the forward LSTM and the hidden state output by the reverse LSTM at each position are spliced by position. Obtain the complete hidden state sequence, output pi, where _p i is the probability of belonging to the i tag; the third processing unit is used to perform sentence-level sequence labeling in the CRF layer of the BI-LSTM-CRF model to obtain Linear CRF, where, in the calculation formula of the linear CRF, the score for the label of sentence x equal to y is:

y'is any value of all labels.

Optionally, the third processing unit further includes: a processing sub-unit for processing sample data using the following maximization log likelihood function: logP(y ^x |x)=score(x,y ^x )- _{log (Σ y 'exp (score} (x, y'))); where, ^(x, y x) of training samples.

According to another embodiment of the present application, there is also provided a storage medium in which a computer program is stored, wherein the computer program is configured to execute the steps in any one of the foregoing method embodiments when running.

According to another embodiment of the present application, there is also provided an electronic device, including a memory and a processor, the memory is stored with a computer program, and the processor is configured to run the computer program to execute any of the above Steps in the method embodiment.

Through this application, input the target resume into the deep neural linguistic programming DNLP system, and use the DNLP system to determine the resume template used by the target resume, and finally extract the characteristic information in the target resume according to the resume template , By first identifying the template of the resume, and then extracting characteristic information from the corresponding template, the technical problem of the low resume recognition rate in the prior art is solved, and the recognition rate of the resume is improved.

【Explanation of 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 embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative labor, other drawings can be obtained from these drawings.

FIG. 1 is a hardware structure block diagram of a mobile terminal for identifying resumes according to an embodiment of the present application;

Figure 2 is a flowchart of a method for identifying resumes according to an embodiment of the present application;

FIG. 3 is a flowchart of training a BI-LSTM-CRF model in an embodiment of the application;

Fig. 4 is a structural block diagram of a device for identifying resumes according to an embodiment of the present application.

【Detailed ways】

Hereinafter, the application will be described in detail with reference to the drawings and in conjunction with embodiments. It should be noted that the embodiments in this application and the features in the embodiments can be combined with each other if there is no conflict.

It should be noted that the terms "first" and "second" in the description and claims of the application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific sequence or sequence.

Example 1

The method embodiment provided in Embodiment 1 of the present application may be executed in a mobile terminal, a server, a computer terminal, or a similar computing device. Taking running on a computer terminal as an example, FIG. 1 is a hardware structural block diagram of a computer terminal for identifying resumes according to an embodiment of the present application. As shown in FIG. 1, the computer terminal 10 may include one or more (only one is shown in FIG. 1) processor 102 (the processor 102 may include, but is not limited to, a processing device such as a microprocessor MCU or a programmable logic device FPGA. ) And a memory 104 for storing data. Optionally, the aforementioned computer terminal may also include a transmission device 106 and an input/output device 108 for communication functions. A person of ordinary skill in the art can understand that the structure shown in FIG. 1 is only for illustration, and does not limit the structure of the foregoing computer terminal. For example, the computer terminal 10 may also include more or fewer components than those shown in FIG. 1, or have a different configuration from that shown in FIG.

The memory 104 may be used to store computer programs, for example, software programs and modules of application software, such as the computer programs corresponding to the method for identifying resumes in the embodiment of the present application. The processor 102 executes the computer programs stored in the memory 104 by running Various functional applications and data processing, namely to achieve the above methods. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely provided with respect to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the aforementioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

The transmission device 106 is used to receive or send data via a network. The above-mentioned specific examples of the network may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC for short), which can be connected to other network devices through a base station to communicate with the Internet. In an example, the transmission device 106 may be a radio frequency (Radio Frequency, referred to as RF) module, which is used to communicate with the Internet in a wireless manner.

In this embodiment, a method for identifying a resume is provided. FIG. 2 is a flowchart of the method for identifying a resume according to an embodiment of the present application. As shown in FIG. 2, the process includes the following steps:

Step S202, receiving a target resume to be identified;

Step S204, input the target resume into a deep neural language programming DNLP system, where the DNLP system is obtained by training using a bidirectional long and short-term memory cyclic neural network BI-LSTM-CRF model;

Step S206: Use the DNLP system to determine a resume template used by the target resume; the resume template includes multiple physical sections;

The resume template of this embodiment refers to the resume style or resume layout adopted by the target resume. In different resume templates, the content of the same physical section (such as work experience) is distributed in different positions of the text. The resume template of the target resume is determined by Can determine the position of each text content to be determined in the target resume;

Step S208: Extract feature information in the target resume according to the resume template.

Through the solution of this embodiment, the target resume is input into the deep neural linguistic programming DNLP system, and the DNLP system is used to determine the resume template used by the target resume, and finally the target resume is extracted according to the resume template By first identifying the template of the resume, and then extracting the characteristic information from the corresponding template, the technical problem of low resume recognition rate in the prior art is solved, and the recognition rate of resumes is improved.

In this embodiment, after extracting the feature information in the target resume according to the resume template, the feature information can be re-typeset according to the specified template set by the user, so as to facilitate centralized collection, or only the feature information that the user pays attention to ( For example, the graduate school) is extracted and bound with the resume logo or other key information, and then formatted and displayed, so as to reduce the time for users to find the key information in the complicated resume.

In this embodiment, before inputting the target resume into the deep neural linguistic programming DNLP system, the method further includes: determining a plurality of resume samples; using the plurality of resume samples to train the initial neural network of the BI-LSTM-CRF model , To obtain the DNLP system.

Fig. 3 is a flowchart of training the BI-LSTM-CRF model according to an embodiment of the present application. As shown in Fig. 3, the initial neural network for training the BI-LSTM-CRF model using the multiple resume samples includes:

S302: Use a supervised classification method to segment the resume text of each resume sample to obtain multiple text blocks that can correspond to manual labels, where each text block corresponds to a category attribute in the resume;

Specifically, dividing the resume text of each resume sample by means of supervised classification includes: dividing the following resume text (physical section) in each resume sample: self-introduction, education experience, work experience, learning experience, project Experience; use label information to mark the resume text. In a resume sample, a complete resume is composed of multiple resume texts, but for resumes with different templates, the same resume text may be distributed in different positions; this part is the process of learning each entity section of the resume;

S304: Perform word segmentation on the text block, and extract feature words of each text block; key feature words can be extracted by performing word segmentation and synonym matching on the marked text block.

Specifically, the solution for extracting the characteristic words of each text block includes: using the word frequency-inverse document frequency TF-IDF algorithm to extract the characteristic words of each text block; where tfidf=tf*idf, and each text block takes the top of tfidf n is a feature word, n is a positive integer greater than 1, preferably, n=15; where,

|D| is the total number of files in the resume sample, |{j: t _i ∈ d _j }| is the number of files containing the word t _i .

TF-IDF can filter out common words, keep important words, and extract characteristic words.

S306. Train the initial neural network of the BI-LSTM-CRF model by using the text block and the corresponding feature words.

By dividing the resume text of the sample into different entity modules (resume text), and then learn different entity modules.

In an implementation of this embodiment, the BI-LSTM-CRF model pair is trained and learned by text blocks of each category, and the recognition model of each category can be obtained including: the word-based Bi-LSTM-CRF can be used, such as B- PER, I-PER represent the first character of a person’s name, and the name is not the first character, B-SCH, I-SCH represent the first character of the school, the non-initial character of the school, etc., to train and learn the recognition model of each entity module. The neural network of the BI-LSTM-CRF model includes a three-layer logical structure. Training the initial neural network of the BI-LSTM-CRF model using the text block and the corresponding feature words includes:

In the BI layer (also called the search layer) of the BI-LSTM-CRF model, each word in the sentence of the text block is mapped to a low-dimensional by one-hot vector using a pre-trained or randomly initialized embedding matrix For dense word vectors, set detachment before entering the next layer to alleviate overfitting;

In the LSTM layer of the BI-LSTM-CRF model, extract sentence features, use each feature word sequence of a sentence as the input of each time step of the bidirectional LSTM, and then combine the hidden state sequence output by the forward LSTM and the reverse LSTM The hidden states output at each position are spliced by position to obtain a complete hidden state sequence, and output pi, where pi is the probability of belonging to the i tag;

In the CRF layer of the BI-LSTM-CRF model, the sentence-level sequence labeling is performed to obtain a linear CRF, where in the calculation formula of the linear CRF, the score for the tag of sentence x equal to y is divided into:

Among them, the sentence length tag sequence y=(y1,y2,...,yn), A is the transition matrix of the CRF layer; the normalized probability obtained by Softmax is:

y'is any value of all labels.

The softmax of this embodiment only takes partial considerations, that is, the tag of the current word is not affected by other tags.

Optionally, when training the initial neural network of the BI-LSTM-CRF model, in the CRF layer of the BI-LSTM-CRF model, the following maximum log likelihood function is used to process the sample data: logP ^{(y x | x) = score} (x, y x) -log (Σ y 'exp (score (x, y'))); where, ^(x, y x) of training samples. The scoring of the entire sequence in this embodiment is equal to the sum of the scoring of each position, and the scoring of each position is obtained from two parts, one part is determined by the pi output by the LSTM, and the other part is determined by the transition matrix A of the CRF.

Through the description of the above embodiments, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus the necessary general hardware platform, of course, it can also be implemented by hardware, but in many cases the former is Better implementation. Based on this understanding, the technical solution of this application essentially or the part that contributes to the existing technology can be embodied in the form of a software product, and the computer software product is stored in a storage medium (such as ROM/RAM, magnetic disk, The optical disc) includes several instructions to enable a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the method described in each embodiment of the present application.

Example 2

In this embodiment, a device for recognizing resumes is also provided. The device is used to implement the above-mentioned embodiments and preferred implementations. What has been explained will not be repeated. As used below, the term "module" can implement a combination of software and/or hardware with predetermined functions. Although the devices described in the following embodiments are preferably implemented by software, hardware or a combination of software and hardware is also possible and conceived.

Fig. 4 is a structural block diagram of a device for identifying resumes according to an embodiment of the present application. As shown in Fig. 4, the device includes:

The receiving module 40 is used to receive the target resume to be identified;

The input module 42 is configured to input the target resume into a deep neural language programming DNLP system, where the DNLP system is obtained by training using a bidirectional long and short-term memory cyclic neural network BI-LSTM-CRF model;

The determining module 44 is configured to use the DNLP system to determine the resume template used by the target resume;

The extraction module 46 is configured to extract feature information in the target resume according to the resume template.

Optionally, the training module includes: a first processing unit configured to use a pre-trained or randomly initialized embedding matrix to convert each sentence in the text block in the BI layer of the BI-LSTM-CRF model Words are mapped from one-hot vectors to low-dimensional dense word vectors. Before inputting the next layer, set detachment to relieve overfitting; the second processing unit is used in the LSTM layer of the BI-LSTM-CRF model In extracting sentence features, each feature word sequence of a sentence is used as the input of each time step of the bidirectional LSTM, and then the hidden state sequence output by the forward LSTM and the hidden state output by the reverse LSTM at each position are spliced by position. Obtain the complete hidden state sequence, output pi, where pi is the probability of belonging to the i tag; the third processing unit is used to perform sentence-level sequence labeling in the CRF layer of the BI-LSTM-CRF model to obtain linear CRF, wherein, in the calculation formula of the linear CRF, the score for the label of sentence x equal to y is:

y'is any value of all labels.

It should be noted that each of the above modules can be implemented by software or hardware. For the latter, it can be implemented in the following manner, but not limited to this: the above modules are all located in the same processor; or, the above modules are combined in any combination The forms are located in different processors.

Example 3

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

The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the objectives of the solutions of the embodiments.

In addition, the functional units in each embodiment of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware, or may be implemented in the form of hardware plus software functional units.

The above-mentioned integrated unit implemented in the form of a software functional unit may be stored in a computer readable storage medium. The above-mentioned software functional unit is stored in a storage medium and includes several instructions to make a computer device (which may be a personal computer, a server, or a network device, etc.) or a processor (Processor) execute the method described in each embodiment of the present application Part of the steps. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), magnetic disk or optical disk and other media that can store program code .

The embodiment of the present application also provides a storage medium in which a computer program is stored, wherein the computer program is set to execute the steps in any one of the foregoing method embodiments when running.

Optionally, in this embodiment, the foregoing storage medium may be configured to store a computer program for executing the following steps:

S1, receive the target resume to be identified;

S2. Input the target resume into a deep neural language programming DNLP system, where the DNLP system is obtained by training using a bidirectional long and short-term memory cyclic neural network BI-LSTM-CRF model;

S3, using the DNLP system to determine a resume template used by the target resume;

S4: Extract feature information in the target resume according to the resume template.

Optionally, in this embodiment, the foregoing storage medium may include, but is not limited to: U disk, Read-Only Memory (Read-Only Memory, ROM for short), Random Access Memory (Random Access Memory, RAM for short), Various media that can store computer programs, such as mobile hard disks, magnetic disks, or optical disks.

The embodiment of the present application also provides an electronic device, including a memory and a processor, the memory is stored with a computer program, and the processor is configured to run the computer program to execute the steps in any of the foregoing method embodiments.

Optionally, the aforementioned electronic device may further include a transmission device and an input-output device, wherein the transmission device is connected to the aforementioned processor, and the input-output device is connected to the aforementioned processor.

Optionally, in this embodiment, the foregoing processor may be configured to execute the following steps through a computer program:

S1, receive the target resume to be identified;

The above are only the preferred embodiments of this application and are not intended to limit this application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this application shall be included in this application Within the scope of protection.

Claims

A method for identifying resumes, the method comprising:

Receive the target resume to be identified;

Input the target resume into a deep neural language programming DNLP system, where the DNLP system is obtained by training using a bidirectional long and short-term memory loop neural network BI-LSTM-CRF model;

Use the DNLP system to determine the resume template used by the target resume;

The feature information in the target resume is extracted according to the resume template.
The method according to claim 1, before inputting the target resume into a deep neural language programming DNLP system, the method further comprises:

Identify multiple resume samples;

The initial neural network of the BI-LSTM-CRF model is trained using the multiple resume samples to obtain the DNLP system.
The method according to claim 2, wherein training the initial neural network of the BI-LSTM-CRF model using the plurality of resume samples comprises:

Using a supervised classification method to segment the resume text of each resume sample to obtain multiple text blocks that can correspond to manual labels, where each text block corresponds to a category attribute in the resume;

Perform word segmentation on the text block, and extract the characteristic words of each text block;

The initial neural network of the BI-LSTM-CRF model is trained using the text block and the corresponding feature words.
The method according to claim 3, adopting a supervised classification method to segment the resume text of each resume sample includes:

Divide the following resume texts in each resume sample: self-introduction, education experience, work experience, learning experience, project experience;

Use label information to mark the resume text.
According to the method of claim 3, extracting the characteristic words of each text block comprises:

Use word frequency-reverse document frequency TF-IDF algorithm to extract the characteristic words of each text block;

Among them, tfidf=tf*idf, each text block takes top n of tfidf as the characteristic word, and n is a positive integer greater than 1;

among them,
n i, j is the number of occurrences of the current word in the text block d j , the denominator is the sum of the number of occurrences of all words in d j , and k is any value of i;

|D| is the total number of files in the resume sample, |{j: t i ∈ d j }| is the number of files containing the word t i .
According to the method of claim 3, training the initial neural network of the BI-LSTM-CRF model by using the text block and the corresponding feature words comprises:

In the BI layer of the BI-LSTM-CRF model, each word in the sentence of the text block is mapped from a one-hot vector to a low-dimensional dense word vector using a pre-trained or randomly initialized embedding matrix. Before entering the next layer, set detachment to relieve overfitting;

In the LSTM layer of the BI-LSTM-CRF model, extract sentence features, use each feature word sequence of a sentence as the input of each time step of the bidirectional LSTM, and then combine the hidden state sequence output by the forward LSTM and the reverse LSTM The hidden states output at each position are spliced by position to obtain a complete hidden state sequence, and output pi, where pi is the probability of belonging to the i tag;

In the CRF layer of the BI-LSTM-CRF model, the sentence-level sequence labeling is performed to obtain a linear CRF, where in the calculation formula of the linear CRF, the score for the tag of sentence x equal to y is divided into:

Among them, the sentence length tag sequence y=(y1,y2,...,yn), A is the transition matrix of the CRF layer;

The normalized probability obtained by Softmax is:

y'is any value of all labels.
According to the method of claim 6, when training the initial neural network of the BI-LSTM-CRF model, in the CRF layer of the BI-LSTM-CRF model, the following maximization log likelihood function is used for the sample Data processing:

logP (y x | x) = score (x, y x) -log (Σ y 'exp (score (x, y')));

Among them, (x, y x ) are training samples.
A device for identifying resumes, the device comprising:

The receiving module is used to receive the target resume to be identified;

The input module is used to input the target resume into a deep neural language programming DNLP system, where the DNLP system is obtained by training using a bidirectional long and short-term memory cyclic neural network BI-LSTM-CRF model;

The determining module is configured to use the DNLP system to determine the resume template used by the target resume;

The extraction module is used to extract feature information in the target resume according to the resume template.
The device according to claim 8, further comprising:

The determining module is used to determine a plurality of resume samples before the input module inputs the target resume into the deep neural linguistic programming DNLP system; the training module is used to train BI-LSTM-CRF using the multiple resume samples The initial neural network of the model is used to obtain the DNLP system.
The device according to claim 9, wherein the training module comprises:

The segmentation unit is used to segment the resume text of each resume sample in a supervised classification method to obtain multiple text blocks that can correspond to manual tags, where each text block corresponds to a category attribute in the resume; the extraction unit uses To segment the text block and extract the feature words of each text block; the training unit is used to train the initial neural network of the BI-LSTM-CRF model by using the text block and the corresponding feature words.
The device according to claim 10, the dividing unit comprises:

The segmentation subunit is used to segment the following resume text in each resume sample: self-introduction, education experience, work experience, learning experience, and project experience; label the resume text with label information.
The device according to claim 10, the extracting unit comprises:

The extraction subunit is used to extract the feature words of each text block by using the word frequency-inverse document frequency TF-IDF algorithm; where tfidf=tf*idf, each text block takes the top n of tfidf as the feature word, and n is greater than 1. Positive integer; where,
n i, j is the number of occurrences of the current word in the text block d j , the denominator is the sum of the number of occurrences of all words in d j , and k is any value of i;
|D| is the total number of files in the resume sample, |{j: t i ∈ d j }| is the number of files containing the word t i .
The device according to claim 10, the training module comprises:

The first processing unit is configured to use a pre-trained or randomly initialized embedding matrix in the BI layer of the BI-LSTM-CRF model to map each word in the sentence of the text block by one-hot vector to low The word vector with dense dimensions is set to escape before entering the next layer to alleviate overfitting; the second processing unit is used to extract sentence features in the LSTM layer of the BI-LSTM-CRF model, and combine the Each feature word sequence is used as the input of each time step of the bidirectional LSTM, and then the hidden state sequence output by the forward LSTM and the hidden state output by the reverse LSTM at each position are spliced by position to obtain a complete hidden state sequence, and output pi, Among them, pi is the probability of belonging to the i tag; the third processing unit is used to perform sentence-level sequence labeling in the CRF layer of the BI-LSTM-CRF model to obtain a linear CRF, wherein the calculation of the linear CRF In the formula, the label of sentence x is equal to y score:
Among them, a tag sequence y=(y1,y2,...,yn) whose length is equal to the length of the sentence; the normalized probability obtained by Softmax is:
y'is any value of all labels.
The apparatus according to claim 13, wherein the third processing unit further comprises:

Processing sub-unit, for maximizing the following log likelihood function of sample data processing: logP (y x | x) = score (x, y x) -log (Σ y 'exp (score (x, y'))); where (x, y x ) are training samples.
A computer device includes a memory and a processor, the memory stores a computer program, and the steps of a method for identifying a resume when the processor executes the computer program include:

Receive the target resume to be identified;

Input the target resume into a deep neural language programming DNLP system, where the DNLP system is obtained by training using a bidirectional long and short-term memory loop neural network BI-LSTM-CRF model;

Use the DNLP system to determine the resume template used by the target resume;

The feature information in the target resume is extracted according to the resume template.
The computer device according to claim 15, before inputting the target resume into a deep neural language programming DNLP system, the method further comprises:

Identify multiple resume samples;

The initial neural network of the BI-LSTM-CRF model is trained using the multiple resume samples to obtain the DNLP system.
The computer device according to claim 15, wherein using the plurality of resume samples to train the initial neural network of the BI-LSTM-CRF model comprises:

Using a supervised classification method to segment the resume text of each resume sample to obtain multiple text blocks that can correspond to manual labels, where each text block corresponds to a category attribute in the resume;

Perform word segmentation on the text block, and extract the characteristic words of each text block;

The initial neural network of the BI-LSTM-CRF model is trained using the text block and the corresponding feature words.
A computer storage medium having a computer program stored thereon, and the steps of implementing a method for identifying a resume when the computer program is executed by a processor include:

Receive the target resume to be identified;

Input the target resume into a deep neural language programming DNLP system, where the DNLP system is obtained by training using a bidirectional long and short-term memory loop neural network BI-LSTM-CRF model;

Use the DNLP system to determine the resume template used by the target resume;

The feature information in the target resume is extracted according to the resume template.
The computer device according to claim 18, before inputting the target resume into the deep neural language programming DNLP system, the method further comprises:

Identify multiple resume samples;

Use the multiple resume samples to train the initial neural network of the BI-LSTM-CRF model to obtain the DNLP system.
The computer device according to claim 18, using the plurality of resume samples to train the initial neural network of the BI-LSTM-CRF model comprises:

Using a supervised classification method to segment the resume text of each resume sample to obtain multiple text blocks that can correspond to manual labels, where each text block corresponds to a category attribute in the resume;

Perform word segmentation on the text block, and extract the characteristic words of each text block;

The initial neural network of the BI-LSTM-CRF model is trained using the text block and the corresponding feature words.