WO2022142108A1 - Method and apparatus for training interview entity recognition model, and method and apparatus for extracting interview information entity - Google Patents

Abstract

A method and apparatus for training an interview entity recognition model, and a method and apparatus for extracting an interview information entity. The method for training an interview entity recognition model comprises: performing standard label prediction on first interview sample data by means of a direct prediction module in a preset recognition model, so as to obtain a standard label distribution and an interview coding vector; performing auxiliary label prediction on the first interview sample data by means of each auxiliary prediction module and according to the coding vector, so as to obtain an auxiliary label distribution output by each auxiliary prediction module; determining a total loss value of the preset recognition model according to each auxiliary label distribution and the standard label distribution; and when the total loss value does not meet a preset convergence condition, updating and iterating a first initial parameter of the preset recognition model until the total loss value meets the preset convergence condition, and recording the preset recognition model after convergence is realized as an interview entity recognition model. By means of the method, the model training efficiency and the model recognition accuracy are improved.

Description

Interview entity recognition model training, interview information entity extraction method and device

This application claims the priority of the Chinese patent application filed on December 30, 2020 with the application number 202011620124.1 and the invention titled "Interview Entity Recognition Model Training, Interview Information Entity Extraction Method and Device", the entire contents of which are approved by Reference is incorporated in this application.

technical field

The present application relates to the technical field of prediction models, and in particular, to a method, apparatus, device and medium for training an interview entity recognition model and extracting an interview information entity.

Background technique

Named entity recognition is essentially a sequence labeling problem, that is, by inputting a sentence, outputting the entity corresponding to each word in the sentence, that is, identifying entities with specific meanings in the document, such as person names, place names, school names and proper nouns. For example, for the self-introduction of the interviewer in the intelligent recruitment process, it may be necessary to extract the company name and the name of the school, so as to facilitate the subsequent extraction and use of the interviewer's information.

The inventor realized that for the problem of named entity recognition, if named entity recognition is performed through supervised learning, the demand for the amount of data is very large, and through unsupervised learning algorithms, such as using pretrained language models, named entity recognition, This limits the accuracy of the model.

Application content

The embodiments of the present application provide an interview entity recognition model training and interview information entity extraction method, device, equipment and medium, so as to solve the problem that supervised learning requires a large amount of data and unsupervised learning limits the accuracy of the model.

An interview entity recognition model training method, comprising:

obtaining a preset interview sample data set; the preset interview sample data set includes at least one first interview sample data without an interview label;

Inputting the first interview sample data into a preset recognition model including the first initial parameter, and performing standard label prediction on the first interview sample data through the direct prediction module in the preset recognition model to obtain a standard label distribution and an interview coding vector corresponding to the first interview sample data;

Through each auxiliary prediction module in the preset recognition model, carry out auxiliary label prediction on the first interview sample data according to the interview coding vector, and obtain the auxiliary label distribution output by each of the auxiliary prediction modules;

Determine the total loss value of the preset recognition model according to each of the auxiliary label distribution and the standard label distribution;

When the total loss value does not reach the preset convergence condition, update and iterate the first initial parameter of the preset recognition model, until the total loss value reaches the preset convergence condition, all the The aforementioned preset recognition model is recorded as the interview entity recognition model.

A method for extracting interview information entities, comprising:

Obtain the interview information of the target interviewee; the interview information includes at least one interview sentence; one of the interview sentences includes a plurality of interview information words;

The interview sentence is input into the interview entity recognition model, and the interview information words in the interview sentence are extracted and recognized to obtain entity recognition results corresponding to the interview information words; the interview entity recognition model is Obtained according to the above interview entity recognition model training method;

Insert the entity recognition result into a preset interview information storage template according to preset matching rules.

An interview entity recognition model training device, comprising:

a sample data acquisition module, configured to acquire a preset interview sample data set; the preset interview sample data set includes at least one first interview sample data without an interview label;

The standard label prediction module is used to input the first interview sample data into a preset recognition model including the first initial parameter, and perform the first interview sample data through the direct prediction module in the preset recognition model. Standard label prediction, to obtain standard label distribution and interview coding vector corresponding to the first interview sample data;

The auxiliary label prediction module is used to perform auxiliary label prediction on the first interview sample data according to the interview coding vector through each auxiliary prediction module in the preset recognition model, and obtain the output of each auxiliary prediction module. Auxiliary label distribution;

a total loss value determination module, configured to determine the total loss value of the preset recognition model according to each of the auxiliary label distributions and the standard label distribution;

A model training module, configured to update and iterate the first initial parameter of the preset recognition model when the total loss value does not reach the preset convergence condition, until the total loss value reaches the preset convergence condition , and record the preset recognition model after convergence as an interview entity recognition model.

A device for extracting interview information entities, comprising:

an interview information obtaining module, used for obtaining interview information of a target interviewee; the interview information includes at least one interview sentence; one interview sentence includes a plurality of interview information words;

An entity extraction and recognition module is used to input the interview sentence into the interview entity recognition model, extract and recognize the interview information words in the interview sentence, and obtain entity recognition results corresponding to each of the interview information words; The interview entity recognition model is obtained according to the above-mentioned interview entity recognition model training method;

An information storage module, configured to insert the entity recognition result into a preset interview information storage template according to preset matching rules.

A computer device, comprising a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor, the processor implementing the following steps when executing the computer-readable instructions:

obtaining a preset interview sample data set; the preset interview sample data set includes at least one first interview sample data without an interview label;

Inputting the first interview sample data into a preset recognition model including the first initial parameter, and performing standard label prediction on the first interview sample data through the direct prediction module in the preset recognition model to obtain a standard label distribution and an interview coding vector corresponding to the first interview sample data;

Through each auxiliary prediction module in the preset recognition model, carry out auxiliary label prediction on the first interview sample data according to the interview coding vector, and obtain the auxiliary label distribution output by each of the auxiliary prediction modules;

Determine the total loss value of the preset recognition model according to the distribution of each auxiliary label and the standard label distribution; when the total loss value does not reach the preset convergence condition, update the first step of iterating the preset recognition model. An initial parameter, until the total loss value reaches the preset convergence condition, the preset recognition model after convergence is recorded as the interview entity recognition model.

A computer device, comprising a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor, the processor implementing the following steps when executing the computer-readable instructions:

Obtain the interview information of the target interviewee; the interview information includes at least one interview sentence; one of the interview sentences includes a plurality of interview information words;

The interview sentence is input into the interview entity recognition model, and the interview information words in the interview sentence are extracted and recognized to obtain entity recognition results corresponding to the interview information words; the interview entity recognition model is Obtained according to the above interview entity recognition model training method;

Insert the entity recognition result into a preset interview information storage template according to preset matching rules.

One or more readable storage media storing computer-readable instructions that, when executed by one or more processors, cause the one or more processors to perform the following steps:

obtaining a preset interview sample data set; the preset interview sample data set includes at least one first interview sample data without an interview label;

Inputting the first interview sample data into a preset recognition model including the first initial parameter, and performing standard label prediction on the first interview sample data through the direct prediction module in the preset recognition model to obtain a standard label distribution and an interview coding vector corresponding to the first interview sample data;

Through each auxiliary prediction module in the preset recognition model, carry out auxiliary label prediction on the first interview sample data according to the interview coding vector, and obtain the auxiliary label distribution output by each of the auxiliary prediction modules;

Determine the total loss value of the preset recognition model according to each of the auxiliary label distribution and the standard label distribution;

When the total loss value does not reach the preset convergence condition, update and iterate the first initial parameter of the preset recognition model, until the total loss value reaches the preset convergence condition, all the The aforementioned preset recognition model is recorded as the interview entity recognition model.

One or more readable storage media storing computer-readable instructions that, when executed by one or more processors, cause the one or more processors to perform the following steps:

Obtain the interview information of the target interviewee; the interview information includes at least one interview sentence; one of the interview sentences includes a plurality of interview information words;

The interview sentence is input into the interview entity recognition model, and the interview information words in the interview sentence are extracted and recognized to obtain entity recognition results corresponding to the interview information words; the interview entity recognition model is Obtained according to the above interview entity recognition model training method;

Insert the entity recognition result into a preset interview information storage template according to preset matching rules.

The above interview entity recognition model training, interview information entity extraction method, device, equipment and medium, the interview entity recognition model training method obtains a preset interview sample data set; the preset interview sample data set includes at least one that does not have an interview label. The first interview sample data; the first interview sample data is input into a preset recognition model containing the first initial parameter, and the first interview sample data is analyzed by the direct prediction module in the preset recognition model. Standard label prediction, obtain standard label distribution and interview coding vector corresponding to the first interview sample data; through each auxiliary prediction module in the preset recognition model, according to the interview coding vector Perform auxiliary label prediction on the data to obtain auxiliary label distributions output by each of the auxiliary prediction modules; determine the total loss value of the preset recognition model according to each of the auxiliary label distributions and the standard label distribution; When the value does not reach the preset convergence condition, update and iterate the first initial parameter of the preset recognition model, until the total loss value reaches the preset convergence condition, the preset recognition model after convergence will be updated. Recorded as Interview Entity Recognition Model.

In this application, by combining the direct prediction module of supervised learning and the auxiliary prediction module of unsupervised learning, on the basis of the direct prediction module, the auxiliary prediction module can give the model more different data features (such as each auxiliary prediction module). The auxiliary label distribution output by the module) improves the training efficiency of the interview entity recognition model and improves the accuracy of the trained interview entity recognition model.

The details of one or more embodiments of the application are set forth in the accompanying drawings and the description below, and other features and advantages of the application will become apparent from the description, drawings, and claims.

Description of drawings

In order to illustrate the technical solutions of the embodiments of the present application more clearly, the following briefly introduces the drawings that are used in the description of the embodiments of the present application. Obviously, the drawings in the following description are only some embodiments of the present application. , for those of ordinary skill in the art, other drawings can also be obtained from these drawings without creative labor.

1 is a schematic diagram of an application environment of an interview entity recognition model training method and an interview information entity extraction method in an embodiment of the present application;

Fig. 2 is a flowchart of an interview entity recognition model training method in an embodiment of the present application;

Fig. 3 is a flowchart of step S30 in the training method of interview entity recognition model in an embodiment of the present application;

Fig. 4 is another flowchart of the training method of the interview entity recognition model in an embodiment of the present application;

5 is a flowchart of a method for extracting interview information entities in an embodiment of the present application;

6 is a schematic block diagram of an interview entity recognition model training device in an embodiment of the present application;

7 is a schematic block diagram of an auxiliary label prediction module in an interview entity recognition model training device in an embodiment of the present application;

8 is a schematic block diagram of an apparatus for extracting interview information entities in an embodiment of the present application;

FIG. 9 is a schematic diagram of a computer device in an embodiment of the present application.

Detailed ways

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

The interview entity recognition model training method provided by the embodiment of the present application can be applied in the application environment shown in FIG. 1 . Specifically, the interview entity recognition model training method is applied in an interview entity recognition model training system. The interview entity recognition model training system includes a client and a server as shown in FIG. 1 , and the client and the server communicate through a network for Solve the problem that supervised learning requires a large amount of data and unsupervised learning limits the accuracy of the model. Among them, the client, also known as the client, refers to the program corresponding to the server and providing local services for the client. Clients can be installed on, but not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented as an independent server or a server cluster composed of multiple servers.

In one embodiment, as shown in FIG. 2 , a method for training an interview entity recognition model is provided, and the method is applied to the server in FIG. 1 as an example for description, including the following steps:

S10: Obtain a preset interview sample data set; the preset interview sample data set includes at least one first interview sample data without an interview label;

Understandably, the first interview sample data is data that does not have pre-marked interview labels; generally, a large amount of manually labeled data is required for model training and learning in supervised learning, but the demand for manually labeled data is large. , the manual labeling method wastes time and cannot output huge labeled data. Therefore, one of the problems to be solved in this application is how to train and learn the model more accurately and quickly in the absence of labeled data. Further, the first interview sample data can be selected according to different scenarios. Exemplarily, the first interview sample data can be the interviewee's self-introduction, or the interviewer's resume; In the editing scenario, the first interview sample data can be replaced with sentences in the movie script.

S20: Input the first interview sample data into a preset recognition model including the first initial parameter, and perform standard label prediction on the first interview sample data through the direct prediction module in the preset recognition model, to obtain Standard label distribution and interview coding vector corresponding to the first interview sample data;

Understandably, in this application, the preset recognition model is a semi-supervised learning model formed by combining supervised learning and unsupervised learning; the direct prediction module is obtained by training a small amount of data with interview labels, that is, The direct prediction module is a module that has been trained. When the standard label prediction is performed on the first interview sample data that does not have an interview label through the direct prediction module, it is not necessary to train an additional prediction module, which improves the efficiency of model training.

Further, after inputting the first interview sample data into the preset recognition model including the first initial parameter, the first interview sample data is used as the input of the direct prediction module, which includes a bidirectional recurrent neural network encoder. , the bidirectional cyclic neural network encoder is used for vector encoding the first interview sample data, and then obtains the interview encoding vector corresponding to the first interview sample data; the direct prediction module also includes an annotation classification trained with interview annotation labels. Then, after vector encoding the first interview sample data through the bidirectional cyclic neural network encoder to obtain the interview encoding vector, the labeling classifier is used to perform direct label prediction on the interview encoding vector to obtain the corresponding first interview sample data. Standard label distribution.

S30: Through each auxiliary prediction module in the preset recognition model, perform auxiliary label prediction on the first interview sample data according to the interview coding vector, and obtain the auxiliary label distribution output by each of the auxiliary prediction modules;

Understandably, the auxiliary prediction module refers to a module that performs entity prediction on a certain word according to different word combinations. The auxiliary prediction module is used to combine with the direct prediction module to form a semi-supervised mode. It should be noted that, in order to extract as much representation data of each word in the first interview sample data as possible, the features of the first interview sample data extracted by each auxiliary prediction module are all are different, that is, the basis for each auxiliary prediction module to discriminate the entities of the words in the first interview sample data is different, thereby improving the accuracy of the model's entity recognition; exemplarily, such as predicting words through the target Entity prediction is performed on the first three words, or entity prediction is performed on the last three words of the target prediction word, etc.

Specifically, in inputting the first interview sample data into a preset recognition model including first initial parameters, standard label prediction is performed on the first interview sample data through the direct prediction module in the preset recognition model , after obtaining the standard label distribution and the interview coding vector corresponding to the first interview sample data, through each auxiliary prediction module in the preset recognition model, the first interview sample data is subjected to different views of auxiliary label prediction according to the interview coding vector It can be understood that the above description has pointed out that the basis of each auxiliary prediction module for the entity discrimination of words in the first interview sample data is different, that is, each auxiliary prediction module uses different word views to predict auxiliary labels. , and then output the entity prediction result for each word in the first interview sample data, that is, the auxiliary label distribution.

In a specific embodiment, as shown in FIG. 3 , step S30 includes:

S301: Obtain at least two second forward coding vectors and at least two second reverse coding vectors in the interview coding vectors; the second forward coding refers to performing the forward coding on each of the unlabeled sample words according to the normal The second reverse encoding refers to encoding each of the unlabeled sample words in reverse order to obtain;

Understandably, the unlabeled sample words refer to each segmented word obtained after segmenting the first interview sample data. An auxiliary prediction module is associated with a second forward coding vector.

Further, after encoding the first interview sample data by the encoder in the direct prediction module in the preset recognition model, it can be understood that the encoder is a bidirectional cyclic neural network encoder, so the output encoding vector contains The second forward coding vector obtained by encoding the unlabeled sample words in the first interview sample data in the forward order (that is, starting from the first unlabeled sample word in the first interview sample data, from front to back coding), and the second reverse coding vector obtained by encoding each unlabeled sample word in the first interview sample data in reverse order (that is, starting from the last unlabeled sample word in the first interview sample data, Encoding from back to front).

Exemplarily, in this embodiment, obtaining at least two second forward coding vectors in the coding of the sample vector means that the second forward coding vector may be the first t unlabeled sample words in the unlabeled sample word. A coding vector sequence composed of words; the second forward coding vector can also be a coding vector sequence composed of t-1 unlabeled sample words before the unlabeled sample word; similarly, at least two second reverse The coding vector refers to that the second reverse coding vector may be a coding vector sequence consisting of t unlabeled sample words after the unlabeled sample word; A sequence of encoded vectors consisting of t+1 unlabeled sample words.

S302: Determine the distribution of forward auxiliary labels corresponding to each of the unlabeled sample words according to each of the second forward coding vectors; Reverse auxiliary label distribution for words.

Understandably, after obtaining at least two second forward coding vectors and at least two second reverse coding vectors in the interview coding vector, each unlabeled sample word in the corresponding second forward vector coding is Perform entity prediction to obtain the forward auxiliary label distribution corresponding to each second forward coding vector, that is, after a second forward coding vector is input to the auxiliary prediction module, a corresponding forward auxiliary label distribution will be output. Similarly, entity prediction is performed by each unlabeled sample word in the corresponding second reverse vector encoding, and the reverse auxiliary label distribution corresponding to each second reverse encoding vector is obtained.

Exemplarily, it is assumed that entity prediction needs to be performed on the unlabeled sample word A in the first interview sample data, and the second forward encoding vector may be composed of t unlabeled sample words before the unlabeled sample word A. Therefore, in the auxiliary prediction module, entity prediction is performed on the unlabeled sample word A through the first t unlabeled sample words; The encoding vector sequence composed of t-1 unlabeled sample words, in another auxiliary prediction module, entity prediction is performed on the unlabeled sample word A through the first t-1 unlabeled sample words. It can be understood that the distribution of the forward auxiliary labels output after the entity prediction of the unlabeled sample word A by the above-mentioned two second forward coding vectors is different, because the unlabeled sample words are compared to the unlabeled sample words through the first t unlabeled sample words. When the word A is used for entity prediction, the unlabeled sample word A can be touched; and when the unlabeled sample word A is predicted by the first t-1 unlabeled sample words, the unlabeled sample word A is also before the unlabeled sample word A. There is another unlabeled sample word, and the unlabeled sample word A cannot be touched at this time, thereby forming the view difference when the auxiliary prediction module performs entity prediction on the unlabeled sample word A. The same is true for the second reverse encoding vector , and will not be repeated here.

S40: Determine the total loss value of the preset recognition model according to the distribution of each auxiliary label and the distribution of the standard label;

It can be understood that, through each auxiliary prediction module in the preset recognition model, auxiliary label prediction is performed on the first interview sample data according to the coding vector, and the auxiliary label distribution output by each auxiliary prediction module is obtained. After that, determine the KL (Kullback–Leibler divergence, relative entropy) divergence between each auxiliary label distribution and the standard label distribution, which can be specifically determined according to the following expression:

Among them, D _KL (p||q) refers to the KL divergence between the auxiliary label distribution and the interview label distribution; p(x _i ) represents the ith unlabeled sample word in the first interview sample data The auxiliary label distribution output by the corresponding auxiliary prediction module; q(x _i ) represents the standard label distribution corresponding to the unlabeled sample words of p(x _i ).

Further, the total loss value of the preset recognition model is determined by the following expression:

Among them, L _VCT (θ) is the total loss value of the preset recognition model; |D _ul | is the number of the first interview sample data in the preset interview sample data set; k is the number of auxiliary prediction modules in the preset recognition model ; q _θ (y|x _i ) is the standard label distribution corresponding to the i-th unlabeled sample word in the θ-th first interview sample data;

is the auxiliary label distribution output by the jth auxiliary prediction module of the ith unlabeled sample word in the θth first interview sample data;

is the KL divergence between each auxiliary label distribution and the standard label distribution of the i-th unlabeled sample word in the θ-th first interview sample data.

S50: When the total loss value does not reach the preset convergence condition, update and iterate the first initial parameter of the preset recognition model, until the total loss value reaches the preset convergence condition, after the convergence The preset recognition model is recorded as the interview entity recognition model.

Understandably, the convergence condition can be the condition that the total loss value is less than the set threshold, that is, when the total loss value is less than the set threshold, the training is stopped; the convergence condition can also be that the total loss value after 10,000 calculations is If the condition is very small and will not decrease again, that is, when the total loss value is small and will not decrease after 10,000 calculations, the training is stopped, and the preset recognition model after convergence is recorded as the interview entity recognition model.

Further, after determining the total loss value of the preset recognition model according to the distribution of each auxiliary label and the standard label distribution corresponding to the first interview sample data, after the total loss value does not reach the preset convergence condition , adjust the first initial parameters of the preset recognition model according to the total loss value, and re-input the first interview sample data into the preset recognition model after adjusting the initial parameters, so that the first interview sample data corresponds to the When the total loss value reaches the preset convergence condition, select another first interview sample data in the preset interview sample data set, and execute the above steps S10 to S50, and obtain the predicted loss value corresponding to the first interview sample data, And when the total loss value does not reach the preset convergence condition, adjust the first initial parameter of the preset recognition model again according to the total loss value, so that the total loss value corresponding to the first interview sample data reaches the preset convergence condition .

In this way, after training the preset recognition model with all the first interview sample data in the preset interview sample data set, the output results of the preset recognition model can be continuously approached to the accurate results, so that the recognition accuracy rate is getting higher and higher. Until the total loss value corresponding to all the first interview sample data reaches the preset convergence condition, the preset recognition model after convergence is recorded as the interview entity recognition model.

In this embodiment, by combining the direct prediction module of supervised learning and the auxiliary prediction module of unsupervised learning, on the basis of the direct prediction module, the auxiliary prediction module can give the model more different data features (such as each The auxiliary label distribution output by the auxiliary prediction module) improves the training efficiency of the interview entity recognition model and improves the accuracy of the trained interview entity recognition model.

In another specific embodiment, in order to ensure the privacy and security of the interview entity recognition model in the above embodiment, the interview entity recognition model may be stored in the blockchain. Among them, Blockchain is a storage structure of encrypted and chained transactions formed by blocks.

For example, the header of each block can include not only the hash values of all transactions in the block, but also the hash values of all transactions in the previous block, so that the transactions in the block can be tamper-proof based on the hash value. And anti-counterfeiting; the newly generated transaction is filled into the block and after the consensus of the nodes in the blockchain network, it will be appended to the end of the blockchain to form a chain growth.

In one embodiment, as shown in FIG. 4 , the preset interview sample data set also includes at least one second interview sample data with the interview label; Before the preset encoder in the recognition model performs standard label prediction on the first interview sample data, the method further includes:

S01: Input the second interview sample data into the preset recognition model, and perform direct label prediction on the second interview sample data through a preset prediction module that includes a second initial parameter in the preset recognition model , obtain the direct prediction label corresponding to the second interview sample data;

It is understandable that the interview label refers to the manual labeling of each word in the second interview sample data in advance. For example, "I go to school in Peking University", "Peking University" will be marked as a school in advance. Name entity.

Further, after obtaining the preset interview sample data set, input the second interview sample data including the interview label into the preset recognition model, and use the preset prediction module including the second parameter in the preset recognition model Perform direct label prediction on the interview sample data to obtain direct predicted labels corresponding to each word in the second interview sample data.

In a specific embodiment, step S01 includes:

Perform word segmentation processing on the second interview sample data to obtain each marked sample word corresponding to the second interview sample data;

Exemplarily, word segmentation processing may be performed on the second interview sample data through a stammering word segmentation method to obtain each marked sample word corresponding to the second interview sample data.

The encoder in the preset prediction module performs encoding processing on each of the labeled sample words to obtain a first forward encoding vector and a first reverse encoding vector; the first forward encoding refers to The labeled sample words are obtained by encoding in a forward order; the first reverse encoding refers to encoding each of the labeled sample words in a reverse order;

The encoder is a bidirectional cyclic neural network encoder, so the output encoding vector includes the first forward encoding vector obtained by encoding the labeled sample words in the second interview sample data in the forward order, and the second The first reverse encoding vector obtained by encoding each labeled sample word in the interview sample data in reverse order.

According to the first forward coding vector and the first reverse coding vector, the labeling classifier in the preset prediction module is used to label each of the labelled sample words, and obtain the same label as each labelled sample. The direct prediction label corresponding to the term.

It can be understood that after the encoder in the preset prediction module performs encoding processing on each of the labeled sample words to obtain the first forward encoding vector and the first reverse encoding vector, the preset prediction module The labeling classifier performs label classification for each labeling sample word according to the first forward coding vector and the first reverse coding vector, and obtains a direct prediction label corresponding to each labeling sample word. Exemplarily, assume that the second interview sample data is "I study at Peking University", where "Peking University" is manually marked as a school name entity, and then after the second interview sample data is input into the preset prediction module, the Each marked word (“I”, “Zai”, “Peking University”, “Learning”, etc.) in the second interview sample data is coded to obtain the first forward coding starting from “I” and coded from front to back The encoding vector, and the second reverse encoding vector encoded from the back to the front starting from "Peking University", to classify "Peking University" according to the second forward encoding vector and the second reverse encoding vector. The direct prediction label corresponding to "Peking University".

S02: Determine the prediction loss value of the preset encoder according to the direct prediction label and the interview label;

It can be understood that, according to the first forward coding vector and the first reverse coding vector, the labeling classifier in the preset prediction module performs label classification on each of the labeled sample words, and obtains the same After the direct prediction label corresponding to each labeled sample word, the direct prediction label corresponding to each labeled sample word is matched with the interview labeled label corresponding to each labeled sample word, and then the prediction loss value of the preset encoder is determined. .

Specifically, the interview label includes a plurality of sample entity labels; in step S02, including:

(1) Obtain the sample entity labels corresponding to each of the labeled sample words;

(2) According to the sample entity label and the direct prediction label corresponding to the same labeled sample word, determine the label loss value corresponding to the labeled sample word;

(3) According to the label loss value corresponding to each labeled sample word, the predicted loss value is determined by the cross entropy loss function.

Understandably, the sample entity label is the label corresponding to each labeled sample word that exists in a second interview sample data, that is, there may be two or more labeled words in a second interview sample data. . Therefore, when the second interview sample data is input into the preset recognition model, the second interview sample data is directly labeled by the preset prediction module including the second initial parameter in the preset recognition model Predict, after obtaining the direct prediction label corresponding to the second interview sample data, obtain the sample entity label corresponding to each of the labeled sample words, according to the sample entity label corresponding to the same labeled sample word and the direct prediction label, to determine the label loss value corresponding to the labeled sample word; according to the label loss value corresponding to each labeled sample word, determine the predicted loss value through the cross entropy loss function.

Further, the predicted loss value can be determined according to the following expression:

Among them, L _sup (β) is the prediction loss value; |D _l | is the number of all the second interview sample data;

refers to the sample entity label corresponding to the δth labeled sample word in the βth second interview sample data; p _β (b|a _δ ) refers to the δth labeled sample in the βth second interview sample data The direct prediction label corresponding to the word; CE() is the cross entropy loss function.

Understandably, for the entire interview entity recognition model, in the training process, it first trains the direct prediction module through the second interview sample data with the interview label, and then uses the direct prediction module to analyze the first interview without the interview label through the direct prediction module. The standard label prediction is performed on the interview sample data, and the auxiliary label prediction is performed on the first interview sample data through each auxiliary prediction module, thereby characterizing the overall loss value of the interview entity recognition model in this application. A superposition of the predicted loss value and the total loss value in step S40.

S03: When the predicted loss value does not reach the preset convergence condition, update and iterate the second initial parameter of the preset preset prediction module, until the predicted loss value reaches the preset convergence condition, change the The preset prediction module after convergence is recorded as the direct prediction module.

Understandably, the convergence condition can be the condition that the predicted loss value is less than the set threshold, that is, when the predicted loss value is less than the set threshold, the training is stopped; the convergence condition can also be that the predicted loss value after 10,000 calculations is The condition that is very small and will not decrease again, that is, when the predicted loss value is small and will not decrease after 10,000 calculations, stop training, and record the preset prediction module after convergence as the direct prediction module .

Further, after determining the predicted loss value of the preset prediction module according to the direct prediction label of the second interview sample data and the interview label, when the predicted loss value does not reach the preset convergence condition, according to the prediction The loss value adjusts the second initial parameter of the preset prediction module, and re-inputs the second interview sample data into the preset prediction module after adjusting the initial parameters, so that the predicted loss value corresponding to the second interview sample data reaches the preset value. When the convergence condition is set, select another second interview sample data in the preset interview sample data set, and perform the above steps S01 to S03, and obtain the predicted loss value corresponding to the second interview sample data, and in the predicted loss value When the preset convergence condition is not reached, the second initial parameter of the preset prediction module is adjusted again according to the predicted loss value, so that the predicted loss value corresponding to the second interview sample data reaches the preset convergence condition.

In this way, after the preset prediction module is trained by using all the second interview sample data in the preset interview sample data set, the results output by the preset prediction module can be continuously approached to the accurate results, so that the recognition accuracy rate is getting higher and higher. Until all the prediction loss values corresponding to the second interview sample data reach the preset convergence condition, the preset prediction module after convergence is recorded as the direct prediction module.

In one embodiment, as shown in FIG. 5 , the present application also proposes a method for extracting interview information entities, which is described by taking the method applied to the server in FIG. 1 as an example, including the following steps:

S60: Obtain interview information of the target interviewee; the interview information includes at least one interview sentence; one of the interview sentences includes a plurality of interview information words;

Understandably, the interview information can be the information in the paper resume provided by the target interviewer, or it can be obtained by the target interviewee by taking the voice information of the target interviewer and converting it into text information during the self-introduction process.

S70: Input the interview sentence into the interview entity recognition model, extract and identify the interview information words in the interview sentence, and obtain entity recognition results corresponding to the interview information words; the interview entity recognition The model is obtained according to the above-mentioned interview entity recognition model training method;

Specifically, after obtaining the interview information of the target interviewee, split the interview information with a period at the end of the interview information to obtain an interview sentence, input the interview sentence into the interview entity recognition model, and analyze the interview information in the interview sentence. Entity recognition is performed on the words to determine the entity recognition result corresponding to each interview information word. Understandably, the entity recognition results include specific entity results, that is, the entity names corresponding to the interview information words are identified, such as school name entities, name entities, etc.; entity recognition results also include non-entity recognition results, such as " Interview information words such as "I" and "Yes" are non-entity.

S80: Insert the entity recognition result into a preset interview information storage template according to a preset matching rule.

Specifically, after the interview sentence is input into the interview entity recognition model, the interview information words in the interview sentence are extracted and recognized, and the entity recognition result corresponding to each interview information word is obtained. The recognition result is inserted into the preset interview information storage template according to the preset matching rules. The preset interview information storage template includes multiple slots to be filled, for example, the template includes specific names, specific schools, specific genders, etc. to be filled, and then the entity results are matched with the slots to be filled, so as to "Peking University" is filled into the specific school slot, and then an interview information page corresponding to the target interviewee is formed.

It should be understood that the size of the sequence numbers of the steps in the above embodiments does not mean the sequence of execution, and the execution sequence of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

In one embodiment, an interview entity recognition model training device is provided, and the interview entity recognition model training device corresponds one-to-one with the interview entity recognition model training method in the above embodiment. As shown in FIG. 5 , the interview entity recognition model training device includes a sample data acquisition module 10 , a standard label prediction module 20 , an auxiliary label prediction module 30 , a total loss value determination module 40 and a model training module 50 . The detailed description of each functional module is as follows:

A sample data acquisition module 10, configured to acquire a preset interview sample data set; the preset interview sample data set includes at least one first interview sample data without an interview label;

The standard label prediction module 20 is used to input the first interview sample data into a preset recognition model including the first initial parameter, and the first interview sample data is analyzed by the direct prediction module in the preset recognition model. Perform standard label prediction to obtain standard label distribution and interview coding vector corresponding to the first interview sample data;

The auxiliary label prediction module 30 is configured to perform auxiliary label prediction on the first interview sample data according to the coding vector through each auxiliary prediction module in the preset recognition model, and obtain the output of each auxiliary prediction module. Auxiliary label distribution;

A total loss value determination module 40, configured to determine the total loss value of the preset recognition model according to each of the auxiliary label distribution and the standard label distribution;

A model training module 50, configured to update and iterate the first initial parameter of the preset recognition model when the total loss value does not reach a preset convergence condition, until the total loss value reaches the preset convergence condition , the preset recognition model after convergence is recorded as the interview entity recognition model.

Preferably, the interview entity recognition model training device further includes:

The direct label prediction module is used to input the second interview sample data into the preset recognition model, and the second interview sample is analyzed by the preset prediction module including the second initial parameter in the preset recognition model. performing direct label prediction on the data to obtain a direct prediction label corresponding to the second interview sample data;

a predicted loss value determination module, configured to determine the predicted loss value of the preset encoder according to the direct prediction label and the interview label;

A parameter update module, configured to update and iterate the second initial parameter of the preset preset prediction module when the predicted loss value does not reach a preset convergence condition, until the predicted loss value reaches the preset convergence When conditions are met, the preset prediction module after convergence is recorded as the direct prediction module.

Preferably, the direct label prediction module includes:

A word segmentation processing unit, configured to perform word segmentation processing on the second interview sample data to obtain each labeled sample word corresponding to the second interview sample data;

an encoding processing unit, configured to perform encoding processing on each of the labeled sample words by an encoder in the preset prediction module to obtain a first forward encoding vector and a first reverse encoding vector; the first forward encoding Encoding refers to encoding each of the labeled sample words in a forward order; the first reverse encoding refers to encoding each of the labeled sample words in a reverse order;

The label classification unit is configured to perform label classification on each of the labeled sample words by the labeling classifier in the preset prediction module according to the first forward coding vector and the first reverse coding vector, to obtain Directly predicted labels corresponding to each of the labeled sample words.

Preferably, the predicted loss value determination module includes:

a sample entity label obtaining unit, configured to obtain the sample entity label corresponding to each of the labeled sample words;

A label loss value determination unit, configured to determine a label loss value corresponding to the labeled sample word according to the sample entity label and the direct prediction label corresponding to the same labeled sample word;

The predicted loss value determining unit is configured to determine the predicted loss value through the cross entropy loss function according to the label loss value corresponding to each of the labeled sample words.

Preferably, as shown in FIG. 6 , the auxiliary label prediction module 30 includes:

A coding vector obtaining unit 301, configured to obtain at least two second forward coding vectors and at least two second reverse coding vectors in the coding vectors; the second forward coding refers to The words are obtained by encoding in a forward order; the second reverse encoding refers to encoding each of the unlabeled sample words in a reverse order;

Auxiliary label distribution determining unit 302, configured to determine the distribution of forward auxiliary labels corresponding to each of the unlabeled sample words according to each of the second forward coding vectors; at the same time, determine according to the second reverse coding vector Reverse auxiliary label distribution corresponding to each of the unlabeled sample words.

In one embodiment, as shown in FIG. 8, a device for extracting interview information entities is provided, including:

The interview information obtaining module 60 is used to obtain the interview information of the target interviewee; the interview information includes at least one interview sentence; one of the interview sentences includes a plurality of interview information words;

The entity extraction and recognition module 70 is used to input the interview sentence into the interview entity recognition model, extract and recognize the interview information words in the interview sentence, and obtain the entity recognition result corresponding to each interview information word ; Described interview entity recognition model is obtained according to above-mentioned interview entity recognition model training method;

The information storage module 80 is configured to insert the entity recognition result into a preset interview information storage template according to preset matching rules.

For the specific limitations of the interview entity recognition model training device and the interview information entity extraction device, please refer to the limitations on the interview entity recognition model training method and the interview information entity extraction method above, which will not be repeated here. Each module in the above-mentioned interview entity recognition model training device and interview information entity extraction device can be implemented in whole or in part by software, hardware and combinations thereof. The above modules can be embedded in or independent of the processor in the computer device in the form of hardware, or stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

In one embodiment, a computer device is provided, and the computer device may be a server, and its internal structure diagram may be as shown in FIG. 9 . The computer device includes a processor, memory, a network interface, and a database connected by a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a readable storage medium, an internal memory. The readable storage medium stores an operating system, computer readable instructions and a database. The internal memory provides an environment for the execution of the operating system and computer-readable instructions in the readable storage medium. The database of the computer device is used to store the data used in the interview entity recognition model training method in the above-mentioned embodiment, or the data used in the interview information entity extraction method in the above-mentioned embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. The computer-readable instructions, when executed by the processor, implement a method for training an interview entity recognition model, or the computer-readable instructions, when executed by the processor, implement a method for extracting interview information entities. The readable storage medium provided by this embodiment includes a non-volatile readable storage medium and a volatile readable storage medium.

In one embodiment, there is provided a computer apparatus comprising a memory, a processor, and computer readable instructions stored in the memory and executable on the processor, the processor executing the computer readable instructions When implementing the following steps:

obtaining a preset interview sample data set; the preset interview sample data set includes at least one first interview sample data without an interview label;

Inputting the first interview sample data into a preset recognition model including the first initial parameter, and performing standard label prediction on the first interview sample data through the direct prediction module in the preset recognition model to obtain a standard label distribution and an interview coding vector corresponding to the first interview sample data;

Through each auxiliary prediction module in the preset recognition model, carry out auxiliary label prediction on the first interview sample data according to the interview coding vector, and obtain the auxiliary label distribution output by each of the auxiliary prediction modules;

Determine the total loss value of the preset recognition model according to the distribution of each auxiliary label and the standard label distribution; when the total loss value does not reach the preset convergence condition, update the first step of iterating the preset recognition model. an initial parameter, until the total loss value reaches the preset convergence condition, the preset recognition model after convergence is recorded as the interview entity recognition model.

In one embodiment, there is provided a computer apparatus comprising a memory, a processor, and computer readable instructions stored in the memory and executable on the processor, the processor executing the computer readable instructions When implementing the following steps:

Obtain the interview information of the target interviewee; the interview information includes at least one interview sentence; one of the interview sentences includes a plurality of interview information words;

The interview sentence is input into the interview entity recognition model, and the interview information words in the interview sentence are extracted and recognized to obtain entity recognition results corresponding to the interview information words; the interview entity recognition model is Obtained according to the above interview entity recognition model training method;

Insert the entity recognition result into a preset interview information storage template according to preset matching rules.

In one embodiment, one or more readable storage media are provided storing computer-readable instructions that, when executed by one or more processors, cause the one or more processors to perform the following: step:

obtaining a preset interview sample data set; the preset interview sample data set includes at least one first interview sample data without an interview label;

Inputting the first interview sample data into a preset recognition model including the first initial parameter, and performing standard label prediction on the first interview sample data through the direct prediction module in the preset recognition model to obtain a standard label distribution and an interview coding vector corresponding to the first interview sample data;

Through each auxiliary prediction module in the preset recognition model, carry out auxiliary label prediction on the first interview sample data according to the interview coding vector, and obtain the auxiliary label distribution output by each of the auxiliary prediction modules;

Determine the total loss value of the preset recognition model according to each of the auxiliary label distribution and the standard label distribution;

When the total loss value does not reach the preset convergence condition, update and iterate the first initial parameter of the preset recognition model, until the total loss value reaches the preset convergence condition, all the The aforementioned preset recognition model is recorded as the interview entity recognition model.

In one embodiment, one or more readable storage media are provided storing computer-readable instructions that, when executed by one or more processors, cause the one or more processors to perform the following: step:

Obtain the interview information of the target interviewee; the interview information includes at least one interview sentence; one of the interview sentences includes a plurality of interview information words;

The interview sentence is input into the interview entity recognition model, and the interview information words in the interview sentence are extracted and recognized to obtain entity recognition results corresponding to the interview information words; the interview entity recognition model is Obtained according to the above interview entity recognition model training method;

Insert the entity recognition result into a preset interview information storage template according to preset matching rules.

Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a non-volatile computer. In a readable storage medium or a volatile computer-readable storage medium, the computer-readable instructions, when executed, may include the processes of the foregoing method embodiments. Wherein, any reference to memory, storage, database or other medium used in the various embodiments provided in this application may include non-volatile and/or volatile memory. Nonvolatile memory may include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain Road (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

Those skilled in the art can clearly understand that, for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. Module completion, that is, dividing the internal structure of the device into different functional units or modules to complete all or part of the functions described above.

The above-mentioned embodiments are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the above-mentioned embodiments, those of ordinary skill in the art should understand that: it can still be used for the above-mentioned implementations. The technical solutions described in the examples are modified, or some technical features thereof are equivalently replaced; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions in the embodiments of the application, and should be included in the within the scope of protection of this application.

Claims (20)

1. An interview entity recognition model training method, which includes:

   obtaining a preset interview sample data set; the preset interview sample data set includes at least one first interview sample data without an interview label;

   Inputting the first interview sample data into a preset recognition model including the first initial parameter, and performing standard label prediction on the first interview sample data through the direct prediction module in the preset recognition model to obtain a standard label distribution and an interview coding vector corresponding to the first interview sample data;

   Through each auxiliary prediction module in the preset recognition model, carry out auxiliary label prediction on the first interview sample data according to the interview coding vector, and obtain the auxiliary label distribution output by each of the auxiliary prediction modules;

   Determine the total loss value of the preset recognition model according to each of the auxiliary label distribution and the standard label distribution;

   When the total loss value does not reach the preset convergence condition, update and iterate the first initial parameter of the preset recognition model, until the total loss value reaches the preset convergence condition, all the The aforementioned preset recognition model is recorded as the interview entity recognition model.
2. The method for training an interview entity recognition model according to claim 1, wherein the preset interview sample data set further includes at least one second interview sample data with the interview label; Before performing standard label prediction on the first interview sample data, the preset encoder in , includes:

   Input the second interview sample data into the preset recognition model, and perform direct label prediction on the second interview sample data through the preset prediction module that includes the second initial parameter in the preset recognition model, to obtain the direct prediction label corresponding to the second interview sample data;

   Determine the prediction loss value of the preset encoder according to the direct prediction label and the interview label;

   When the predicted loss value does not reach the preset convergence condition, update and iterate the second initial parameter of the preset preset prediction module until the predicted loss value reaches the preset convergence condition, after the convergence The preset prediction module is recorded as the direct prediction module.
3. The method for training an interview entity recognition model according to claim 2, wherein the direct label prediction is performed on the second interview sample data by using a preset prediction module including a second initial parameter in the preset recognition model, to obtain The direct prediction label corresponding to the second interview sample data, including:

   Perform word segmentation processing on the second interview sample data to obtain each marked sample word corresponding to the second interview sample data;

   The encoder in the preset prediction module performs encoding processing on each of the labeled sample words to obtain a first forward encoding vector and a first reverse encoding vector; the first forward encoding refers to The labeled sample words are obtained by encoding in a forward order; the first reverse encoding refers to encoding each of the labeled sample words in a reverse order;

   According to the first forward coding vector and the first reverse coding vector, the labeling classifier in the preset prediction module is used to label each of the labelled sample words, and obtain the same label as each labelled sample. The direct prediction label corresponding to the term.
4. The method for training an interview entity recognition model according to claim 2, wherein the interview label includes a plurality of sample entity labels; the preset encoder is determined according to the direct prediction label and the interview label The predicted loss value of , including:

   obtaining the sample entity labels corresponding to each of the labeled sample words;

   Determine the label loss value corresponding to the labeled sample word according to the sample entity label and the directly predicted label corresponding to the same labeled sample word;

   According to the label loss value corresponding to each of the labeled sample words, the predicted loss value is determined through a cross entropy loss function.
5. The method for training an interview entity recognition model according to claim 1, wherein the auxiliary label prediction is performed on the first interview sample data according to the coding vector through each auxiliary prediction module in the preset recognition model, Obtain the auxiliary label distribution output by each of the auxiliary prediction modules, including:

   Obtain at least two second forward encoding vectors and at least two second reverse encoding vectors in the interview encoding vectors; the second forward encoding refers to encoding each unlabeled sample word in a forward order Obtained; the second reverse encoding refers to encoding each of the unlabeled sample words in reverse order to obtain;

   According to each of the second forward coding vectors, determine the distribution of each forward auxiliary label corresponding to each of the unlabeled sample words; at the same time, determine the distribution of each of the unlabeled sample words according to the second reverse coding vector. The corresponding reverse auxiliary label distribution.
6. A method for extracting interview information entities, comprising:

   Obtain the interview information of the target interviewee; the interview information includes at least one interview sentence; one of the interview sentences includes a plurality of interview information words;

   The interview sentence is input into the interview entity recognition model, and the interview information words in the interview sentence are extracted and recognized to obtain entity recognition results corresponding to the interview information words; the interview entity recognition model is Obtained according to the interview entity recognition model training method as described in any one of claims 1 to 5;

   Insert the entity recognition result into a preset interview information storage template according to preset matching rules.
7. An interview entity recognition model training device, comprising:

   a sample data acquisition module, configured to acquire a preset interview sample data set; the preset interview sample data set includes at least one first interview sample data without an interview label;

   The standard label prediction module is used to input the first interview sample data into a preset recognition model including the first initial parameter, and perform the first interview sample data through the direct prediction module in the preset recognition model. Standard label prediction, to obtain standard label distribution and interview coding vector corresponding to the first interview sample data;

   The auxiliary label prediction module is used to perform auxiliary label prediction on the first interview sample data according to the coding vector through each auxiliary prediction module in the preset recognition model, and obtain the auxiliary label output with each of the auxiliary prediction modules. label distribution;

   a total loss value determination module, configured to determine the total loss value of the preset recognition model according to each of the auxiliary label distributions and the standard label distribution;

   A model training module, configured to update and iterate the first initial parameter of the preset recognition model when the total loss value does not reach the preset convergence condition, until the total loss value reaches the preset convergence condition , and record the preset recognition model after convergence as an interview entity recognition model.
8. A device for extracting interview information entities, comprising:

   an interview information obtaining module, used for obtaining interview information of a target interviewee; the interview information includes at least one interview sentence; one interview sentence includes a plurality of interview information words;

   An entity extraction and recognition module is used to input the interview sentence into the interview entity recognition model, extract and recognize the interview information words in the interview sentence, and obtain entity recognition results corresponding to each of the interview information words; The interview entity recognition model is obtained according to the interview entity recognition model training method as described in any one of claims 1 to 5;

   An information storage module, configured to insert the entity recognition result into a preset interview information storage template according to preset matching rules.
9. A computer device comprising a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer-readable instructions:

   obtaining a preset interview sample data set; the preset interview sample data set includes at least one first interview sample data without an interview label;

   Inputting the first interview sample data into a preset recognition model including the first initial parameter, and performing standard label prediction on the first interview sample data through the direct prediction module in the preset recognition model to obtain a standard label distribution and an interview coding vector corresponding to the first interview sample data;

   Through each auxiliary prediction module in the preset recognition model, carry out auxiliary label prediction on the first interview sample data according to the interview coding vector, and obtain the auxiliary label distribution output by each of the auxiliary prediction modules;

   Determine the total loss value of the preset recognition model according to the distribution of each auxiliary label and the standard label distribution; when the total loss value does not reach the preset convergence condition, update the first step of iterating the preset recognition model. An initial parameter, until the total loss value reaches the preset convergence condition, the preset recognition model after convergence is recorded as the interview entity recognition model.
10. The computer device according to claim 9, wherein the preset interview sample data set further includes at least one second interview sample data with the interview label; Before the encoder performs standard label prediction on the first interview sample data, the processor further implements the following steps when executing the computer-readable instructions:

    Input the second interview sample data into the preset recognition model, and perform direct label prediction on the second interview sample data through the preset prediction module that includes the second initial parameter in the preset recognition model, to obtain the direct prediction label corresponding to the second interview sample data;

    Determine the prediction loss value of the preset encoder according to the direct prediction label and the interview label;

    When the predicted loss value does not reach the preset convergence condition, update and iterate the second initial parameter of the preset preset prediction module until the predicted loss value reaches the preset convergence condition, after the convergence The preset prediction module is recorded as the direct prediction module.
11. The computer device according to claim 10, wherein the direct label prediction is performed on the second interview sample data through a preset prediction module including a second initial parameter in the preset recognition model, and a result that is the same as the first one is obtained. The direct prediction labels corresponding to the second interview sample data, including:

    Perform word segmentation processing on the second interview sample data to obtain each marked sample word corresponding to the second interview sample data;

    The encoder in the preset prediction module performs encoding processing on each of the labeled sample words to obtain a first forward encoding vector and a first reverse encoding vector; the first forward encoding refers to The labeled sample words are obtained by encoding in a forward order; the first reverse encoding refers to encoding each of the labeled sample words in a reverse order;

    According to the first forward coding vector and the first reverse coding vector, the labeling classifier in the preset prediction module is used to label each of the labelled sample words, and obtain the same label as each labelled sample. The direct prediction label corresponding to the term.
12. The computer device according to claim 10, wherein the interview annotation label includes a plurality of sample entity labels; the prediction loss value of the preset encoder is determined according to the direct prediction label and the interview annotation label ,include:

    obtaining the sample entity labels corresponding to each of the labeled sample words;

    Determine the label loss value corresponding to the labeled sample word according to the sample entity label and the directly predicted label corresponding to the same labeled sample word;

    According to the label loss value corresponding to each of the labeled sample words, the predicted loss value is determined through a cross entropy loss function.
13. The computer device according to claim 9, wherein the auxiliary label prediction is performed on the first interview sample data according to the coding vector through each auxiliary prediction module in the preset recognition model, and the corresponding prediction is obtained by each auxiliary prediction module. The auxiliary label distribution output by the auxiliary prediction module, including:

    Obtain at least two second forward encoding vectors and at least two second reverse encoding vectors in the interview encoding vectors; the second forward encoding refers to encoding each unlabeled sample word in a forward order Obtained; the second reverse encoding refers to encoding each of the unlabeled sample words in reverse order to obtain;

    According to each of the second forward coding vectors, determine the distribution of each forward auxiliary label corresponding to each of the unlabeled sample words; at the same time, determine the distribution of each of the unlabeled sample words according to the second reverse coding vector. The corresponding reverse auxiliary label distribution.
14. A computer device comprising a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer-readable instructions:

    Obtain interview information of the target interviewee; the interview information includes at least one interview sentence; one interview sentence includes multiple interview information words;

    The interview sentence is input into the interview entity recognition model, and the interview information words in the interview sentence are extracted and recognized to obtain entity recognition results corresponding to the interview information words; the interview entity recognition model is Obtained according to the interview entity recognition model training method as described in any one of claims 1 to 5;

    Insert the entity recognition result into a preset interview information storage template according to preset matching rules.
15. One or more readable storage media storing computer-readable instructions, wherein the computer-readable instructions, when executed by one or more processors, cause the one or more processors to perform the following steps:

    obtaining a preset interview sample data set; the preset interview sample data set includes at least one first interview sample data without an interview label;

    Inputting the first interview sample data into a preset recognition model including the first initial parameter, and performing standard label prediction on the first interview sample data through the direct prediction module in the preset recognition model to obtain a standard label distribution and an interview coding vector corresponding to the first interview sample data;

    Through each auxiliary prediction module in the preset recognition model, carry out auxiliary label prediction on the first interview sample data according to the interview coding vector, and obtain the auxiliary label distribution output by each of the auxiliary prediction modules;

    Determine the total loss value of the preset recognition model according to each of the auxiliary label distribution and the standard label distribution;

    When the total loss value does not reach the preset convergence condition, update and iterate the first initial parameter of the preset recognition model, until the total loss value reaches the preset convergence condition, all the The aforementioned preset recognition model is recorded as the interview entity recognition model.
16. The readable storage medium according to claim 15, wherein the preset interview sample data set further includes at least one second interview sample data with the interview label; Before the preset encoder performs standard label prediction on the first interview sample data, when the computer-readable instructions are executed by one or more processors, the one or more processors further perform the following steps:

    Input the second interview sample data into the preset recognition model, and perform direct label prediction on the second interview sample data through the preset prediction module that includes the second initial parameter in the preset recognition model, to obtain the direct prediction label corresponding to the second interview sample data;

    Determine the prediction loss value of the preset encoder according to the direct prediction label and the interview label;

    When the predicted loss value does not reach the preset convergence condition, update and iterate the second initial parameter of the preset preset prediction module until the predicted loss value reaches the preset convergence condition, after the convergence The preset prediction module is recorded as the direct prediction module.
17. The readable storage medium according to claim 16, wherein the direct label prediction is performed on the second interview sample data by a preset prediction module including the second initial parameter in the preset recognition model, and the result is obtained with the The direct prediction labels corresponding to the second interview sample data, including:

    Perform word segmentation processing on the second interview sample data to obtain each marked sample word corresponding to the second interview sample data;

    The encoder in the preset prediction module performs encoding processing on each of the labeled sample words to obtain a first forward encoding vector and a first reverse encoding vector; the first forward encoding refers to The labeled sample words are obtained by encoding in a forward order; the first reverse encoding refers to encoding each of the labeled sample words in a reverse order;

    According to the first forward coding vector and the first reverse coding vector, the labeling classifier in the preset prediction module is used to label each of the labelled sample words, and obtain the same label as each labelled sample. The direct prediction label corresponding to the term.
18. The readable storage medium according to claim 16, wherein the interview annotation label includes a plurality of sample entity labels; the prediction of the preset encoder is determined according to the direct prediction label and the interview annotation label Loss values, including:

    obtaining the sample entity labels corresponding to each of the labeled sample words;

    Determine the label loss value corresponding to the labeled sample word according to the sample entity label and the directly predicted label corresponding to the same labeled sample word;

    According to the label loss value corresponding to each of the labeled sample words, the predicted loss value is determined through a cross entropy loss function.
19. The readable storage medium according to claim 15, wherein the auxiliary label prediction is performed on the first interview sample data according to the coding vector through each auxiliary prediction module in the preset recognition model, and the result is obtained with The auxiliary label distribution output by each of the auxiliary prediction modules, including:

    Obtain at least two second forward encoding vectors and at least two second reverse encoding vectors in the interview encoding vectors; the second forward encoding refers to encoding each unlabeled sample word in a forward order Obtained; the second reverse encoding refers to encoding each of the unlabeled sample words in reverse order to obtain;

    According to each of the second forward coding vectors, determine the distribution of each forward auxiliary label corresponding to each of the unlabeled sample words; at the same time, according to the second reverse coding vector The corresponding reverse auxiliary label distribution.
20. One or more readable storage media storing computer-readable instructions, wherein the computer-readable instructions, when executed by one or more processors, cause the one or more processors to perform the following steps:

    Obtain the interview information of the target interviewee; the interview information includes at least one interview sentence; one of the interview sentences includes a plurality of interview information words;

    The interview sentence is input into the interview entity recognition model, and the interview information words in the interview sentence are extracted and recognized to obtain entity recognition results corresponding to each of the interview information words; the interview entity recognition model is Obtained according to the interview entity recognition model training method as described in any one of claims 1 to 5;

    Inserting the entity recognition result into a preset interview information storage template according to preset matching rules.

Images