An intelligent interview method, device and terminal device

Abstract

This application is applicable to the field of artificial intelligence technology, and provides an intelligent interview method, device and terminal device. The method includes: obtaining the response information of a candidate during an interview; using a preset language model to convert multiple response sentences into corresponding sentence vector information respectively; determining the sentence set vector information corresponding to the response information according to the sentence vector information of the multiple response sentences; using the sentence set vector information to calculate the relationship probability between multiple entities included in the response information; extracting target relationship information from the multiple entities according to the relationship probability; and generating an interview question for the candidate based on the target relationship information. The above method can quickly extract the important parts in the candidate's answer content, facilitate the artificial intelligence interviewer to give necessary and reasonable follow-up questions, and then generate an interview evaluation report. In addition, the interview evaluation report can be uploaded to the blockchain to ensure its security and fairness and transparency.

Description

Technical Field

[0001] This application belongs to the technical field of artificial intelligence, and particularly relates to an intelligent interview method, device, and terminal device. Background Art

[0002] Recruitment interviews are time-consuming and laborious. Especially when there is a large volume of recruitment, due to a large number of candidates waiting for interviews but limited interviewers, interviewers usually need to conduct multiple sessions of interviews continuously, seriously affecting the interview efficiency. To save interview time and improve recruitment efficiency, intelligent interviews have emerged. Intelligent interviews can interact with candidates through machines and automatically complete the evaluation of candidates.

[0003] When conducting an intelligent interview, in order for the machine to accurately ask questions to candidates, it is necessary to process the candidate's answer to the previous question, extract the important part of the candidate's previous answer as a reference basis. Relation extraction plays an important role in this link.

[0004] In the prior art, relation extraction can be achieved in two ways. One is to use the concepts in the text and the corresponding relation instances in the knowledge base to heuristically generate labeled data, and then use the labeled data to train the relation extraction model. However, the labeled data generated in this way will produce noisy labels, resulting in incorrect judgments and incomplete knowledge base information, making the machine ask inaccurate questions. The other way can perform multi-instance learning based on the pre-provided semantic and syntactic knowledge and guide the training of the model according to the learning results. However, the model trained based on the pre-provided semantic and syntactic knowledge can only recognize a certain part or a certain type of relations corresponding to the provided semantic and syntactic knowledge, performs poorly in the recognition of other types of relations, has a narrow scope of application, and cannot be widely applied to intelligent interviews in various different scenarios. Summary of the Invention

[0005] In view of this, the embodiments of this application provide an intelligent interview method, device, and terminal device to solve the problem in the prior art that during intelligent interviews, due to inaccurate relation extraction, the credibility of interview results is relatively low, and it helps to make a more objective evaluation of candidates more accurately.

[0006] The first aspect of the embodiments of this application provides an intelligent interview method, including:

[0007] Obtain the reply information of the candidate during the interview, where the reply information includes multiple reply sentences;

[0008] Adopt a preset language model to respectively convert the multiple reply sentences into corresponding sentence vector information;

[0009] Determine the sentence set vector information corresponding to the reply information according to the sentence vector information of the multiple reply sentences;

[0010] Use the sentence set vector information to calculate the relationship probability between multiple entities included in the reply information;

[0011] Extract the target relationship information from the multiple entities according to the relationship probability;

[0012] Generate an interview question for the candidate based on the target relationship information.

[0013] The second aspect of the embodiments of the present application provides an intelligent interview device, including:

[0014] An acquisition module, configured to acquire the reply information of the candidate during the interview, where the reply information includes multiple reply sentences;

[0015] A conversion module, configured to respectively convert the multiple reply sentences into corresponding sentence vector information by using a preset language model;

[0016] A determination module, configured to determine the sentence set vector information corresponding to the reply information according to the sentence vector information of the multiple reply sentences;

[0017] A calculation module, configured to calculate the relationship probability between multiple entities included in the reply information by using the sentence set vector information;

[0018] An extraction module, configured to extract the target relationship information from the multiple entities according to the relationship probability;

[0019] A generation module, configured to generate an interview question for the candidate based on the target relationship information.

[0020] The third aspect of the embodiments of the present application provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the intelligent interview method described in the first aspect above is implemented.

[0021] The fourth aspect of the embodiments of the present application provides a computer-readable storage medium, where the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the intelligent interview method described in the first aspect above is implemented.

[0022] The fifth aspect of the embodiments of the present application provides a computer program product. When the computer program product runs on a terminal device, the terminal device is enabled to execute the intelligent interview method described in any item of the first aspect above.

[0023] Compared with the prior art, the embodiments of the present application include the following advantages:

[0024] In the embodiments of the present application, by obtaining the response information of a candidate during an interview and using a preset language model, a plurality of response statements in the response information are converted into corresponding sentence vector information. Thus, based on the sentence vector information of each response statement, the sentence set vector information of the entire response information can be calculated, which facilitates the terminal device to quickly extract the important parts from the candidate's response information according to the sentence set vector information, reduces the amount of data for subsequent machine understanding, and improves the speed at which the terminal device gives necessary and reasonable follow-up questions. In actual interview applications, targeted processing is performed based on the important part content in the candidate's response information, making the evaluation of the candidate by the terminal device more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. 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 be obtained based on these drawings without creative efforts.

[0026] Figure 1 is a schematic flowchart of the steps of an intelligent interview method according to an embodiment of the present application;

[0027] Figure 2 is a schematic flowchart of the steps of another intelligent interview method according to an embodiment of the present application;

[0028] Figure 3 is a schematic diagram of an intelligent interview device according to an embodiment of the present application;

[0029] Figure 4 is a schematic diagram of a terminal device according to an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0030] In the following description, specific details such as specific system structures and technologies are proposed for the purpose of illustration rather than limitation, so as to thoroughly understand the embodiments of the present application. However, those skilled in the art should clearly understand that the present application can also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid unnecessary details from interfering with the description of the present application.

[0031] The technical solutions of the present application will be described below through specific embodiments.

[0032] Refer to Figure 1, which shows the schematic diagram of the step flow of an intelligent interview method according to an embodiment of the present application, may specifically include the following steps:

[0033] S101. Obtain the response information of the candidate during the interview, where the response information includes multiple response sentences;

[0034] It should be noted that this method can be applied to intelligent interviews, that is, through machines such as terminal devices to interact with the interview candidates and complete the interview evaluation of the candidates.

[0035] The terminal device in this embodiment can be an electronic device such as a mobile phone, a tablet computer, or a personal computer. The specific type of the terminal device is not limited in this embodiment.

[0036] In the embodiment of the present application, the response information of the candidate can be input into the terminal device in text form by the candidate through the input device of the terminal device during the interview; or, the terminal device can also interview the candidate in the form of voice interaction, collect the voice information of the candidate during the interview, and convert the voice information into text content to obtain the response information of the candidate. This embodiment does not limit this.

[0037] In the embodiment of the present application, the response information of the candidate can be the information for responding to a certain question proposed by the terminal device. At the beginning of the interview, the terminal device can require the candidate to introduce himself first. The content of the candidate's self-introduction is the response information that needs to be obtained, and the terminal device can process the response information to make the entire interview process proceed smoothly.

[0038] S102. Use a preset language model to convert the multiple response sentences into corresponding sentence vector information respectively;

[0039] Generally, the response information of the candidate may include multiple response sentences. After each response sentence is converted, the corresponding sentence vector information can be obtained, and this sentence vector information is also the sentence representation of the corresponding response sentence.

[0040] In the embodiment of the present application, a pre-trained language model can be used to perform vector conversion on each response sentence to obtain the corresponding sentence representation.

[0041] In specific implementation, a byte pair encoder (BPE) function can be preset in the language model, and the words in each response sentence are encoded in the way of byte pair encoding to obtain the word vectors of each word; then, the word vectors of each word in each sentence are added to the position vector corresponding to the position of the word to obtain the sentence vector information representing the sentence.

[0042] Alternatively, a language model based on a Natural Language Processing (NLP) transformer can also be used to transform each response statement. Generally, an NLP transformer includes an encoder and a decoder. For a given sequence of statements, through the processing of the encoder and decoder, it can be used to represent the sentence vector of the statement. This embodiment does not limit how to transform the response statement.

[0043] S103. Determine the sentence set vector information corresponding to the response information according to the sentence vector information of the multiple response statements;

[0044] Generally, the sentence vector information obtained from each response statement can only be used to represent the content of the current sentence, and each response of the candidate is often a detailed elaboration on a certain topic, and the concerns reflected in each sentence may be different.

[0045] Therefore, after obtaining the sentence vector information of each response statement respectively, the sentence set vector information of the entire response information can also be determined according to all the sentence vector information, and the content of the entire response information is represented by the sentence set vector information.

[0046] S104. Use the sentence set vector information to calculate the relationship probability between multiple entities included in the response information;

[0047] In the embodiment of the present application, the multiple entities included in the response information can be obtained by segmenting each response statement and determining the part-of-speech of each word after segmentation. For example, each word with a noun part-of-speech can be marked as an entity. These entities will be used as the basis for subsequent questioning of the candidate, that is, corresponding interview questions are generated through these entities.

[0048] Generally, what needs to be questioned of the candidate or what requires the candidate to conduct an in-depth analysis of a certain topic is often the content that the candidate mentions multiple times or focuses on during the interview process. Therefore, in this embodiment, it can be determined which are the contents that need to be focused on through the relationship probability between each entity.

[0049] In specific implementation, the relationship probability between each entity can be output based on the aforementioned language model.

[0050] S105. Extract target relationship information from the multiple entities according to the relationship probability;

[0051] In the embodiment of the present application, a group of entity pairs with the largest probability value can be extracted as the target relationship, or multiple groups of entity pairs with probability values exceeding a certain threshold can be extracted as the target relationship information, as the input statement of the terminal device.

[0052] Extracting some entities based on the magnitude of the relationship probability as the basis for questions in subsequent interviews can reduce the amount of data processed by the machine and improve the questioning efficiency of the terminal device during the interview process.

[0053] S106. Generate interview questions for the candidate based on the target relationship information.

[0054] After receiving the above-mentioned target relationship information, the terminal device can determine the next interview question according to this relationship and continue to interview the candidate.

[0055] In the embodiment of the present application, by obtaining the response information of the candidate during the interview and using a preset language model, multiple response sentences in the response information are converted into corresponding sentence vector information. Thus, based on the sentence vector information of each response sentence, the sentence set vector information of the entire response information can be calculated, which is convenient for the terminal device to quickly extract the important parts in the candidate's response information according to the sentence set vector information, reduce the amount of data for subsequent machine understanding, and improve the speed of the terminal device to give necessary and reasonable follow-up questions. In actual interview applications, targeted processing is performed according to the important parts of the candidate's response information, so that the evaluation of the candidate by the terminal device is more accurate.

[0056] Refer to Figure 2 , which shows the schematic flow chart of steps of another intelligent interview method according to an embodiment of the present application, and may specifically include the following steps:

[0057] S201. Obtain the response information of the candidate during the interview, where the response information includes multiple response sentences;

[0058] This method can be applied to intelligent interviews. By using a terminal device as an Artificial Intelligence (AI) interviewer to interact with the interview candidate, the interview evaluation of the candidate is completed, and the interview efficiency is improved.

[0059] In the embodiment of the present application, the candidate can directly communicate with the AI interviewer using voice. For example, the candidate uses a mobile phone, a tablet computer or a personal computer for the interview. By collecting the voice information during the interview, the AI interviewer can convert the voice information into text content, then understand the specific meaning contained in the text content based on natural language processing technology, and ask questions to the candidate on this basis.

[0060] S202. Identify multiple entities in the target response sentence, and generate a sequence of entities to be processed according to the multiple entities, where the target response sentence is any one of the multiple response sentences;

[0061] In the embodiments of the present application, each response of a candidate may contain multiple response sentences. The AI interviewer can process each response sentence separately to identify the key content in the candidate's entire response information.

[0062] In the embodiments of the present application, the AI interviewer can first identify the entities in each response sentence and generate a sequence of entities to be processed.

[0063] It should be noted that the entities in each response sentence can refer to each word with a noun part of speech in that response sentence.

[0064] In a specific implementation, each entity can correspond to a word with a word order number in a preset corpus. By arranging the corresponding word order numbers in the order of the entities in the response sentence, the entity sequence of the response sentence can be obtained.

[0065] S203. Input the sequence of entities to be processed into a preset language model to obtain the probability distribution of each entity in the target response sentence;

[0066] In the embodiments of the present application, a language model based on the NLP transformer decoder can be used to convert the candidate's response sentence into a corresponding sentence vector. The above language model based on the NLP transformer decoder can be a transformer decoder model with a masked multi-head self-attention mechanism based on a position feed-forward operation.

[0067] The NLP transformer decoder in this embodiment is different from the original NLP transformer that only decodes. Instead, it includes a masked multi-head self-attention mechanism based on a position feed-forward operation, and can repeatedly encode based on a given input representation on multiple NLP transformer blocks. Moreover, because there is no encoder block, the NLP transformer decoder does not contain any unmasked self-attention mechanisms.

[0068] In a specific implementation, the NLP transformer decoder can be generated by adopting the following encoding method:

[0069] h0 = TW e +W p

[0070]

[0071] where T is a matrix composed of one-hot vectors corresponding to sentences, W e is the token embedding matrix, W p is the position embedding matrix, L is the number of transformer blocks, and h l is the state of the l-th layer transformer block.

[0072] Since NLP transformers do not have the implicit concept of marked positions, the first layer of NLP transformers will add the position embeddings e p ∈R d to each token embedding at position p in the input sequence The architecture of self-attention allows the output state block l-1 to be represented by all input states h This is important for effectively modeling long-range dependencies. However, the NLP transformer decoder in this embodiment needs to restrict self-attention at the same time, so that the model only needs to focus on the context of the current token, rather than the context above the current token. The above tokens are the entities in each response statement.

[0073] S204. Generate the sentence vector information of the target response statement according to the probability distribution of each entity in the target response statement;

[0074] In the embodiment of the present application, based on the above NLP transformer decoder model, for a given entity sequence the objective function can be set to maximize the log-likelihood function:

[0075]

[0076] where k is the context window considered for predicting the next token c i through the conditional probability P.

[0077] Using the above NLP transformer decoder model, the probability distribution of each entity can be calculated as follows:

[0078]

[0079] where h L is the state sequence after the last layer L of the transformer, W e is the embedding matrix, and θ is the model parameter optimized by stochastic gradient descent.

[0080] By outputting the probability distribution of each entity, the sentence vector information corresponding to each response statement can be formed, that is, the sentence representation.

[0081] S205. Determine the weight value of the sentence vector information of each response statement, and perform weighted summation on the sentence vector information of each response statement according to the weight value to obtain the sentence set vector information corresponding to the response information;

[0082] In the embodiment of the present application, the sentence set representation of the entire response information can be obtained by summarizing the sentence representations of each response statement, that is, the sentence set vector information corresponding to the candidate response information.

[0083] In a specific implementation, in order to more clearly express the relationship between each reply statement and the entire reply information, the ratio of the sentence vector information of each reply statement to the sum of the sentence vector information of all reply statements can be calculated respectively, and this ratio can be used as the weight value of the sentence vector information of the corresponding reply statement.

[0084] For example, the weight value of the sentence vector information of each reply statement can be expressed by the following formula:

[0085]

[0086] where α i is the weight value of the sentence vector information of the i-th reply statement, and exp(s i r) is the sentence vector information of the i-th reply statement.

[0087] In the embodiments of the present application, in order to be more effective when using the language model generated based on the above NLP transformer to implement multi-instance learning of the remote supervision data set, the structure of the NLP transformer can also be extended. That is, the model can be pre-trained with equation (1) as the target first, and then the language model can be fine-tuned for the relation extraction task.

[0088] For example, let a labeled data set where each example consists of a label x i =[x 1 ,…,x m , head i and tail i are the positions of the two entities in the relation in the labeled entity sequence, and r i is the relation label corresponding to the distant supervision. Due to the noisy annotations, the response variable trained with the label r i is unreliable. Instead, apply relation classification to the text at the bag level, representing each entity pair as a set where is composed of the entity pairs of all sentences.

[0089] By feeding the entity sequence of the sentence into the model, the previous state L of the final state representation h is used to represent the sentence vector information s i of the reply statement.

[0090] Then, by performing weighted summation on the sentence vector information of each reply statement, the sentence set vector information corresponding to the reply information can be obtained:

[0091]

[0092] S206. Perform a linear transformation and a logistic regression softmax transformation on the sentence set vector information in sequence to obtain the relationship probabilities between multiple entities included in the reply information.

[0093] In an embodiment of the present application, the sentence set vector information used to represent the entire reply information can be linearly transformed and then subjected to a softmax transformation to obtain the output distribution P(l) on the relationship labels:

[0094] P(l|S,θ) = softmax(W r s + b)

[0095] where W r is the representation matrix of the relationship r, and b is the bias vector.

[0096] The above output distribution is the relationship probability between each entity in the candidate's answer content.

[0097] In an embodiment of the present application, in order to improve the accuracy of the distribution probability of the model output, the parameters in the output distribution can also be fine-tuned.

[0098] Specifically, the goal of fine-tuning is to maximize the following likelihood function:

[0099]

[0100] During fine-tuning, the language model can be used as an auxiliary target to improve the generality and convergence speed of the model. Therefore, combining the above formulas (1) and (2), the final objective function can be obtained:

[0101]

[0102] where the scalar value λ is the weight of the language model objective function during the fine-tuning process.

[0103] S207. Extract one or more entity pairs whose relationship probabilities exceed a preset threshold as target relationship information.

[0104] In an embodiment of the present application, a group of entity pairs with the highest probability can be extracted as the target relationship, or multiple groups of entity pairs whose probabilities exceed a certain threshold can be extracted as the target relationship, as the input sentence of the AI interviewer. After receiving the above target relationship information, the AI interviewer can determine the next interview question according to this relationship and continue to interview the candidate.

[0105] As a specific example, if the interviewer asks: "We have a new product, and you can discuss and decide whether to launch this new product."

[0106] Candidate's answer: "I think the clinical trial time for the new product is not long enough. If it is directly put into use, it will bring certain risks to users, and we may not be able to solve these problems. Therefore, I think it should be put on sale after the clinical trial results are sufficient."

[0107] By encoding the candidate's answer and obtaining the corresponding hidden state h L , then weight each sentence to obtain the representation of the whole text, and then output the probability of each entity. For example, <new product, risk> (0.7), <new product, problem> (0.2),.... Then, the relationship with the highest probability can be selected, that is, <new product, risk> as the target relationship information, as the basis for asking questions in the further interview.

[0108] S208. Generate an interview question for the candidate based on the target relationship information;

[0109] S209. Generate an interview evaluation report for the candidate according to the candidate's reply information to multiple interview questions; upload the interview evaluation report to the blockchain.

[0110] In the embodiment of the present application, according to the candidate's reply to each question of the AI interviewer, an interview evaluation report for the candidate can be generated, and the above report can be uploaded to the blockchain to ensure its security and the transparency and fairness of the evaluation result.

[0111] In the subsequent process, for example, when it is necessary to trace the information of a certain candidate, the evaluation report can be downloaded from the blockchain through the user device to check whether the report has been tampered with.

[0112] It should be noted that the blockchain mentioned in this embodiment is a new application mode of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, and encryption algorithm. The blockchain is essentially a decentralized database, a string of data blocks generated by using cryptographic methods, and each data block contains information about a batch of network transactions, which is used to verify the validity of the information (anti-counterfeiting) and generate the next block. The blockchain can include the blockchain underlying platform, the platform product service layer, and the application service layer, etc.

[0113] In the embodiments of the present application, during the intelligent interview process, by using a language model based on remotely supervised NLP transformers for relation extraction, the AI interviewer can quickly extract the important content of the candidate's answer. Since the amount of data used for machine understanding after extraction is reduced, the AI interviewer can quickly give necessary and reasonable follow-up questions. In practical applications, because the judgment is more accurate, the response speed of the hardware is also improved, which not only saves hardware space, but also improves the running speed and the interview experience of the candidate.

[0114] Secondly, by extracting the candidate's answer content, it is possible to follow up on the places where the relationship reflected in the candidate's answer is inconsistent with the actual information, so as to better understand the candidate's answer, which helps to provide a more effective basis for selecting candidates, and avoids the possibility that the interviewer makes a wrong judgment on the candidate's performance due to factors such as the candidate's appearance and the lack of energy caused by a long interview, ensuring the accuracy and reliability of the interview results.

[0115] It should be noted that the magnitudes of the sequence numbers of the steps in the above embodiments do not mean the order of execution. The execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0116] Refer to Figure 3 , which shows a schematic diagram of an intelligent interview device according to an embodiment of the present application, and specifically may include the following modules:

[0117] An acquisition module 301, configured to acquire the reply information of the candidate during the interview process, where the reply information includes a plurality of reply statements;

[0118] A conversion module 302, configured to respectively convert the plurality of reply statements into corresponding sentence vector information by using a preset language model;

[0119] A determination module 303, configured to determine the sentence set vector information corresponding to the reply information according to the sentence vector information of the plurality of reply statements;

[0120] A calculation module 304, configured to calculate the relationship probability between a plurality of entities included in the reply information by using the sentence set vector information;

[0121] An extraction module 305, configured to extract target relationship information from the plurality of entities according to the relationship probability;

[0122] A generation module 306, configured to generate an interview question for the candidate based on the target relationship information.

[0123] In the embodiments of the present application, the conversion module 302 may specifically include the following sub-modules:

[0124] An entity sequence generation sub-module, configured to identify multiple entities in a target reply statement, and generate a to-be-processed entity sequence according to the multiple entities, where the target reply statement is any one of the multiple reply statements;

[0125] A probability distribution calculation sub-module, configured to input the to-be-processed entity sequence into a preset language model, and obtain the probability distribution of each entity in the target reply statement, where the language model is a transformer decoder based on a masked multi-head self-attention mechanism with position-wise feed-forward operations;

[0126] A sentence vector information generation sub-module, configured to generate sentence vector information of the target reply statement according to the probability distribution of each entity in the target reply statement.

[0127] In an embodiment of the present application, the transformer decoder based on a masked multi-head self-attention mechanism with position-wise feed-forward operations is generated by adopting the following encoding method:

[0128] h0 = TW e +W p

[0129]

[0130] where T is a matrix composed of one-hot vectors corresponding to sentences, W e is a token embedding matrix, W p is a position embedding matrix, L is the number of transformer blocks, and h l is the state of the l-th layer transformer block.

[0131] In an embodiment of the present application, the determining module 303 may specifically include the following sub-modules:

[0132] A weight value determining sub-module, configured to determine the weight value of the sentence vector information of each reply statement;

[0133] A sentence set vector information generation sub-module, configured to perform weighted summation on the sentence vector information of each reply statement according to the weight value, so as to obtain the sentence set vector information corresponding to the reply information.

[0134] In an embodiment of the present application, the weight value determining sub-module may specifically include the following units:

[0135] A weight value calculation unit, configured to calculate the ratio of the sentence vector information of each reply statement to the sum of the sentence vector information of all reply statements respectively, and use the ratio as the weight value of the sentence vector information of the corresponding reply statement.

[0136] In an embodiment of the present application, the calculation module 304 may specifically include the following sub-modules:

[0137] A relational probability calculation sub-module, which is used to perform a linear transformation and a logistic regression softmax transformation on the vector information of the statement set in sequence, so as to obtain the relational probabilities between multiple entities included in the reply information.

[0138] In the embodiment of the present application, the extraction module 305 may specifically include the following sub-modules:

[0139] A target relational information extraction sub-module, which is used to extract one or more entity pairs whose relational probabilities exceed a preset threshold as target relational information.

[0140] In the embodiment of the present application, the device may further include the following modules:

[0141] An interview evaluation report generation module, which is used to generate an interview evaluation report for the candidate according to the reply information of the candidate to multiple interview questions;

[0142] An interview evaluation report uploading module, which is used to upload the interview evaluation report to the blockchain.

[0143] For the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple. For related parts, refer to the description in the method embodiment section.

[0144] Refer to Figure 4 , which shows a schematic diagram of a terminal device according to an embodiment of the present application. As Figure 4 shown, the terminal device 400 in this embodiment includes: a processor 410, a memory 420, and a computer program 421 stored in the memory 420 and executable on the processor 410. When the processor 410 executes the computer program 421, it implements the steps in each embodiment of the above intelligent interview method, such as Figure 1 the steps S101 to S106 shown. Or, when the processor 410 executes the computer program 421, it implements the functions of each module / unit in each device embodiment above, such as Figure 3 the functions of the modules 301 to 306 shown.

[0145] Exemplarily, the computer program 421 may be divided into one or more modules / units, which are stored in the memory 420 and executed by the processor 410 to complete this application. The one or more modules / units may be a series of computer program instruction segments capable of performing specific functions, and these instruction segments may be used to describe the execution process of the computer program 421 in the terminal device 400. For example, the computer program 421 may be divided into an acquisition module, a conversion module, a determination module, a calculation module, an extraction module, and a generation module. The specific functions of each module are as follows:

[0146] The acquisition module is used to acquire the response information of the candidate during the interview process, and the response information includes a plurality of response statements;

[0147] The conversion module is used to respectively convert the plurality of response statements into corresponding sentence vector information by using a preset language model;

[0148] The determination module is used to determine the sentence set vector information corresponding to the response information according to the sentence vector information of the plurality of response statements;

[0149] The calculation module is used to calculate the relationship probability between a plurality of entities included in the response information by using the sentence set vector information;

[0150] The extraction module is used to extract target relationship information from the plurality of entities according to the relationship probability;

[0151] The generation module is used to generate an interview question for the candidate based on the target relationship information.

[0152] The terminal device 400 may be a computing device such as a desktop computer, a notebook, or a palm computer. The terminal device 400 may include, but is not limited to, a processor 410 and a memory 420. Those skilled in the art can understand that Figure 4 This is only an example of the terminal device 400 and does not constitute a limitation on the terminal device 400. It may include more or fewer components than shown in the figure, or combine certain components, or different components. For example, the terminal device 400 may further include input / output devices, network access devices, a bus, etc.

[0153] The processor 410 may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0154] The memory 420 may be an internal storage unit of the terminal device 400, such as the hard disk or memory of the terminal device 400. The memory 420 may also be an external storage device of the terminal device 400, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, etc. equipped on the terminal device 400. Further, the memory 420 may also include both the internal storage unit and the external storage device of the terminal device 400. The memory 420 is used to store the computer program 421 and other programs and data required by the terminal device 400. The memory 420 may also be used to temporarily store data that has been output or is to be output.

[0155] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included within the protection scope of the present application.

Claims

1. An intelligent interview method, characterized in that, Including: Obtain the response information of the candidate during the interview process, where the response information includes multiple response statements; Adopt a preset language model to respectively convert the multiple response statements into corresponding sentence vector information. The language model is a transformer decoder with masked multi-head self-attention mechanism based on position-wise feed-forward operation, and the language model does not contain any unmasked self-attention mechanism; According to the sentence vector information of the multiple response statements, determine the sentence set vector information corresponding to the response information; Adopt the sentence set vector information to calculate the relationship probabilities among multiple entities included in the response information; According to the relationship probabilities, extract the target relationship information from the multiple entities; Based on the target relationship information, generate interview questions for the candidate; Wherein, the transformer decoder with masked multi-head self-attention mechanism based on position-wise feed-forward operation is generated by adopting the following encoding method: h0 = TW e +W p Among them, T is a matrix composed of one-hot vectors corresponding to sentences, and W e is the token embedding matrix, and W p is the position embedding matrix, L is the number of transformer blocks, and h l is the state of the l-th layer of the transformer block.

2. The method according to claim 1, characterized in that, The step of adopting a preset language model to respectively convert the multiple response statements into corresponding sentence vector information includes: Identify multiple entities in the target response statement, and generate an entity sequence to be processed according to the multiple entities. The target response statement is any one of the multiple response statements; Input the entity sequence to be processed into a preset language model to obtain the probability distribution of each entity in the target response statement; According to the probability distribution of each entity in the target response statement, generate the sentence vector information of the target response statement.

3. The method according to claim 1 or 2, characterized in that, The step of determining the sentence set vector information corresponding to the response information according to the sentence vector information of the multiple response statements includes: Determine the weight value of the sentence vector information of each response statement; According to the weight value, perform weighted summation on the sentence vector information of each response statement to obtain the sentence set vector information corresponding to the response information.

4. The method according to claim 3, characterized in that, The step of determining the weight value of the sentence vector information of each response statement includes: Respectively calculate the ratio of the sentence vector information of each response statement to the sum of the sentence vector information of all response statements, and use the ratio as the weight value of the sentence vector information of the corresponding response statement.

5. The method according to any one of claims 1 or 2 or 4, characterized in that, The step of adopting the sentence set vector information to calculate the relationship probabilities among multiple entities included in the response information includes: Perform linear transformation and logistic regression softmax transformation on the sentence set vector information in sequence to obtain the relationship probabilities among multiple entities included in the response information.

6. The method according to claim 5, characterized in that, The step of extracting the target relationship information from the multiple entities according to the relationship probabilities includes: Extract one or more entity pairs whose relationship probabilities exceed a preset threshold as the target relationship information.

7. The method according to any one of claims 1 or 2 or 4 or 6, characterized in that, It also includes: Generate an interview evaluation report for the candidate according to the candidate's response information to multiple interview questions; Upload the interview evaluation report to the blockchain.

8. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the intelligent interview method according to any one of claims 1 to 7.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the intelligent interview method according to any one of claims 1 to 7.

Images