Artificial intelligence-based interview content refining method and device, equipment and medium

Abstract

The application discloses an interview content refining method and device based on artificial intelligence, equipment and medium, the method comprises the following steps: obtaining an interview recording, and converting the interview recording into a self-introduction text and an interview answer text; performing text analysis on the self-introduction text to obtain interviewer basic information; performing sentence classification on the interview answer text to obtain classified texts; extracting sentences from each type of classified text through a language extraction model to obtain extracted sentences; and refining the extracted sentences by using a Transformer model to obtain interview refined corpus, so that the core content can be accurately refined from the interview record content with a large amount of data, the accuracy of content refinement is improved, the accuracy of intelligent interview evaluation is improved, the interviewer basic information and the interview refined corpus are stored in a blockchain, and are simultaneously sent to a management end for evaluation, so that the evaluation result can meet the requirements, and the accuracy and efficiency of intelligent interview result evaluation are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence, and in particular to an interview content refining method and device based on artificial intelligence, equipment and medium. BACKGROUND

[0002] In the recruitment season of large enterprises, there are often many interviewees participating in the interview. At present, most employers and interviewees conduct interviews through on-site or video conferencing. The employer often evaluates the interviewee after the interview based on the interviewee's interview answers. The usual artificial interview at least stores the following questions: (1) different interviewers have different preferred angles of questioning, and the same interviewer may have different judgments due to different job experience, interview skills and emotional state; (2) high labor cost and interview time cost. In view of this, some enterprises use artificial intelligence-based interview robots to conduct interviews and provide the obtained interview content to decision makers for result evaluation, which is beneficial to improving the fairness of the interview, but at the same time, it also leads to a new problem. When there are many interviewees, the obtained interview content will also be more, which also increases the time cost of decision evaluation, resulting in low efficiency of intelligent interview.

[0003] The existing solution mainly matches the interview content with keywords to obtain key sentences, or uses a natural language processing (NLP) model to perform semantic recognition. When using the keyword matching method, the interviewee may answer questions in different ways during the answering process, which may not match the preset keywords, resulting in low final interview estimation accuracy. When using a general natural language processing model for semantic recognition, the semantic recognition accuracy often does not meet the requirements. SUMMARY

[0004] The embodiments of the present application provide an interview content refining method, device and medium based on artificial intelligence to improve the accuracy of interview content evaluation in intelligent interviews.

[0005] To solve the above technical problems, the present application provides an interview content refining method based on artificial intelligence, comprising:

[0006] obtaining an interview recording and converting the interview recording into an interview text, wherein the interview text includes a self-introduction text and an interview answer text;

[0007] performing text analysis on the self-introduction text to obtain basic information of the interviewee;

[0008] classifying the interview answer text according to the interview angle involved to obtain a classified text;

[0009] extracting a sentence from each of the classified texts through a language extraction model to obtain an extracted sentence, and refining the extracted sentence by using a Transformer model to obtain an interview refined corpus;

[0010] sending the interviewee basic information and the interview refined corpus to a management end, so that the management end determines an interview result according to the interviewee basic information and the interview refined corpus.

[0011] Optionally, the converting the interview recording into an interview text comprises:

[0012] recognizing a question and answer start identifier contained in the interview recording;

[0013] performing text conversion on the interview recording by using a voice-to-text conversion mode, and converting the recording content before the question and answer start identifier into a text as a self-introduction text and converting the recording content before the question and answer start identifier into a text as an interview response text.

[0014] Optionally, the classifying the interview response text according to the interview angles involved to obtain classified texts comprises:

[0015] taking each sentence in the interview response text as a basic sentence, and performing word segmentation processing on the basic sentence by using a preset word segmentation mode to obtain basic word segmentation;

[0016] converting the basic word segmentation into a word vector, and performing clustering on the word vector by using a clustering algorithm to obtain a clustering center corresponding to the basic sentence;

[0017] for each basic sentence, calculating the Euclidean distance between the clustering center corresponding to the basic sentence and the word vector corresponding to each preset interview angle, and taking the preset interview angle with the smallest distance as a target classification of the basic sentence, and taking the basic sentence as the classified text corresponding to the target classification.

[0018] Optionally, the performing word segmentation processing on the basic sentence by using a preset word segmentation mode to obtain basic word segmentation comprises:

[0019] performing word segmentation on the basic sentence by using a conditional random field model to obtain initial word segmentation;

[0020] obtaining the word frequency of each initial word segmentation from historical interview response texts;

[0021] generating the weight of the initial word segmentation based on the word frequency of the initial word segmentation, and taking the initial word segmentation labeled with the weight as the basic word segmentation.

[0022] Optionally, the language extraction model is a bidirectional long short-term memory network model, the bidirectional long short-term memory network model comprises a sentence encoder and a document encoder, the extracting sentences from each category of the classified texts by the language extraction model comprises:

[0023] The text in the classified text is split by the sentence encoder according to characters to obtain basic characters;

[0024] The basic characters are encoded to obtain the encoding content corresponding to the basic characters;

[0025] The encoding content is input to an initialized weight character encoding layer, each encoding is mapped to a character vector by the character encoding layer, and each character vector is taken as a sentence encoding result;

[0026] The sentence encoding result is spliced into a hidden layer vector in a forward and reverse hidden layer output, and the hidden layer vector is input to the document encoder;

[0027] The hidden layer vector is weighted by the document encoder to obtain a document feature vector, and the document feature vector is decoded, and the output result obtained by decoding is taken as the extracted sentence.

[0028] Optionally, the weighting of the hidden layer vector by the document encoder to obtain the document feature vector comprises:

[0029] The document feature vector is determined by the following formula:

[0030]

[0031] Wherein, C i is the i-th document feature vector, j is the serial number of the embedding code, n is the number of embedding codes, b ij is the weight of the i-th document feature vector for the j-th hidden layer vector, h j is the j-th hidden layer vector, wherein the embedding code is generated based on the hidden state of the bidirectional long short-term memory network model.

[0032] Optionally, after the extracted sentences are refined by the Transformer model to obtain interview refined corpus, the interview content refinement method based on artificial intelligence further comprises: storing the interviewee basic information and the interview refined corpus into a blockchain network.

[0033] To solve the above technical problems, the embodiment of the present application also provides an interview content refinement device based on artificial intelligence, comprising:

[0034] The text acquisition module is configured to acquire an interview audio and convert the interview audio into an interview text, wherein the interview text comprises a self-introduction text and an interview response text.

[0035] The text analysis module is configured to analyze the self-introduction text to obtain basic information of an interviewee.

[0036] The text classification module is configured to classify sentences of the interview response text according to an interview angle involved to obtain classified texts.

[0037] The corpus extraction module is configured to extract sentences from each of the classified texts by using a language extraction model to obtain extracted sentences, and refine the extracted sentences by using a Transformer model to obtain interview refined corpora.

[0038] The information sending module is configured to send the basic information of the interviewee and the interview refined corpora to a management terminal, so that the management terminal determines an interview result according to the basic information of the interviewee and the interview refined corpora.

[0039] Optionally, the text acquisition module comprises:

[0040] The identifier identification unit is configured to identify a question-and-answer start identifier contained in the interview audio.

[0041] The text determination unit is configured to convert the interview audio into text by using a voice-to-text manner, and convert the text obtained by converting the audio content before the question-and-answer start identifier into the self-introduction text and the text obtained by converting the audio content before the question-and-answer start identifier into the interview response text.

[0042] Optionally, the text classification module comprises:

[0043] The word segmentation unit is configured to take each sentence in the interview response text as a basic sentence, and perform word segmentation processing on the basic sentence by using a preset word segmentation manner to obtain basic word segmentation.

[0044] The clustering unit is configured to convert the basic word segmentation into a word vector, and perform clustering on the word vector by using a clustering algorithm to obtain a clustering center corresponding to the basic sentence.

[0045] The classification unit is configured to calculate, for each basic sentence, an Euclidean distance between the clustering center corresponding to the basic sentence and a word vector corresponding to each preset interview angle, and take a preset interview angle with the smallest distance as a target classification of the basic sentence, and take the basic sentence as a classified text corresponding to the target classification.

[0046] Optionally, the word segmentation unit comprises:

[0047] An initial word segmentation unit is configured to perform word segmentation on the basic sentence by using a conditional random field model to obtain initial word segmentation results.

[0048] A word frequency obtaining sub-unit is configured to obtain a word frequency of each initial word segmentation result from historical interview answer texts.

[0049] A word segmentation weighting unit is configured to generate a weight of each initial word segmentation result based on the word frequency of the initial word segmentation result, and take the initial word segmentation result with the weight as the basic word segmentation result.

[0050] Optionally, the corpus extraction module comprises:

[0051] A splitting unit is configured to split the text in the classification text into basic characters according to characters by using the sentence encoder.

[0052] An encoding unit is configured to encode the basic characters to obtain encoding content corresponding to the basic characters.

[0053] A mapping unit is configured to input the encoding content into a character encoding layer with an initial weight, map each encoding into a character vector by using the character encoding layer, and take each character vector as a sentence encoding result.

[0054] A concatenating unit is configured to concatenate the sentence encoding result in a forward hidden layer output and a reverse hidden layer output into a hidden layer vector, and input the hidden layer vector into the document encoder.

[0055] A weighting unit is configured to weight the hidden layer vector by using the document encoder to obtain a document feature vector, and decode the document feature vector to obtain an output result as the extracted sentence.

[0056] Optionally, the weighting decoding unit comprises:

[0057] A calculating sub-unit is configured to determine the document feature vector by using the following formula:

[0058]

[0059] wherein, C i is the i-th document feature vector, j is a serial number of an embedding code, n is a number of embedding codes, b ij is a weight of the i-th document feature vector for the j-th hidden layer vector, h j is the j-th hidden layer vector, and the embedding code is generated based on a hidden state of the bidirectional long short-term memory network model.

[0060] Optionally, the interview content refining device based on artificial intelligence further comprises:

[0061] a storage module, configured to store the interviewee basic information and the interview refined corpus into a blockchain network.

[0062] To solve the above technical problems, the embodiment of the present application also provides a computer device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above interview content refining method based on artificial intelligence when executing the computer program.

[0063] To solve the above technical problems, the embodiment of the present application also provides a computer readable storage medium, which stores a computer program, wherein the computer program is executed by a processor to implement the steps of the above interview content refining method based on artificial intelligence.

[0064] The interview content refining method, device, equipment and medium based on artificial intelligence provided by the embodiment of the present application, by obtaining an interview recording and converting the interview recording into an interview text, wherein the interview text comprises a self-introduction text and an interview answer text, performing text analysis on the self-introduction text to obtain interviewee basic information, classifying sentences of the interview answer text according to interview angles involved to obtain classified texts, extracting sentences from each type of classified text by a language extraction model to obtain extracted sentences, and refining the extracted sentences by a Transformer model to obtain interview refined corpus, the core content can be accurately refined from the interview record content with large data volume, the accuracy of content refinement is improved, which is conducive to improving the accuracy of intelligent interview evaluation, and finally the interviewee basic information and the interview refined corpus are sent to a management end, so that the management end determines an interview result according to the interviewee basic information and the interview refined corpus, avoiding inaccurate evaluation results caused by direct semantic recognition, and being conducive to improving the accuracy and efficiency of intelligent interview result evaluation. BRIEF DESCRIPTION OF DRAWINGS

[0065] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be 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, and other drawings can also be obtained by those skilled in the art without creative labor.

[0066] Figure 1 is an exemplary system architecture diagram to which the present application can be applied;

[0067] Figure 2 is a flowchart of one embodiment of the interview content refining method based on artificial intelligence of the present application;

[0068] Figure 3This is a schematic diagram of a structure of an embodiment of the AI-based interview content refinement device according to this application;

[0069] Figure 4 This is a schematic diagram of the structure of one embodiment of the computer device according to this application. Detailed Implementation

[0070] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0071] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0072] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] Please see Figure 1 ,like Figure 1 As shown, system architecture 100 may include terminal devices 101, 102, and 103, a network 104, and a server 105. Network 104 serves as the medium for providing communication links between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0074] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc.

[0075] The terminal devices 101, 102, and 103 can be various electronic devices with a display screen and supporting web browsing, including but not limited to a smart phone, a tablet computer, an e-book reader, an MP3 (Moving Picture Experts Group Audio Layer III) player, an MP4 (Moving Picture Experts Group Audio Layer IV) player, a laptop computer, a desktop computer, and the like.

[0076] The server 105 can be a server providing various services, for example, a background server providing support for a page displayed on the terminal devices 101, 102, and 103.

[0077] It should be noted that the method for refining interview content based on artificial intelligence provided in the embodiments of the present application is executed by a server, and accordingly, the device for refining interview content based on artificial intelligence is arranged in the server.

[0078] It should be understood that, Figure 1 The number of terminal devices, networks, and servers in the system is merely illustrative. According to the implementation needs, there can be any number of terminal devices, networks, and servers, and the terminal devices 101, 102, and 103 in the embodiments of the present application can specifically correspond to an application system in actual production.

[0079] Please refer to Figure 2 , Figure 2 A method for refining interview content based on artificial intelligence is shown, which is taken as an example of the server in Figure 1 for illustration, and is described in detail as follows:

[0080] S201: Obtain an interview recording and convert the interview recording into an interview text, wherein the interview text includes a self-introduction text and an interview answer text.

[0081] Specifically, in the process of interview recruitment in an enterprise, a plurality of interviewees participate in the interview. Since the number of posts for interview is limited, there are cases where multiple interviewees interview for the same post. To avoid confusion or forgetfulness of the information of the interviewees, the embodiments of the present application record the interview process of the multiple interviewees in the interview process, convert the recording content into an interview text after the interview, and perform subsequent processing. The interview text includes a self-introduction text and an interview answer text.

[0082] The self-introduction text refers to a text converted from the voice of the self-introduction of the interviewee, and the answer text refers to a text of the answer of the interviewee after the self-introduction.

[0083] It needs to be explained that the interviewer mentioned in the embodiment can be a person or a question and answer robot participating in the intelligent interview, which is not limited here.

[0084] It should be understood that the general interview time is 30-40 minutes or even longer, so the content of the interviewee's answer is relatively large. In view of this situation, the embodiment takes the self-introduction as the starting point, because the information in the self-introduction part can summarize the ability of the interviewee in a large part, and the other links of the interview, such as the skill examination and the business acuteness examination, can be used as a reference for the training data to supplement and verify the self-introduction of the interviewee, and obtain a more comprehensive result.

[0085] In the embodiment, the interview recording is converted into an interview text, which can use a tool supporting voice conversion text or a voice conversion text algorithm, which is not limited here. The specific implementation process of dividing the interview text into self-introduction text and interview answer text can refer to the description of the subsequent embodiments, and will not be repeated here to avoid repetition.

[0086] S202: Text analysis is performed on the self-introduction text to obtain the basic information of the interviewee.

[0087] Specifically, since the self-introduction text generally includes personal basic information, experience information, areas of expertise and skills, past honors and self-evaluation, etc., the content modules involved are relatively similar, in order to improve efficiency, the embodiment adopts a text analysis method based on regular expressions to analyze the self-introduction text and quickly extract the content in the self-introduction text to obtain the basic information of the interviewee.

[0088] Among them, the basic information of the interviewee includes but is not limited to: name, household registration, school, major, work experience, etc. Personal fixed information, and personal professional information such as honors obtained, enterprises served, work experience and skills mastered, etc.

[0089] It should be noted that since the content dimensions in the self-introduction text are similar, the basic information of the interviewee to be obtained is divided into multiple dimensions, at least one regular expression is set for each dimension to match and analyze the self-introduction text, and the content corresponding to the dimension is obtained as the analysis content of the dimension.

[0090] Among them, the regular expression (regular expression) describes a pattern of string matching, which can be used to check whether a string contains a certain substring, replace the matching substring, or take out the substring that meets a certain condition from a certain string, etc.

[0091] For example, in a specific embodiment, text analysis is performed from the seven dimensions of name, household registration, school, major, work experience, work experience, and skills mastered. For the household registration dimension, keywords containing specific characters can be set for matching, such as matching sentence patterns containing "I am a person from XXX", "I come from XXX", "I am a person from XXX", "I grew up in XXX", etc.

[0092] S203: Classify the interview response text according to the interview angle involved, and obtain classified text.

[0093] Specifically, during the process of the interviewer asking questions, questions are usually asked about work experience, areas of expertise, and skills. In this embodiment, these interview angles are pre-set according to actual needs. After obtaining the interview response text, the interview response text is classified according to the interview angle involved, and classified text is obtained. This allows subsequent extraction and refinement of key sentences according to the category of the classified text, which is beneficial to improving the accuracy of content refinement.

[0094] The interview angle involved refers to the focus of the question and answer, such as salary requirements, awards received, work experience, professional skills, etc.

[0095] Further, the specific implementation process of classifying the interview response text according to the interview angle involved and classifying the text according to the semantic recognition result in this embodiment can be referred to the description of the subsequent embodiments. To avoid repetition, it will not be repeated here.

[0096] Specifically, the classification of the text according to the semantic recognition result can be clustering the recognition result to obtain a clustering result, and calculating the Euclidean distance between the clustering result and the word vector corresponding to each interview angle, and then the interview angle with the closest distance is taken as the interview angle corresponding to the clustering result.

[0097] S204: Extract sentences from each type of classified text through a language extraction model to obtain extracted sentences, and refine the extracted sentences using a Transformer model to obtain interview refined corpus.

[0098] Specifically, sentences are extracted from each type of classified text through a language extraction model to obtain extracted sentences, and the extracted sentences are refined using a Transformer model to obtain interview refined corpus.

[0099] The language extraction model includes but is not limited to: deep semantic representation (Embedding from Language Model, ELMo) algorithm, OpenAI GPT and pre-training bidirectional encoder semantic (Bidirectional Encoder Representations from Transformers, BERT) model.

[0100] Preferably, in the present embodiment, the improved OpenAI GPT model is used as the semantic extraction model, and the implementation process of the specific sentence extraction can refer to the description of the subsequent embodiments. To avoid repetition, it will not be repeated here.

[0101] It should be noted that the extracted sentences obtained in the present embodiment can also be in the form of vectors for subsequent rapid input into the Transformer model for fine extraction.

[0102] The Transformer model can quickly extract sentences with higher importance according to the weight through the attention mechanism.

[0103] It should be noted that in the decoding stage of the Transformer model, the sum of the generated document feature vectors is input into the decoder. This autoregressive long short-term time network will predict the next sentence to be extracted, and the output result will be connected to the input in the next sentence decoding. The biggest difference between the decoder used in the Transformer model and other commonly used decoders is that in the attention acquisition process through dot product, if the same index appears twice in succession, the entire extraction process is ended, avoiding information redundancy caused by multiple extraction of similar information.

[0104] It should be understood that in the present embodiment, steps S203 to S204 and step S202 do not have a certain logical sequence, and they can also be executed in parallel, which is not limited here.

[0105] S205: Send the interviewee's basic information and interview refined corpus to the management end, so that the management end determines the interview result according to the interviewee's basic information and interview refined corpus.

[0106] Specifically, the extracted interviewee's basic information and interview refined corpus are sent to the management end to ensure the accuracy and refinement of the extracted content, so that the subsequent users of the management end can accurately and quickly determine the evaluation result according to the extracted content, which is beneficial to improve the accuracy and efficiency of intelligent interview.

[0107] In the embodiment, the interview recording is obtained, and the interview recording is converted into an interview text, wherein the interview text includes a self-introduction text and an interview answer text, the self-introduction text is subjected to text analysis to obtain interviewer basic information, the interview answer text is subjected to sentence classification according to an interview angle involved to obtain classified texts, the language extraction model is used to extract sentences from each type of classified text to obtain extracted sentences, and the Transformer model is used to refine the extracted sentences to obtain interview refined corpus, so that the core content is accurately refined from the interview record content with a large amount of data, the accuracy of content refinement is improved, the accuracy of intelligent interview evaluation is improved, and finally the interviewer basic information and the interview refined corpus are sent to a management end, so that the management end determines an interview result according to the interviewer basic information and the interview refined corpus, the inaccurate evaluation result caused by direct semantic recognition is avoided, and the accuracy and efficiency of intelligent interview result evaluation are improved.

[0108] In an embodiment, the obtained interviewer basic information and interview refined corpus can be stored on a blockchain network, so that the data information can be shared between different platforms through blockchain storage, and the data can also be prevented from being tampered with.

[0109] The blockchain is a new application mode of distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm and other computer technologies. The blockchain is essentially a decentralized database, and is a series of data blocks associated using cryptographic methods. Each data block contains information about a batch of network transactions, and is used to verify the validity (anti-fake) of the information and generate the next block. The blockchain can include a blockchain underlying platform, a platform product service layer and an application service layer.

[0110] In some optional implementation manners of the embodiment, in step S201, converting the interview recording into the interview text includes:

[0111] identifying a question and answer start identifier included in the interview recording;

[0112] text conversion is performed on the interview recording by using a voice-to-text conversion mode, and the text obtained by converting the recording content before the question and answer start identifier is taken as the self-introduction text, and the text obtained by converting the recording content before the question and answer start identifier is taken as the interview answer text.

[0113] Specifically, before the voice-to-text conversion, the interview recording file is traversed to find a voice segment with the same voice information as a preset question and answer start identifier as a demarcation point, the text obtained by converting the voice before the voice segment is taken as the self-introduction text, and the text obtained by converting the voice after the voice segment is taken as the interview answer text.

[0114] The voice segment with the same voice information as the preset question and answer start identifier is found, specifically, amplitude normalization processing, pre-emphasis processing and frame windowing are performed on the voice signal to obtain a voice frame set, and then the same voice frame segment as the voice frame of the preset question and answer start identifier is found from the voice frame set through traversal comparison, and the voice frame segment is determined as the voice segment with the same voice information as the preset question and answer start identifier.

[0115] The preset question and answer start identifier is a voice identifier for reminding that the self-introduction stage is completed and the question and answer stage starts, for example, a voice prompt such as "thank you for your introduction, now I want to ask you a few questions", and the preset can be specifically made according to actual conditions, which is not limited here.

[0116] The voice conversion text can use a voice recognition algorithm or a third-party tool with a voice conversion function, and the specific implementation is not limited. The voice conversion text algorithm includes but is not limited to a voice recognition algorithm based on a sound channel model, a voice template matching recognition algorithm and / or an artificial neural network voice recognition algorithm.

[0117] In this embodiment, the interview recording text is converted into a self-introduction text and an interview answer text, so that the two types of texts are processed separately in the subsequent processing, which is more targeted and the processing result is more accurate.

[0118] In some optional implementation manners of this embodiment, in step S203, the interview answer text is classified according to the interview angle involved, to obtain a classified text, which includes:

[0119] Each sentence in the interview answer text is taken as a basic sentence, and the basic sentence is processed by a preset word segmentation manner to obtain a basic word segmentation;

[0120] The basic word segmentation is converted into a word vector, and the word vector is clustered by a clustering algorithm to obtain a clustering center corresponding to the basic sentence;

[0121] For each basic sentence, the Euclidean distance between the clustering center corresponding to the basic sentence and the word vector corresponding to each preset interview angle is calculated, and the preset interview angle with the smallest distance is taken as the target classification of the basic sentence, and the basic sentence is taken as the classified text corresponding to the target classification.

[0122] Specifically, each sentence in the interview answer text is processed by word segmentation and clustering to obtain a clustering center corresponding to each sentence, and then the clustering center and a word vector corresponding to a preset interview angle are calculated to determine the classification to which each sentence belongs.

[0123] The preset word segmentation manner includes but is not limited to a third-party word segmentation tool or a word segmentation algorithm.

[0124] Among them, the common third-party word segmentation tools include but are not limited to: Stanford NLP word segmentation tool, ICTClAS word segmentation system, ansj word segmentation tool and HanLP Chinese word segmentation tool, etc.

[0125] Among them, the word segmentation algorithms include but are not limited to: rule-based word segmentation method, statistical-based word segmentation method, understanding-based word segmentation method and neural network word segmentation method.

[0126] The rule-based word segmentation method mainly includes: Minimum Matching, Maximum Matching, Reverse Directional Maximum Matching, Bi-Direction Maximum Matching (BMM), flag segmentation method, full segmentation path selection method and Association-Backtracking Method (AB method), etc.

[0127] The statistical-based word segmentation method mainly includes: N-Gram model, Hidden Markov Model (HMM) sequence labeling method, Maximum Entropy Model (MEM) sequence labeling method, Maximum Entropy Markov Model (MEMM) sequence labeling method and Conditional Random Fields (CRF) sequence labeling method, etc.

[0128] Preferably, the improved CRF model is used for word segmentation in the embodiment, and the specific implementation process can refer to the description of the subsequent embodiments. To avoid repetition, it will not be described here.

[0129] It is easy to understand that the basic word segmentation is extracted by the word segmentation method, on the one hand, some meaningless words in the text can be effectively filtered out, and on the other hand, it is also conducive to generating word vectors using these texts subsequently.

[0130] Among them, the cluster algorithm is also called group analysis, which is a statistical analysis method for sample or index classification problem, and is also an important algorithm of data mining. The cluster algorithm includes but is not limited to: K-means clustering algorithm, mean shift clustering algorithm, density-based clustering (Density-Based Spatial Clustering of Applications with Noise, DBSCAN) method, maximum expectation clustering based on Gaussian mixture model, agglomerative hierarchical clustering and graph community detection (Graph Community Detection) algorithm, etc.

[0131] Preferably, in the embodiment, the K-means clustering algorithm is adopted.

[0132] In the embodiment, by clustering and calculating semantic similarity, the classification of each sentence in the interview answer text is determined, which is beneficial to subsequent refinement of different classified sentences.

[0133] In some optional implementation manners of the embodiment, the basic sentence is processed by a preset word segmentation manner to obtain basic word segmentation, which includes:

[0134] The conditional random field model is adopted to segment the basic sentence to obtain initial word segmentation;

[0135] The word frequency of each initial word segmentation is obtained from the historical interview answer text;

[0136] Based on the word frequency of the initial word segmentation, the weight of the initial word segmentation is generated, and the initial word segmentation marked with the weight is taken as the basic word segmentation.

[0137] Specifically, the conditional random field model is adopted to segment the basic sentence to obtain initial word segmentation, and the word frequency of each initial word segmentation is obtained from the historical interview answer text, and the weight corresponding to the initial word segmentation is generated according to the word frequency to obtain the basic word segmentation with weight information, so that the proportion of each basic word segmentation is more in line with the needs of the interview scene when the basic word segmentation is marked subsequently.

[0138] Among them, the conditional random field (CRF) model is a discriminative probabilistic model, which is a kind of random field, commonly used for labeling or analyzing sequence data, representing the Markov random field of another group of output random variables Y given a group of input random variables X. It has good effect in sequence labeling tasks such as word segmentation, part-of-speech tagging and named entity recognition.

[0139] The historical interview response text refers to an interview response text generated by a historical interview. The word frequency of the historical interview response text can reflect the proportion of some words in the interview.

[0140] In this embodiment, the initial segmentation obtained by segmenting the conditional random field model is given a weight to obtain a basic segmentation that is more suitable for the intelligent interview scenario, which is conducive to improving the classification accuracy.

[0141] In some optional implementation manners of this embodiment, in step S204, the language extraction model is a bidirectional long short-term memory network model, the bidirectional long short-term memory network model includes a sentence encoder and a document encoder, and the language extraction model is used to extract sentences from each type of classification text to obtain the extracted sentences, which include:

[0142] The text in the classification text is split according to characters by the sentence encoder to obtain basic characters;

[0143] The basic characters are encoded to obtain the encoding content corresponding to the basic characters;

[0144] The encoding content is input into the character encoding layer with an initialized weight, each encoding is mapped into a character vector by the character encoding layer, and each character vector is taken as a sentence encoding result;

[0145] The sentence encoding result is spliced into a hidden layer vector in the forward and reverse hidden layers, and the hidden layer vector is input into the document encoder;

[0146] The hidden layer vector is weighted by the document encoder to obtain a document feature vector, and the document feature vector is decoded, and the output result obtained by decoding is taken as the extracted sentence.

[0147] Specifically, the text in the classification text is split and encoded according to characters by the sentence encoder to obtain the encoding content, the encoding content is input into the character encoding layer, each encoding corresponds to a character vector, each character vector is taken as a sentence encoding result, and the sentence encoding result is transmitted to the document encoder through the hidden layer, and the document encoder is weighted to obtain the extracted sentence.

[0148] It is worth noting that based on the sentence encoding result, the forward and reverse hidden layer outputs of each character in the model are spliced into a hidden layer vector: where the forward direction is indicated by a superscript +, the reverse direction is indicated by a superscript -, and the i-th character is indicated by a subscript i.

[0149] The long short-term memory network (LSTM) is a kind of time recurrent neural network, which is suitable for processing and predicting important events with relatively long intervals and delays in time series.

[0150] It should be noted that the unidirectional LSTM can be memorized from the first word to the last word of a sentence in the reading order of human beings, and such LSTM structure can only capture the above information and cannot capture the below information, while the bidirectional LSTM is composed of two different LSTMs, one LSTM reads data from front to back according to the word order in the sentence, and the other LSTM reads data from back to front according to the reverse direction of the word order in the sentence, so that the first LSTM obtains the above information and the other LSTM obtains the below information, and the combination of the two LSTMs is the context information of the entire sentence, and the context information is provided by the entire sentence and naturally contains abstract semantic information (the meaning of the sentence). The advantage of this method is to fully utilize the processing advantage of LSTM on sequence data with time sequence characteristics, and since the position feature is input, the entity direction information contained in the position feature can be extracted after being encoded by the bidirectional LSTM.

[0151] In the embodiment, the classified sentences are parsed and extracted by the sentence encoder and the document encoder from the bidirectional long short-term memory network at two different levels, so as to improve the accuracy of key sentence extraction.

[0152] In some optional implementation modes of the embodiment, the document feature vector is obtained by weighting the hidden layer vectors by the document encoder, and the document feature vector includes:

[0153] The document feature vector is determined by using the following formula:

[0154]

[0155] Wherein, C i is the i-th document feature vector, j is the sequence number of the embedding code, n is the number of embedding codes, b ij is the weight of the i-th document feature vector for the j-th hidden layer vector, h j is the j-th hidden layer vector, wherein the embedding code is generated based on the hidden state of the bidirectional long short-term memory network model.

[0156] In the embodiment, the generation mode of the document feature vector is obtained by weighting calculation, which is beneficial to accurately extract key sentences.

[0157] It should be understood that the size of the sequence number of each step in the above embodiment does not mean the order of execution, and the execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the application.

[0158] Figure 3 The principle block diagram of the artificial intelligence-based interview content refining device corresponding to the artificial intelligence-based interview content refining method in the above embodiment is shown. As shown inFigure 3 As shown, the artificial intelligence-based interview content refining device includes a text acquisition module 31, a text analysis module 32, a text classification module 33, a corpus extraction module 34, and an information sending module 35. The functions of each module are described in detail as follows:

[0159] The text acquisition module 31 is configured to acquire an interview recording and convert the interview recording into an interview text, wherein the interview text includes a self-introduction text and an interview response text.

[0160] The text analysis module 32 is configured to perform text analysis on the self-introduction text to obtain basic information of the interviewee.

[0161] The text classification module 33 is configured to perform sentence classification on the interview response text according to the interview angle involved to obtain a classified text.

[0162] The corpus extraction module 34 is configured to perform sentence extraction from each type of classified text by using a language extraction model to obtain extracted sentences, and refine the extracted sentences by using a Transformer model to obtain interview refined corpus.

[0163] The information sending module 35 is configured to send the basic information of the interviewee and the interview refined corpus to a management terminal, so that the management terminal determines an interview result according to the basic information of the interviewee and the interview refined corpus.

[0164] Optionally, the text acquisition module 31 includes:

[0165] An identifier recognition unit is configured to recognize a question-answer start identifier included in the interview recording.

[0166] A text determination unit is configured to convert the interview recording into text by using a voice-to-text conversion manner, convert the text obtained from the recording content before the question-answer start identifier into the self-introduction text, and convert the text obtained from the recording content before the question-answer start identifier into the interview response text.

[0167] Optionally, the text classification module 33 includes:

[0168] A word segmentation unit is configured to take each sentence in the interview response text as a basic sentence, perform word segmentation processing on the basic sentence by using a preset word segmentation manner, and obtain basic word segmentation.

[0169] A clustering unit is configured to convert the basic word segmentation into a word vector, perform clustering on the word vector by using a clustering algorithm, and obtain a clustering center corresponding to the basic sentence.

[0170] The classification unit is configured to calculate, for each basic sentence, an Euclidean distance between a cluster center corresponding to the basic sentence and a word vector corresponding to each preset interview angle, and take a preset interview angle with the smallest distance as a target classification of the basic sentence, and take the basic sentence as a classification text corresponding to the target classification.

[0171] Optionally, the word segmentation unit comprises:

[0172] The initial word segmentation unit is configured to perform word segmentation on the basic sentence by using a conditional random field model to obtain initial word segmentation.

[0173] The word frequency acquisition sub-unit is configured to acquire a word frequency of each initial word segmentation from the historical interview answer text.

[0174] The word segmentation weighting unit is configured to generate a weight of the initial word segmentation based on the word frequency of the initial word segmentation, and take the initial word segmentation with the weight as the basic word segmentation.

[0175] Optionally, the corpus extraction module 34 comprises:

[0176] The splitting unit is configured to split the text in the classification text into basic characters according to characters by using a sentence encoder.

[0177] The encoding unit is configured to encode the basic characters to obtain encoding content corresponding to the basic characters.

[0178] The mapping unit is configured to input the encoding content to a character encoding layer with an initial weight, map each encoding to a character vector by using the character encoding layer, and take each character vector as a sentence encoding result.

[0179] The concatenating unit is configured to concatenate the sentence encoding result in a forward and reverse hidden layer output into a hidden layer vector, and input the hidden layer vector to a document encoder.

[0180] The weighting unit is configured to weight the hidden layer vector by using the document encoder to obtain a document feature vector, and decode the document feature vector to take an output result obtained by decoding as an extracted sentence.

[0181] Optionally, the weighting decoding unit comprises:

[0182] The calculation sub-unit is configured to determine the document feature vector by using the following formula:

[0183]

[0184] wherein, C i is the i-th document feature vector, j is a serial number of an embedding code, n is a number of embedding codes, b ij is a weight of the i-th document feature vector for the j-th hidden layer vector, h jis the jth hidden layer vector, wherein the embedding code is generated based on hidden states of a bidirectional long short-term memory network model.

[0185] Optionally, the AI-based interview content refining device further comprises:

[0186] The storage module is configured to store the interviewee basic information and the interview refining corpus into the blockchain network.

[0187] The specific limitations of the AI-based interview content refining device can be found in the limitations of the AI-based interview content refining method described above, which will not be repeated here. Each module in the AI-based interview content refining device described above can be realized by software, hardware and their combinations in whole or in part. The above-mentioned modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to each module.

[0188] To solve the above technical problems, the embodiments of the present application further provide a computer device. For details, please refer to Figure 4 , Figure 4 The basic structure block diagram of the computer device of the present embodiment is shown in the figure.

[0189] The computer device 4 comprises a memory 41, a processor 42 and a network interface 43 which are connected to each other through a system bus. It should be pointed out that only the computer device 4 with the components of the memory 41, the processor 42 and the network interface 43 is shown in the figure, but it should be understood that it is not required to implement all the shown components, and more or fewer components can be alternatively implemented. Among them, those skilled in the art can understand that the computer device here is a device capable of automatically performing numerical calculation and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessor, application specific integrated circuit (ASIC), field programmable gate array (FPGA), digital signal processor (DSP), embedded device, etc.

[0190] The computer device can be a desktop computer, a notebook computer, a palm computer and a cloud server, etc. The computer device can interact with the user through a keyboard, a mouse, a remote controller, a touchpad or a voice control device, etc.

[0191] The memory 41 includes at least one type of readable storage medium, such as a flash memory, a hard disk, a multimedia card, a card-type memory (e.g., an SD or D interface display memory, etc.), a random access memory (RAM), a static random access memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a magnetic disk, an optical disk, etc. In some embodiments, the memory 41 can be an internal storage unit of the computer device 4, such as a hard disk or a memory of the computer device 4. In other embodiments, the memory 41 can also be an external storage device of the computer device 4, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the computer device 4. Of course, the memory 41 can also include both the internal storage unit and the external storage device of the computer device 4. In this embodiment, the memory 41 is generally used to store an operating system and various application software installed on the computer device 4, such as program codes for controlling electronic files, etc. In addition, the memory 41 can also be used to temporarily store various data that have been output or will be output.

[0192] The processor 42 can be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip in some embodiments. The processor 42 is generally used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to run program codes or process data stored in the memory 41, such as running program codes for controlling electronic files.

[0193] The network interface 43 can include a wireless network interface or a wired network interface, and is generally used to establish a communication connection between the computer device 4 and other electronic devices.

[0194] The present application also provides another embodiment, i.e., to provide a computer readable storage medium storing an interface display program, the interface display program being executable by at least one processor to cause the at least one processor to perform the steps of the artificial intelligence-based interview content refining method as described above.

[0195] Those skilled in the art can clearly understand the above-mentioned embodiment method can be realized by means of software and the necessary general hardware platform, of course, it can also be realized by hardware, but in many cases, the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes a plurality of instructions for making a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) execute the method described in each embodiment of the present application.

[0196] Obviously, the above-described embodiments are only some of the embodiments of the present application, not all the embodiments, and the drawings show the preferred embodiments of the present application, but do not limit the patent scope of the present application. The present application can be implemented in many different forms, and conversely, the purpose of providing these embodiments is to make the disclosure of the present application more thorough and comprehensive. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions recorded in the foregoing specific embodiments, or make equivalent replacements to some of the technical features. Any equivalent structure made by using the content of the specification and drawings, directly or indirectly applied to other related technical fields, is also within the scope of the patent protection of the present application.

Claims

1. An interview content refining method based on artificial intelligence, characterized by, The interview content refining method based on artificial intelligence comprises: acquiring an interview recording and converting the interview recording into an interview text, wherein the interview text comprises a self-introduction text and an interview answer text; performing text analysis on the self-introduction text to obtain interviewer basic information; classifying sentences in the interview answer text according to involved interview angles to obtain classified texts; extracting sentences from each type of the classified texts by a language extraction model to obtain extracted sentences, and refining the extracted sentences by a Transformer model to obtain interview refined corpus; sending the interviewer basic information and the interview refined corpus to a management end to enable the management end to determine an interview result according to the interviewer basic information and the interview refined corpus; the involved interview angles refer to the focus of questions and answers, including salary requirements, awards obtained, work experience and professional skills, and the classification of sentences in the interview answer text according to the involved interview angles to obtain classified texts comprises: taking each sentence in the interview answer text as a basic sentence, and performing word segmentation on the basic sentence by a preset word segmentation mode to obtain basic word segmentation; converting the basic word segmentation into a word vector, and clustering the word vector by a clustering algorithm to obtain a clustering center corresponding to the basic sentence; for each basic sentence, calculating the Euclidean distance between the clustering center corresponding to the basic sentence and a word vector corresponding to each preset interview angle, and taking the preset interview angle with the smallest distance as a target classification of the basic sentence, and taking the basic sentence as the classified text corresponding to the target classification; the word segmentation of the basic sentence by the preset word segmentation mode to obtain the basic word segmentation comprises: performing word segmentation on the basic sentence by a conditional random field model to obtain initial word segmentation; acquiring the word frequency of each initial word segmentation from historical interview answer texts; generating the weight of the initial word segmentation based on the word frequency of the initial word segmentation, and taking the initial word segmentation with the weight labeled as the basic word segmentation; the language extraction model is a bidirectional long short-term memory network model, which comprises a sentence encoder and a document encoder, and the sentence extraction from each type of the classified texts by the language extraction model to obtain the extracted sentences comprises: splitting the text in the classified text according to characters by the sentence encoder to obtain basic characters; encoding the basic characters to obtain the encoding content corresponding to the basic characters; inputting the encoding content into an initialized weight character encoding layer, mapping each encoding into a character vector by the character encoding layer, and taking each character vector as a sentence encoding result; concatenating the sentence encoding result in the forward and reverse hidden layers into a hidden layer vector, and inputting the hidden layer vector into the document encoder; weighting the hidden layer vector by the document encoder to obtain a document feature vector, and decoding the document feature vector to obtain an output result as the extracted sentence.

2. The artificial intelligence-based interview content refining method of claim 1, wherein, The converting the interview audio into an interview text comprises: identifying a question-and-answer start identifier contained in the interview audio; performing text conversion on the interview audio by using a voice-to-text conversion manner, and converting the audio content before the question-and-answer start identifier into a text as a self-introduction text and converting the audio content before the question-and-answer start identifier into a text as an interview response text. 3.The AI-based interview content refining method of claim 1, wherein, The weighting the hidden layer vector by the document encoder to obtain a document feature vector comprises: determining the document feature vector by using the following formula: wherein C i is the i-th document feature vector, j is the serial number of the embedding code, n is the number of embedding codes, b ij is the weight of the i-th document feature vector for the j-th hidden layer vector, h j is the j-th hidden layer vector, wherein the embedding code is generated based on the hidden state of the bidirectional long short-term memory network model. 4.The AI-based interview content refining method of claim 1, wherein, After the refining the extracted sentence by using the Transformer model to obtain the interview refined corpus, the method further comprises: storing the interviewee basic information and the interview refined corpus into a blockchain network. 5.An interview content refining apparatus based on artificial intelligence, characterized by, The AI-based interview content refining device implements the steps of the AI-based interview content refining method according to any one of claims 1 to 4, and the AI-based interview content refining device comprises: a text acquisition module configured to acquire an interview audio and convert the interview audio into an interview text, wherein the interview text comprises a self-introduction text and an interview response text; a text analysis module configured to perform text analysis on the self-introduction text to obtain interviewee basic information; a text classification module configured to perform sentence classification on the interview response text according to an interview angle involved to obtain classified texts; a corpus extraction module configured to perform sentence extraction from each of the classified texts by using a language extraction model to obtain extracted sentences, and refine the extracted sentences by using a Transformer model to obtain interview refined corpus; an information sending module configured to send the interviewee basic information and the interview refined corpus to a management end, so that the management end determines an interview result according to the interviewee basic information and the interview refined corpus.

6. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor implements the AI-based interview content refining method according to any one of claims 1 to 4 when executing the computer program.

7. A computer-readable storage medium storing a computer program, wherein the computer program comprises the following steps of: receiving a request for a resource from a client; determining whether the client is authorized to access the resource; and if the client is authorized to access the resource, providing the resource to the client. The computer program implements the AI-based interview content refining method according to any one of claims 1 to 4 when executed by the processor.

Images