CN110347787A - A kind of interview method, apparatus and terminal device based on AI secondary surface examination hall scape - Google Patents
A kind of interview method, apparatus and terminal device based on AI secondary surface examination hall scape Download PDFInfo
- Publication number
- CN110347787A CN110347787A CN201910504604.2A CN201910504604A CN110347787A CN 110347787 A CN110347787 A CN 110347787A CN 201910504604 A CN201910504604 A CN 201910504604A CN 110347787 A CN110347787 A CN 110347787A
- Authority
- CN
- China
- Prior art keywords
- interview
- similar
- engine
- key message
- template
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/3332—Query translation
- G06F16/3334—Selection or weighting of terms from queries, including natural language queries
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3343—Query execution using phonetics
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/10—Office automation; Time management
- G06Q10/105—Human resources
- G06Q10/1053—Employment or hiring
Abstract
The present invention is suitable for computer application technology, provides a kind of interview method, apparatus and terminal device based on AI secondary surface examination hall scape, method includes: the voice content for obtaining interviewer's statement, and voice content is converted to content of text；Word segmentation processing is carried out to content of text, obtains participle database；Key message is obtained in participle database；Current interview scene type is judged according to key message, and it chooses whether to enable according to interview scene type and similar asks engine, start similar when asking engine, question template is obtained according to the similar processing result for asking engine, and Rating Model is selected to score the answer of interviewee；Do not start similar when asking engine, question template is obtained according to key message, and Rating Model is selected to score the answer of interviewee.Through the invention can retain true man interview flexible and subjective initiative simultaneously, improve AI interview evaluation accuracy rate.
Description
Technical field
The present invention relates to computer application technology more particularly to a kind of interview sides based on AI secondary surface examination hall scape
Method, device and terminal device.
Background technique
AI auxiliary interview, i.e. true man interviewer carry out candidate under the assistance of AI auxiliary interview exam system live or long-range
The behavior of interview.Compared to tradition interview scene is played, AI auxiliary interview can provide candidate message for interviewer, interview is recommended to ask
Topic records interview process, and can generate appraisal report to interview result, has the advantages of procedure structure, standards of grading.
In the scape of AI secondary surface examination hall, AI assists interview exam system can be based on current interview position ability portrait, from exam pool
Corresponding issue list is generated, is played the role of for interviewer's prompter and to problem score.
But due to the limitation of AI technology current stage, function of giving a mark is only applicable to already present problem in exam pool, right
Interim impromptu the problem of getting do not have marking ability in interviewer.On the other hand, in practical interview process, even exam pool
The problems in, the enquirement of interviewer sequence and concrete syntax expression are also likely to different from standard way to put questions, this is to AI auxiliary interview
Stability propose challenge.
Summary of the invention
It is a primary object of the present invention to propose that a kind of interview method, apparatus based on AI secondary surface examination hall scape and terminal are set
It is standby, when solving the problem of in the prior art using the auxiliary interview of AI technology it is interim for interviewer it is impromptu get not to have beat
The ability of dividing, and when interviewer's interview, using the problem having in exam pool, there is also scores to determine inaccuracy, and interview result analysis is not
Reliable and stable problem.
To achieve the above object, first aspect of the embodiment of the present invention provides a kind of interview side based on AI secondary surface examination hall scape
Method, comprising:
The voice content of interviewer's statement is obtained, and the voice content is converted into content of text；
Word segmentation processing is carried out to the content of text, obtains participle database；
Key message is obtained in the participle database；
Current interview scene type is judged according to the key message, and chooses whether to open according to the interview scene type
Engine is asked with similar, it is described similar to ask engine for according to the problems in participle term vector matching problem database template；
Start it is similar ask engine when, the processing result for asking engine according to similar obtains question template, and selects scoring mould
Type scores to the answer of interviewee；
Do not start it is similar ask engine when, question template is obtained according to the key message, and selects Rating Model opposite
The answer of examination person is scored.
In conjunction with first aspect present invention, in first aspect present invention first embodiment, judged according to the key message
Current interview scene type, and choose whether to enable according to the interview scene type and similar ask engine, comprising:
The problem of detecting short sentence quantity and the key message in participle database ID quantity；
The problem of key message ID quantity with it is described participle database in short sentence quantity it is identical when, then work as front
Examination scene type is people's machine side examination hall scape, does not enable and similar asks engine；
The problem of key message ID quantity and it is described participle database in short sentence quantity difference when, then work as front
Trying scene type is live secondary surface examination hall scape, and enabling is similar to ask engine.
In conjunction with first aspect present invention, in first aspect present invention second embodiment, start it is similar ask engine when, root
Question template is obtained according to the similar processing result for asking engine, and Rating Model is selected to score the answer of interviewee, comprising:
Start it is similar ask engine when, call trained matching neural network model in the similar engine, pass through institute
Matching neural network model is stated, the key message is converted into the participle term vector；
The problems in described problem database template is matched according to the participle term vector；
Rating Model corresponding to the problem of calling successful match template, the answer according to the Rating Model to interviewee
It scores.
In conjunction with first aspect present invention second embodiment, in third embodiment of the invention, according to the participle word to
The problems in flux matched described problem database template, comprising:
Obtain the template term vector of each question template；
The participle term vector and the template term vector are compared, cosine similarity is calculated；
According to the cosine similarity judge the participle term vector whether with described problem template matching.
In conjunction with first aspect present invention, in the 4th embodiment of first aspect present invention, do not start it is similar ask engine when,
Question template is obtained according to the key message, and Rating Model is selected to score the answer of interviewee, comprising:
Do not start it is similar ask engine when, according to the problems in key message ID match described problem database in
Question template；
Rating Model corresponding to the problem of calling successful match template, the answer according to the Rating Model to interviewee
It scores.
In conjunction with first aspect present invention, in the 5th embodiment of first aspect present invention, the Rating Model includes rule
Rating Model and comprehensive grade model；
Wherein, when the interview scene type is people's machine side examination hall scape, the Rating Model includes regular Rating Model；
When the interview scene is live secondary surface examination hall scape, the Rating Model includes comprehensive grade model.
In conjunction with first aspect present invention, in first aspect present invention sixth embodiment, carried out to the content of text
Word segmentation processing obtains before segmenting database, comprising:
Noise information replacement is carried out to the content of text, filters non-critical information.
Second aspect of the present invention provides a kind of interview device based on AI secondary surface examination hall scape, comprising:
Text obtains module, for obtaining the voice content of interviewer's statement, and the voice content is converted to text
Content；
Word segmentation processing module obtains participle database for carrying out word segmentation processing to the content of text；
Key message obtains module, for obtaining key message in the participle database；
Problem matching module, for judging current interview scene type according to the key message, and according to the interview
Scene type choose whether to enable it is similar ask engine, it is described similar to ask engine for according to the participle term vector matching problem number
According to the problems in library template；
Grading module, for start it is similar ask engine when, the processing result for asking engine according to similar obtains question template,
And Rating Model is selected, it scores the answer of interviewee；
Do not start it is similar ask engine when, question template is obtained according to the key message, and selects Rating Model opposite
The answer of examination person is scored.
The third aspect of the embodiment of the present invention provides a kind of terminal device, including memory, processor and is stored in
In above-mentioned memory and the computer program that can be run on above-mentioned processor, when above-mentioned processor executes above-mentioned computer program
The step of realizing method provided by first aspect as above.
The fourth aspect of the embodiment of the present invention provides a kind of computer readable storage medium, above-mentioned computer-readable storage
Media storage has computer program, and above-mentioned computer program realizes method provided by first aspect as above when being executed by processor
The step of.
The embodiment of the present invention proposes a kind of interview method based on AI secondary surface examination hall scape, and interviewer is asked by voice
The problem of be converted to content of text, then word segmentation processing is carried out to content of text, obtains the participle data based on interview content of text
Library, to participle database analyze, obtain can be used for judging currently interview scene type key message, finally further according to
Interview scene selects the similar use for asking engine, so that arbitrary face examination hall scape can be inscribed according to the enquirement accurate match of interviewer
The problems in library template, the suitable Rating Model of simultaneous selection, so that interviewer higher can carry out interviewee with freedom degree
It puts question to, and interviewee can also be scored accordingly for the answer of problem described by interviewer in AI interview exam system, because
This, the embodiment of the present invention realize retain true man interview flexible and subjective initiative simultaneously, improve AI interview evaluation
Accuracy rate.
Detailed description of the invention
Fig. 1 is the implementation process signal for the interview method based on AI secondary surface examination hall scape that the embodiment of the present invention one provides
Figure；
Fig. 2 is that the implementation process of the interview method provided by Embodiment 2 of the present invention based on AI secondary surface examination hall scape is illustrated
Figure；
Fig. 3 is the matching Artificial Neural Network Structures schematic diagram that the embodiment of the present invention three provides；
Fig. 4 is the implementation process signal for the interview method based on AI secondary surface examination hall scape that the embodiment of the present invention three provides
Figure；
Fig. 5 is the composed structure signal for the interview device based on AI secondary surface examination hall scape that the embodiment of the present invention four provides
Figure.
The embodiments will be further described with reference to the accompanying drawings for the realization, the function and the advantages of the object of the present invention.
Specific embodiment
It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not intended to limit the present invention.
It should be noted that, in this document, the terms "include", "comprise" or its any other variant are intended to non-row
His property includes, so that the process, method, article or the device that include a series of elements not only include those elements, and
And further include other elements that are not explicitly listed, or further include for this process, method, article or device institute it is intrinsic
Element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including being somebody's turn to do
There is also other identical elements in the process, method of element, article or device.
Herein, using the suffix for indicating such as " module ", " component " or " unit " of element only for advantageous
In explanation of the invention, there is no specific meanings for itself.Therefore, " module " can be used mixedly with " component ".
In subsequent description, inventive embodiments serial number is for illustration only, does not represent the advantages or disadvantages of the embodiments.
Embodiment one
As shown in Figure 1, passing through offer the embodiment of the invention provides a kind of interview method based on AI secondary surface examination hall scape
Suitable Rating Model auxiliary interview, method includes but is not limited to following steps:
S101, the voice content for obtaining interviewer's statement, and the voice content is converted into content of text.
Front end system can be used in the realization of above-mentioned steps S101；Wherein, front end system can be real by the end PC or mobile phone terminal
It is existing.Front end system is realized out of voice by automatic speech recognition (Automatic Speech Recognition, ASR) technology
Hold the conversion of content of text.
In one embodiment, the content of text obtained can be by text recognition technique, for example, natural language processing
(Natural Language Processing, NLP) technology identifies the mistake place in content of text, for example, semantic, grammer
Or/and the mistake in terms of logic；And relevant error is corrected.
In embodiments of the present invention, after voice content is converted to content of text, content of text be multiple groups short sentence, as :/you
What name is/,/your inaugural position last time what is/.
S102, word segmentation processing is carried out to the content of text, obtains participle database.
In above-mentioned steps S102, word segmentation processing can also be completed by front end system, can also be by subsequent docking front end system
The middle control engine of system is completed；Wherein, middle control engine is the data system on backstage.
In a particular application, execute word segmentation processing task can be segmenter, be also possible to segmentation methods, most such as forward direction
Big matching method, reverse maximum matching method and bi-directional matching participle method etc..
In practical applications, the content of text after word segmentation processing is using word as node, and each word has word
Property mark, be convenient for subsequent key information extraction when, filtered using grammer, such as only retain specific part of speech (such as adjective and noun)
Word.
In embodiments of the present invention, after carrying out word segmentation processing to each short sentence, segmenting the data in database is based on more
Group short sentence multiple groups participle ,/name/,/last time/position/.
In one embodiment, before above-mentioned steps S102, may include:
Noise information replacement is carried out to the content of text, filters non-critical information.
Replaced by noise information, content of text can be simplified, make the key message of step S103 obtain it is more accurate simultaneously
With specific aim.
In a particular application, noise information can be for name, place name, date, money etc., and the replacement of noise information then can be with
It is matched and is replaced using canonical, replace above-mentioned name, place name, date, money etc.；Then, it carries out Chinese word segmentation and (e.g., identifies language
Sentence ingredient, subject and predicate, guest, fixed, shape etc.), and remove stop words.Stop words refers in information retrieval, for save memory space and
Search efficiency is improved, certain words or word are fallen in meeting automatic fitration before or after handling natural language data (or text), these
Word or word are referred to as Stop Words (stop words).
S103, key message is obtained in the participle database.
In above-mentioned steps S103, after completing word segmentation processing, interviewer ID, candidate can be targetedly obtained
The key messages such as ID, company ID, scene ID, position ID, problem ID and question text.
In embodiments of the present invention, the extraction of key message can be based on having filtered non-pass based on participle database
The participle database of key information.
In embodiments of the present invention, the unrelated keyword abstraction algorithm in selection field carries out the acquisition of key message, if any
Supervised learning algorithm and unsupervised learning algorithm
Wherein, keyword abstraction process is considered as two classification problems by supervised learning algorithm, first extracts candidate word, then
Label delimited for each candidate word or is keyword or is not keyword, then trains keyword abstraction classifier.
When newly arrive a document when, extract all candidate words, then utilize trained keyword abstraction classifier, to each time
Word is selected to classify, finally using the candidate word that label is keyword as keyword.
And unsupervised learning algorithm, candidate word is first extracted, is then given a mark to each candidate word, topK is then exported
A highest candidate word of score value is as keyword.It is different according to the strategy of marking, there are different algorithms, such as TF-IDF,
TextRank scheduling algorithm.
S104, current interview scene type is judged according to the key message, and is selected according to the interview scene type
Whether enable and similar asks engine.
Wherein, described similar to ask engine for according to participle the problems in term vector matching problem database template.
In one embodiment, a kind of implementation of above-mentioned steps S104 can be with are as follows:
The problem of detecting short sentence quantity and the key message in participle database ID quantity；
The problem of key message ID quantity it is identical as the short sentence quantity in the participle database when, then current interview
Scene type is people's machine side examination hall scape, does not enable and similar asks engine；
The problem of key message ID quantity and the participle database in short sentence quantity difference when, then current interview
Scene type is live secondary surface examination hall scape, and enabling is similar to ask engine.
In a particular application, when the key message in every group of short sentence includes problem ID, each problem of interviewer is
The problems in issue database, current scene type of interviewing is people's machine side examination hall scape.If the key message in any group of short sentence is not
When including problem ID, the problem of current scene type of interviewing is live secondary surface examination hall scape, interview, may not be from problem number
According to library.
S105, start it is similar ask engine when, the processing result for asking engine according to similar obtains question template, and selects to comment
Sub-model scores to the answer of interviewee；
Do not start it is similar ask engine when, question template is obtained according to the key message, and selects Rating Model opposite
The answer of examination person is scored.
In practical applications, content of text may be the problem having in issue database, it is also possible to not have in database
The problem of having, all differences because interviewing scene type, therefore different sides examination scene type is obtained using different problems template
Mode.
In man-machine interview scene, in the participle database after carrying out word segmentation processing, every group of short sentence is issue database
In the problem that has can match problem template corresponding to every group of short sentence therefore according to the problems in key message ID, complete
The problem of entire content of text, matches, and does not start similar ask engine at this time.
At the scene in the scape of secondary surface examination hall, according only to key message can not in issue database accurate match to corresponding
Question template, need to start at this time it is similar ask engine, ask engine by similar, key message handled.
Wherein, similar to ask that engine passes through the problems in the term vector matching problem database of content of text template, term vector
It can reflect the positional relationship between respectively segmenting and segment in content of text, when extracting key message in participle database,
The feature extraction accuracy to participle is improved, therefore when interviewing scene is live secondary surface examination hall scape, it still can be accurate
It is matched to question template corresponding with content of text.The similar processing for asking engine is explained in the following examples and
Explanation.
In one embodiment, the Rating Model includes regular Rating Model and comprehensive grade model；
Wherein, when the interview scene type is people's machine side examination hall scape, the Rating Model includes regular Rating Model；Institute
When stating interview scene as live secondary surface examination hall scape, the Rating Model includes comprehensive grade model.
In above-mentioned man-machine interview scene, the typical problem usually got by system, and in the scape of live secondary surface examination hall, lead to
It often will appear the problem of interviewer freely plays.
In embodiments of the present invention, Rating Model is for determining which set Rating Model of selection evaluates interviewee.Appoint
The interview scene type of meaning may be introduced into Rating Model and assist interview, and difference is obtained according to the similar processing result for asking engine
The Rating Model for taking question template and selecting question template and selects Rating Model with obtaining according to key message, scores
Standard or code of points are different.
The embodiment of the present invention is also interview scene type with man-machine interview scene and live secondary surface examination hall scape, to upper commentary
Standards of grading or code of points in sub-model are illustrated:
When examination hall scape is people's machine side examination hall scape face to face, the problem of interviewee obtains from issue database, then according to point
The key message that word database extracts, one can surely be matched to related problem in issue database, then in such face
, can be with accurate match question template according to key message under the scape of examination hall, and select regular Rating Model, wherein rule scoring mould
Whether type, which can be understood as machine, can recognize the marking mode of class, be matched in Rating Model from answer fluency, answer content
The dimensions such as score keyword score interviewee, such as:
Answer fluency: whether identification candidate there is stopping for long period (passing through threshold decision) during answering
?；The words and phrases continuously repeated whether are stored in identification candidate's answer content (identification is nervous, or situations such as stammerer).
The score keyword whether answer content is matched in Rating Model: some answers have been preset in Rating Model, and this is asked
There are these keywords as long as recognizing to get to accordingly in some score keywords when topic in the content that candidate answers
Score.
When examination hall scape is live secondary surface examination hall scape face to face, the problem of interviewee obtains, is not necessarily from problem data
Library then according to the key message that participle database extracts is possible to that related problem can not be matched in issue database,
Under such interview scene, should start it is similar ask engine, it is similar to ask that engine carries out term vector processing, according to its processing result can
With question template, and select comprehensive grade model, wherein comprehensive grade model, which can be understood as machine, can recognize class and artificial auxiliary
The marking mode for judging that class be combined with each other is helped, i.e., whether is matched to the score in Rating Model from answer fluency, answer content
The dimensions such as keyword score to interviewee, meanwhile, from 1, interviewer to the evaluation of the answer logicality, 2, interviewer to time
It chooses the evaluation of anti-pressure ability when answering the problem；3, evaluation of emergency capability etc. is tieed up when interviewer answers the problem to candidate
Degree scores to interviewee.In practical applications, human assistance judges the evaluation of class, and system can provide several opinion ratings
It is selected for interviewer, the opinion rating and want the weight answered that system is obtained according to each dimension, when calculating to candidate's answer
Human assistance judges the scoring of class.Finally integrating machine can recognize that class and human assistance judge the score of each dimension of class, obtain
To final score when answering candidate.
Above-mentioned step S101 to step S103 is the realization process of the acquisition and processing of content of text, above-mentioned steps S104
With step S105 be according to the selection of interview scene type how matching problem template, and selection Rating Model, and to candidate
The realization process to score, the interview method provided in an embodiment of the present invention based on AI secondary surface examination hall scape, interviewer is led to
It crosses the problem of voice is asked and is converted to content of text, then word segmentation processing is carried out to content of text, obtain based on interview content of text
Participle database, to participle database analyze, obtain can be used for judging currently interview scene type key message,
The similar use for asking engine finally is selected further according to interview scene, so that arbitrary face examination hall scape can be according to the enquirement of interviewer
The problems in accurate match exam pool template, the suitable Rating Model of simultaneous selection, so that interviewer can be right higher with freedom degree
Interviewee puts question to, and interviewee can also obtain phase in AI interview exam system for the answer of problem described by interviewer
The scoring answered, therefore, the embodiment of the present invention realize retain true man interview flexible and subjective initiative simultaneously, improve
The accuracy rate of AI interview evaluation.
Embodiment two
As shown in Fig. 2, the embodiment of the present invention provides a kind of interview method based on AI secondary surface examination hall scape, implementation is shown
The detailed implementation process of step S105 in example one.In step S105 start it is similar ask engine when, ask engine according to similar
Processing result obtain question template, and select Rating Model, score the answer of interviewee, may include:
S10511, start it is similar ask engine when, call trained matching neural network model in the similar engine,
By the matching neural network model, the key message is converted into term vector.
In above-mentioned steps S10511, by the key message in the content of text stated based on interviewer be converted to word to
Amount, is matched, available accurate matching range by term vector.
In one embodiment, it is contemplated that interviewer may temporarily investigate that interviewee is simpler or more complicated to be asked
Topic then in the case where starting the similar interview scene for asking engine, can will compare logical design to determine whether it can be competent at other position
To match all problems in the said firm's database, that is, the problem of being not limited solely to this position logic.
In embodiments of the present invention, it can be understood by general semantics and HR related text trains neural network, thus
To trained matching neural network model.
The embodiment of the present invention also shows the process using Word2Vec neural network model training term vector:
Firstly, as shown in figure 3, Word2Vec includes two submodels: the part a indicates in CBOW and Skip-Gram, Fig. 3
The part CBOW, b indicates Skip-Gram, includes three input, mapping and output neural net layers, and the w (t) of the part a indicates defeated
Out, the w (t) of the part b indicates input.
Then, above-mentioned two submodel is trained, trained method is: " surrounding word, which stacks up, predicts current word "
(P (wt | Context) P (wt | Context)) and " current word predicts surrounding word respectively " (P (wothers | wt) P (wothers
|wt))。
Wherein, accelerating method can also be used, is trained；It specifically, is that level Softmax and negative sample are adopted respectively
Sample.
Level Softmax is the simplification to Softmax, and the efficiency of prediction probability is directly reduced to O from O (| V |) O (| V |)
(log2 | V |) O (log2 | V |), but comparatively, precision can be slightly poorer than primary Softmax；Negative sample sampling then uses phase
Anti- thinking, it joins together original outputting and inputting to do one two classification as input then to give a mark, form joint
The modeling of probability P (wt, Context) P (wt, Context) and P (wothers, wt) P (wothers, wt), wherein positive sample
Occurred using corpus, and negative sample take out at random it is several.
The embodiment of the present invention also shows the training condition in practical application, includes the following:
Training corpus: the article of wechat public platform, it is multi-field, belong to Chinese balance corpus, interview corpus that HR is given to
Swash the interview scene corpus got for internet；
Corpus quantity: 8,000,000, total word number reaches 65,000,000,000；
Model word number: totally 352196 word, substantially Chinese word include common English words；
Model structure: Skip-Gram+Huffman Softmax；
Vector dimension: 256 dimensions；
Participle tool: stammerer participle joined the dictionary of 500,000 entries, close new word discovery；
Training tool: the Word2Vec of Gensim, server have trained 7 days；
Other situations: window size 10, minimum word frequency are 64, iteration 10 times.
In a particular application, with the use of AI interview exam system, interviewer and interview also be will record in system database
The interview content of person, as the training sample of the matching neural network model, so that the matching neural network model is in use
Further optimized.
In a particular application, after the completion of above-mentioned matching neural network model training, determining that interview scene type is scene
When the scape of secondary surface examination hall, i.e., start it is similar ask engine when, the conversion of term vector asks that engine executes by similar.
S10512, the problems in described problem database template is matched according to the participle term vector.
In the matching process that above-mentioned steps S10512 is term vector and term vector, wherein the participle based on participle database
Term vector is completed by step S10511, and the term vector of question template can directly acquire.
In one embodiment, above-mentioned steps S10512 may include:
Obtain the template term vector of each question template；
The participle term vector and the template term vector are compared, cosine similarity is calculated；
According to the cosine similarity judge the participle term vector whether with described problem template matching.
In a particular application, cosine similarity is in term vector for reflecting a change of position of each participle in sentence
Amount, the cosine similarity of the participle term vector by the cosine similarity in template term vector and based on content of text, can solve
When certainly interviewer proposes random interview question, because what reverse participle occurred, question template matching is unsuccessful, or matching
The problem of inaccuracy.
In practical applications, process term vector and template term vector compared can be with are as follows:
After getting the term vector of content of text, all term vectors are averaged, obtain the vector of sentence.Then will
The sentence vector of interviewee's corpus of input is compared with the sentence vector of per pass typical problem template, calculates its cosine phase
Like degree, the formula of cosine similarity is as follows:
In a particular application, term vector is that the vector of 100-5000 dimension can will own in the same machine learning model
Keyword uniformly convert some vector for fixing dimension, be 256 dimensions in this motion.
In one embodiment, adjustable decision threshold can be set, for auxiliary judgment participle term vector whether with
The matching of template term vector.Such as, if when calculated cosine similarity has data to be greater than this threshold value, then it is assumed that the maximum number of similarity
The problem of according to being as matched to, if being greater than threshold value without data, then it is assumed that be not matched to problem.
S10513, Rating Model corresponding to the problem of successful match template is called, according to the Rating Model to interview
The answer of person is scored.
In above-mentioned steps S10513, Rating Model corresponding to each question template is different, in the embodiment of the present invention
Question template is applicable in secondary surface examination hall scape on site, therefore its Rating Model may include comprehensive grade model, i.e. machine can
Identification class and human assistance judge the marking mode that class be combined with each other.
Embodiment three
As shown in figure 4, the embodiment of the present invention provides a kind of interview method based on AI secondary surface examination hall scape, implementation is shown
The detailed implementation process of step S105 in example one.In step S105 do not start it is similar ask engine when, according to the key
Acquisition of information question template, and Rating Model is selected to score the answer of interviewee, may include:
S10521, do not start it is similar ask engine when, according to the problems in key message ID match described problem number
According to the problems in library template.
S10522, Rating Model corresponding to the problem of successful match template is called, according to the Rating Model to interview
The answer of person is scored.
In above-mentioned steps S10521 and step S10522, does not start and similar ask engine, then it represents that in current interview scene,
It is not in the speaking ability by random question of interviewer, interview process is all completed by machine.It at this time can be accurate according to key message
Allot question template.
Example IV
As shown in figure 5, present invention implementation provides a kind of interview device 50 based on AI secondary surface examination hall scape, comprising:
Text obtains module 51, for obtaining the voice content of interviewer's statement, and the voice content is converted to text
This content；
Word segmentation processing module 52 obtains participle database for carrying out word segmentation processing to the content of text；
Key message obtains module 53, for obtaining key message in the participle database；
Problem matching module 54, for judging current interview scene type according to the key message, and according to the face
Examination scene type choose whether to enable it is similar ask engine, it is described similar to ask engine for according to the participle term vector matching problem
The problems in database template；
Grading module 55, for start it is similar ask engine when, the processing result for asking engine according to similar obtains problem mould
Plate, and Rating Model is selected, it scores the answer of interviewee；
Do not start it is similar ask engine when, question template is obtained according to the key message, and selects Rating Model opposite
The answer of examination person is scored.
The embodiment of the present invention also provide a kind of terminal device include memory, processor and storage on a memory and can be
The computer program run on processor when the processor executes the computer program, is realized as described in embodiment one
The interview method based on AI secondary surface examination hall scape in each step.
The embodiment of the present invention also provides a kind of storage medium, and the storage medium is computer readable storage medium, thereon
It is stored with computer program, when the computer program is executed by processor, is realized auxiliary based on AI as described in embodiment one
Each step in the interview method of principal surface examination hall scape.
Embodiment described above is merely illustrative of the technical solution of the present invention, rather than its limitations；Although previous embodiment
Invention is explained in detail, those skilled in the art should understand that: it still can be to aforementioned each implementation
Technical solution documented by example is modified or equivalent replacement of some of the technical features；And these modification or
Replacement, the spirit and scope for technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution should all include
Within protection scope of the present invention.
Claims (10)
1. a kind of interview method based on AI secondary surface examination hall scape characterized by comprising
The voice content of interviewer's statement is obtained, and the voice content is converted into content of text；
Word segmentation processing is carried out to the content of text, obtains participle database；
Key message is obtained in the participle database；
Current interview scene type is judged according to the key message, and is chosen whether to enable phase according to the interview scene type
Seemingly ask engine；
Wherein, described similar to ask engine for according to participle the problems in term vector matching problem database template；
Start it is similar ask engine when, the processing result for asking engine according to similar obtains question template, and selects Rating Model pair
The answer of interviewee is scored；
Do not start it is similar ask engine when, question template is obtained according to the key message, and selects Rating Model to interviewee
Answer score.
2. the interview method as described in claim 1 based on AI secondary surface examination hall scape, which is characterized in that according to the crucial letter
The current interview scene type of breath judgement, and choose whether to enable according to the interview scene type and similar ask engine, comprising:
The problem of detecting short sentence quantity and the key message in participle database ID quantity；
The problem of key message ID quantity with it is described participle database in short sentence quantity it is identical when, then currently interview scene
Type is people's machine side examination hall scape, does not enable and similar asks engine；
The problem of key message ID quantity and it is described participle database in short sentence quantity difference when, then currently interview scene
Type is live secondary surface examination hall scape, and enabling is similar to ask engine.
3. the interview method as described in claim 1 based on AI secondary surface examination hall scape, which is characterized in that draw starting similar ask
When holding up, question template is obtained according to the similar processing result for asking engine, and Rating Model is selected to comment the answer of interviewee
Point, comprising:
Start it is similar ask engine when, call trained matching neural network model in the similar engine, pass through described
With neural network model, the key message is converted into the participle term vector；
The problems in described problem database template is matched according to the participle term vector；
Rating Model corresponding to the problem of calling successful match template carries out the answer of interviewee according to the Rating Model
Scoring.
4. the interview method as claimed in claim 3 based on AI secondary surface examination hall scape, which is characterized in that according to the participle word
The problems in Vectors matching described problem database template, comprising:
Obtain the template term vector of each question template；
The participle term vector and the template term vector are compared, cosine similarity is calculated；
According to the cosine similarity judge the participle term vector whether with described problem template matching.
5. the interview method as described in claim 1 based on AI secondary surface examination hall scape, which is characterized in that do not starting similar ask
When engine, question template is obtained according to the key message, and Rating Model is selected to score the answer of interviewee, wrapped
It includes:
Do not start it is similar ask engine when, the problems in described problem database is matched according to the problems in key message ID
Template；
Rating Model corresponding to the problem of calling successful match template carries out the answer of interviewee according to the Rating Model
Scoring.
6. the interview method as described in claim 1 based on AI secondary surface examination hall scape, which is characterized in that the Rating Model packet
Include regular Rating Model and comprehensive grade model；
Wherein, when the interview scene type is people's machine side examination hall scape, the Rating Model includes regular Rating Model；
When the interview scene is live secondary surface examination hall scape, the Rating Model includes comprehensive grade model.
7. the interview method as described in claim 1 based on AI secondary surface examination hall scape, which is characterized in that in the text
Hold and carry out word segmentation processing, obtains before segmenting database, comprising:
Noise information replacement is carried out to the content of text, filters non-critical information.
8. a kind of interview device based on AI secondary surface examination hall scape characterized by comprising
Text obtains module, for obtaining the voice content of interviewer's statement, and the voice content is converted to content of text；
Word segmentation processing module obtains participle database for carrying out word segmentation processing to the content of text；
Key message obtains module, for obtaining key message in the participle database；
Problem matching module, for judging current interview scene type according to the key message, and according to the interview scene
Type choose whether to enable it is similar ask engine, it is described similar to ask engine for according to the participle term vector matching problem database
The problems in template；
Grading module, for start it is similar ask engine when, the processing result for asking engine according to similar obtains question template, and selects
Rating Model is selected, is scored the answer of interviewee；
Do not start it is similar ask engine when, question template is obtained according to the key message, and selects Rating Model to interviewee
Answer score.
9. a kind of terminal device, which is characterized in that on a memory and can be on a processor including memory, processor and storage
The computer program of operation, which is characterized in that when the processor executes the computer program, realize such as claim 1 to 6
Each step in described in any item interview methods based on AI secondary surface examination hall scape.
10. a kind of storage medium, the storage medium is computer readable storage medium, is stored thereon with computer program,
It is characterized in that, when the computer program is executed by processor, realizes as auxiliary based on AI in as claimed in any one of claims 1 to 6
Each step in the interview method of principal surface examination hall scape.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910504604.2A CN110347787B (en) | 2019-06-12 | 2019-06-12 | Interview method and device based on AI auxiliary interview scene and terminal equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910504604.2A CN110347787B (en) | 2019-06-12 | 2019-06-12 | Interview method and device based on AI auxiliary interview scene and terminal equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110347787A true CN110347787A (en) | 2019-10-18 |
| CN110347787B CN110347787B (en) | 2022-12-09 |
Family
ID=68181789
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910504604.2A Active CN110347787B (en) | 2019-06-12 | 2019-06-12 | Interview method and device based on AI auxiliary interview scene and terminal equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110347787B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111027305A (en) * | 2019-10-29 | 2020-04-17 | 平安科技（深圳）有限公司 | Intelligent interviewing method and device based on text matching and computer equipment |
| CN111429101A (en) * | 2020-03-30 | 2020-07-17 | 广西通途科技有限公司 | Remote interview system and method based on artificial intelligence |
| CN111695591A (en) * | 2020-04-26 | 2020-09-22 | 平安科技（深圳）有限公司 | AI-based interview corpus classification method, device, computer equipment and medium |
| CN112466308A (en) * | 2020-11-25 | 2021-03-09 | 北京明略软件系统有限公司 | Auxiliary interviewing method and system based on voice recognition |
| CN113241076A (en) * | 2021-05-12 | 2021-08-10 | 北京字跳网络技术有限公司 | Voice processing method and device and electronic equipment |
| CN113470651A (en) * | 2021-06-04 | 2021-10-01 | 广东外语外贸大学 | Voice scoring method and system based on abstract extraction |
| CN113609864A (en) * | 2021-08-06 | 2021-11-05 | 珠海市鸿瑞信息技术股份有限公司 | Text semantic recognition processing system and method based on industrial control system |
| CN114741274A (en) * | 2021-01-08 | 2022-07-12 | 腾讯科技（深圳）有限公司 | Abnormity testing method and device |
| CN111695591B (en) * | 2020-04-26 | 2024-05-10 | 平安科技（深圳）有限公司 | AI-based interview corpus classification method, AI-based interview corpus classification device, AI-based interview corpus classification computer equipment and AI-based interview corpus classification medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104991965A (en) * | 2015-07-23 | 2015-10-21 | 上海智臻网络科技有限公司 | Method and apparatus for creating extended question based on standard question |
| CN106095988A (en) * | 2016-06-21 | 2016-11-09 | 上海智臻智能网络科技股份有限公司 | Automatic question-answering method and device |
| CN106649825A (en) * | 2016-12-29 | 2017-05-10 | 上海智臻智能网络科技股份有限公司 | Voice interaction system, establishment method and device thereof |
| CN107315766A (en) * | 2017-05-16 | 2017-11-03 | 广东电网有限责任公司江门供电局 | A kind of voice response method and its device for gathering intelligence and artificial question and answer |
| US20180211287A1 (en) * | 2017-01-24 | 2018-07-26 | International Business Machines Corporation | Digital content generation based on user feedback |
| CN109492854A (en) * | 2018-09-17 | 2019-03-19 | 平安科技（深圳）有限公司 | Method, apparatus, computer equipment and the storage medium of intelligent robot interview |
-
2019
- 2019-06-12 CN CN201910504604.2A patent/CN110347787B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104991965A (en) * | 2015-07-23 | 2015-10-21 | 上海智臻网络科技有限公司 | Method and apparatus for creating extended question based on standard question |
| CN106095988A (en) * | 2016-06-21 | 2016-11-09 | 上海智臻智能网络科技股份有限公司 | Automatic question-answering method and device |
| CN106649825A (en) * | 2016-12-29 | 2017-05-10 | 上海智臻智能网络科技股份有限公司 | Voice interaction system, establishment method and device thereof |
| US20180211287A1 (en) * | 2017-01-24 | 2018-07-26 | International Business Machines Corporation | Digital content generation based on user feedback |
| CN107315766A (en) * | 2017-05-16 | 2017-11-03 | 广东电网有限责任公司江门供电局 | A kind of voice response method and its device for gathering intelligence and artificial question and answer |
| CN109492854A (en) * | 2018-09-17 | 2019-03-19 | 平安科技（深圳）有限公司 | Method, apparatus, computer equipment and the storage medium of intelligent robot interview |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111027305A (en) * | 2019-10-29 | 2020-04-17 | 平安科技（深圳）有限公司 | Intelligent interviewing method and device based on text matching and computer equipment |
| CN111027305B (en) * | 2019-10-29 | 2023-12-19 | 平安科技（深圳）有限公司 | Text matching-based intelligent interview method and device and computer equipment |
| CN111429101A (en) * | 2020-03-30 | 2020-07-17 | 广西通途科技有限公司 | Remote interview system and method based on artificial intelligence |
| CN111695591A (en) * | 2020-04-26 | 2020-09-22 | 平安科技（深圳）有限公司 | AI-based interview corpus classification method, device, computer equipment and medium |
| CN111695591B (en) * | 2020-04-26 | 2024-05-10 | 平安科技（深圳）有限公司 | AI-based interview corpus classification method, AI-based interview corpus classification device, AI-based interview corpus classification computer equipment and AI-based interview corpus classification medium |
| CN112466308A (en) * | 2020-11-25 | 2021-03-09 | 北京明略软件系统有限公司 | Auxiliary interviewing method and system based on voice recognition |
| CN114741274A (en) * | 2021-01-08 | 2022-07-12 | 腾讯科技（深圳）有限公司 | Abnormity testing method and device |
| CN113241076A (en) * | 2021-05-12 | 2021-08-10 | 北京字跳网络技术有限公司 | Voice processing method and device and electronic equipment |
| CN113470651A (en) * | 2021-06-04 | 2021-10-01 | 广东外语外贸大学 | Voice scoring method and system based on abstract extraction |
| CN113609864A (en) * | 2021-08-06 | 2021-11-05 | 珠海市鸿瑞信息技术股份有限公司 | Text semantic recognition processing system and method based on industrial control system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110347787B (en) | 2022-12-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110347787A (en) | A kind of interview method, apparatus and terminal device based on AI secondary surface examination hall scape | |
| EP3862889A1 (en) | Responding to user queries by context-based intelligent agents | |
| US11704501B2 (en) | Providing a response in a session | |
| CN109388700A (en) | A kind of intension recognizing method and system | |
| CN112468659B (en) | Quality evaluation method, device, equipment and storage medium applied to telephone customer service | |
| CN110175229A (en) | A kind of method and system carrying out online training based on natural language | |
| CN110362819A (en) | Text emotion analysis method based on convolutional neural networks | |
| CN111814454A (en) | Multi-modal network spoofing detection model on social network | |
| CN110717021B (en) | Input text acquisition and related device in artificial intelligence interview | |
| CN114020906A (en) | Chinese medical text information matching method and system based on twin neural network | |
| CN112183075B (en) | Method and device for extracting slots in conversation, electronic equipment and storage medium | |
| CN109740164A (en) | Based on the matched electric power defect rank recognition methods of deep semantic | |
| CN114417097A (en) | Emotion prediction method and system based on time convolution and self-attention | |
| CN114676259B (en) | Conversation emotion recognition method based on causal perception interactive network | |
| CN110597968A (en) | Reply selection method and device | |
| CN114003682A (en) | Text classification method, device, equipment and storage medium | |
| Klaylat et al. | Enhancement of an Arabic speech emotion recognition system | |
| CN109710913A (en) | Semantic hierarchy generation method and terminal based on dependency analysis | |
| CN114003700A (en) | Method and system for processing session information, electronic device and storage medium | |
| CN115905187B (en) | Intelligent proposition system oriented to cloud computing engineering technician authentication | |
| CN113033664A (en) | Question-answering model training method, question-answering method, device, equipment and storage medium | |
| Kusuma et al. | Deep learning based facial expressions recognition system for assisting visually impaired persons | |
| Anikina et al. | Anaphora resolution in dialogue: Description of the DFKI-TalkingRobots system for the CODI-CRAC 2021 shared-task | |
| Michelbacher et al. | A cascaded classification approach to semantic head recognition | |
| Cueva et al. | Crawling to improve multimodal emotion detection |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |