CN109543020A - Inquiry handles method and system - Google Patents
Inquiry handles method and system Download PDFInfo
- Publication number
- CN109543020A CN109543020A CN201811426774.5A CN201811426774A CN109543020A CN 109543020 A CN109543020 A CN 109543020A CN 201811426774 A CN201811426774 A CN 201811426774A CN 109543020 A CN109543020 A CN 109543020A
- Authority
- CN
- China
- Prior art keywords
- question
- answer
- current
- answer pair
- complexity
- 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
- 238000000034 method Methods 0.000 title claims abstract description 40
- 230000004044 response Effects 0.000 claims abstract description 40
- 238000012545 processing Methods 0.000 claims abstract description 30
- 238000013518 transcription Methods 0.000 claims abstract description 21
- 230000035897 transcription Effects 0.000 claims abstract description 21
- 239000013598 vector Substances 0.000 claims description 118
- 238000000605 extraction Methods 0.000 claims description 67
- 238000004364 calculation method Methods 0.000 claims description 15
- 239000000284 extract Substances 0.000 claims description 12
- 230000004927 fusion Effects 0.000 claims description 11
- 238000003672 processing method Methods 0.000 claims description 11
- 238000005516 engineering process Methods 0.000 abstract description 16
- 230000006399 behavior Effects 0.000 description 20
- 230000008569 process Effects 0.000 description 17
- 238000004458 analytical method Methods 0.000 description 8
- 238000012549 training Methods 0.000 description 7
- 238000012163 sequencing technique Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 238000010276 construction Methods 0.000 description 3
- 230000001360 synchronised effect Effects 0.000 description 3
- 241001622623 Coeliadinae Species 0.000 description 2
- 208000001613 Gambling Diseases 0.000 description 2
- 230000033228 biological regulation Effects 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000013499 data model Methods 0.000 description 2
- 235000013399 edible fruits Nutrition 0.000 description 2
- 230000030279 gene silencing Effects 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 238000003058 natural language processing Methods 0.000 description 2
- 230000000750 progressive effect Effects 0.000 description 2
- 230000033764 rhythmic process Effects 0.000 description 2
- 238000000926 separation method Methods 0.000 description 2
- 238000012546 transfer Methods 0.000 description 2
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 description 1
- 241000208340 Araliaceae Species 0.000 description 1
- 238000012935 Averaging Methods 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 238000009825 accumulation Methods 0.000 description 1
- 238000005267 amalgamation Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 239000011449 brick Substances 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000005336 cracking Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000009472 formulation Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000002955 isolation Methods 0.000 description 1
- 239000010977 jade Substances 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/26—Speech to text systems
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Theoretical Computer Science (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Human Computer Interaction (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a kind of inquiries to handle method and system, wherein the treating method comprises: when carrying out inquiry to target response people, question and answer voice being carried out transcription in real time, obtains the question and answer text as unit of question and answer pair;According to the word number and sentence number of question and answer pair in the question and answer text, the complexity of current question and answer pair is calculated in real time;In conjunction with the complexity of current question and answer pair, the Text eigenvector of current question and answer pair is obtained;The Text eigenvector of the Text eigenvector of the current question and answer pair of target response people and the question and answer pair of other respondents is carried out similarity to compare;Comparison result is shown by preset demand strategy.The present invention can be realized online collaboration processing, promote the efficiency handled a case, handled official business;And due to using mature voice processing technology, can not only substantially save manpower and time cost, additionally it is possible to ensure the accuracy handled, reduce False Rate.
Description
Technical field
The present invention relates to natural language processing fields more particularly to a kind of inquiry to handle method and system.
Background technique
With the continuous breakthrough of artificial intelligence technology, the especially promotion of natural language processing the relevant technologies, natural language
Speech processing technique is applied to every field and is gradually taken seriously.Such as the inquiry often occurred during police service handles official business, handles a case
Scene can be combined with current voice processing technology.By taking the hearing for suspect as an example, police are usually needed
Hearing content is recorded, on the one hand testimonial evidence of the notes as suspect;On the other hand, especially for clique
The case that case or more suspects participate in, police by isolation method to personnel concerning the case carry out at times or it is synchronous by
A hearing, and in the later period by the record of trial for summarizing multiple personnel concerning the case, can carry out comprehensive analysis to case, such as from pen
Similarities and differences when more people answer a question are found out in record, and the break-through point etc. of cracking of cases is excavated with this.
But the means such as keyboard record, video and audio recording employed in current hearing process, it can not be while hearing
Quick comparison dynamically is carried out to current inquiry content, and the comprehensive analysis process in later period also needs police to relate to multiple
The record of trial of case personnel is manually compared, and not only program is cumbersome, but also can also while expending police strength and time resource
The artificial erroneous judgement being difficult to avoid that can occur.
Summary of the invention
For the demand, the object of the present invention is to provide a kind of inquiries to handle method and system, realizes at online collaboration
The information comparison in inquiry is managed, office, case handling efficiency and accuracy are promoted with this.
The technical solution adopted by the invention is as follows:
A kind of inquiry processing method, comprising:
When carrying out inquiry to target response people, question and answer voice is subjected to transcription in real time, is obtained as unit of question and answer pair
Question and answer text;
According to the word number and sentence number of question and answer pair in the question and answer text, the complexity of current question and answer pair is calculated in real time;
In conjunction with the complexity of current question and answer pair, the Text eigenvector of current question and answer pair is obtained;
By the text feature of the Text eigenvector of the current question and answer pair of target response people and the question and answer pair of other respondents
Vector carries out similarity comparison;
Comparison result is shown by preset demand strategy.
Optionally, the word number and sentence number according to question and answer pair in the question and answer text, calculates current question and answer in real time
Pair complexity include:
According to the word number of each question and answer pair in the question and answer text, the word complexity phase with history question and answer pair is calculated
The current word complexity of pass;
According to the sentence number of each question and answer pair in the question and answer text, the sentence complexity phase with history question and answer pair is calculated
The current sentence complexity of pass;
According to the current word complexity and the current sentence complexity, the complexity of current question and answer pair is calculated in real time
Degree.
Optionally,
The word number according to each question and answer pair in the question and answer text calculates the word complexity journey with history question and answer pair
Spending relevant current word complexity includes:
The ratio for answering word number and problem word number for calculating each question and answer centering, obtains the word of each question and answer pair
Number complexity;
The weighted sum in this inquiry from history question and answer to all word number complexities to current question and answer pair is calculated, is obtained
The current word complexity;
The sentence number according to each question and answer pair in the question and answer text calculates the sentence complexity journey with history question and answer pair
Spending relevant current sentence complexity includes:
The ratio for answering sentence number and problem sentence number for calculating each question and answer centering, obtains the sentence of each question and answer pair
Number complexity;
The weighted sum in this inquiry from history question and answer to all sentence number complexities to current question and answer pair is calculated, is obtained
The current sentence complexity.
Optionally, described according to the current word complexity and the current sentence complexity, it calculates in real time current
The complexity of question and answer pair includes:
The current word complexity is added with the current sentence complexity, obtains the words and phrases of current question and answer pair
Complexity;
The ratio for calculating the words and phrases complexity Yu preset complexity upper limit value, obtains the complexity of current question and answer pair.
Optionally, the complexity of the current question and answer pair of the combination, the Text eigenvector for obtaining current question and answer pair include:
It is obtained by real-time online feature extraction algorithm or current question and answer pair is obtained by various features extraction algorithm respectively
The problems in sentence and answer statement feature vector;Wherein, the various features extraction algorithm includes that real-time online feature mentions
Take algorithm;
The feature vector mean value of computational problem sentence and answer statement;
Complexity based on current question and answer pair is arranged weight coefficient, and by problem sentence feature vector mean value and answer statement
Feature vector mean value is weighted and sums, and obtains single or multiple foundation characteristics corresponding with feature extraction algorithm used
Vector;
The text that the single foundation characteristic vector that real-time online feature extraction algorithm is obtained is determined as current question and answer pair is special
Vector is levied, or the multiple foundation characteristic vectors obtained based on various features extraction algorithm are merged, then by fusion results
It is determined as the Text eigenvector of current question and answer pair.
Optionally,
The various features extraction algorithm further includes that professional knowledge planting modes on sink characteristic extracts model;
The method also includes:
According to the keyword of question and answer pair in the question and answer text, the knowledge base dependency degree of current question and answer pair is calculated in real time;Its
In, the keyword is the word that the business vocabulary in question and answer centering and preset professional knowledge library matches;
The multiple foundation characteristic vectors obtained based on various features extraction algorithm carry out fusion
Using the knowledge base dependency degree as weight coefficient, foundation characteristic vector relevant to professional knowledge library is added
Power;
Foundation characteristic vector relevant to professional knowledge library after weighting is merged with other foundation characteristic vectors.
Optionally, the keyword according to question and answer pair in the question and answer text, calculates the knowledge of current question and answer pair in real time
Library dependency degree includes:
The keyword number of each question and answer pair and the ratio of the question and answer centering total words are calculated, the pass of each question and answer pair is obtained
Keyword coverage rate;
The weighted sum in this inquiry from history question and answer to the keyword coverage rate to current question and answer pair is calculated, is obtained current
The knowledge base dependency degree of question and answer pair.
Optionally, described be shown comparison result by preset demand strategy includes:
The current question and answer pair of target response people and the comparison result of all question and answer pair of other respondents are ranked up,
It therefrom chooses and shows the question and answer pair for meeting other respondents of preset standard;And/or
Directly show the current question and answer pair of target response people and the asking based on same or similar problem of other respondents
The similarity comparison result answered questions.
Optionally, described that question and answer voice is carried out transcription in real time, obtaining the question and answer text as unit of question and answer pair includes:
Acquire the voice data in this inquiry;
According to acoustic feature, the voice data boundary of different speakers is divided, wherein the difference speaker includes target
Respondent and enquirement people;
Based on the voice switching or mute duration between different speakers, voice data is labeled as to be directed to different speakers
Voice segments, wherein institute's speech segment characterizes a speaker once complete voice data;
According to the time sequencing of each institute's speech segment or according to the semantic analysis result to institute's speech segment, generates and be based on one
The question and answer of secondary question and answer are to voice data;
In real time by the question and answer to voice data transcription at corresponding text data;
The text data of question and answer pair is aggregated into the question and answer text of this inquiry.
A kind of inquiry processing system, comprising:
Question and answer text generation module, for question and answer voice to be carried out transcription in real time when carrying out inquiry to target response people,
Obtain the question and answer text as unit of question and answer pair;
Complicated dynamic behaviour module calculates in real time for the word number and sentence number according to question and answer pair in the question and answer text
The complexity of current question and answer pair;
Feature vector obtains module, for combining the complexity of current question and answer pair, obtains the text feature of current question and answer pair
Vector;
Comparison module, for by the question and answer of the Text eigenvector of the current question and answer pair of target response people and other respondents
Pair Text eigenvector carry out similarity comparison;
Display module, for comparison result to be shown by preset demand strategy.
Optionally, the complicated dynamic behaviour module specifically includes:
Current word complicated dynamic behaviour unit is calculated for the word number according to each question and answer pair in the question and answer text
Current word complexity relevant to the word complexity of history question and answer pair;
Current sentence complicated dynamic behaviour unit is calculated for the sentence number according to each question and answer pair in the question and answer text
Current sentence complexity relevant to the sentence complexity of history question and answer pair;
Question and answer are to complicated dynamic behaviour unit, for complicated according to the current word complexity and the current sentence
Degree, calculates the complexity of current question and answer pair in real time.
Optionally,
The current word complicated dynamic behaviour unit specifically includes:
First computation subunit, for calculating the ratio for answering word number and problem word number of each question and answer centering,
Obtain the word number complexity of each question and answer pair;
Second computation subunit, for calculating in this inquiry from history question and answer to all word numbers to current question and answer pair
The weighted sum of complexity obtains the current word complexity;
The current sentence complicated dynamic behaviour unit specifically includes:
Third computation subunit, for calculating the ratio for answering sentence number and problem sentence number of each question and answer centering,
Obtain the sentence number complexity of each question and answer pair;
4th computation subunit, for calculating in this inquiry from history question and answer to all sentence numbers to current question and answer pair
The weighted sum of complexity obtains the current sentence complexity.
Optionally, the question and answer specifically include complicated dynamic behaviour unit:
Words and phrases complicated dynamic behaviour subelement, for carrying out the current word complexity and the current sentence complexity
It is added, obtains the words and phrases complexity of current question and answer pair;
Current question and answer are to complicated dynamic behaviour subelement, for calculating the words and phrases complexity and preset complexity upper limit value
Ratio, obtain the complexity of current question and answer pair.
Optionally, described eigenvector obtains module and specifically includes:
Feature acquiring unit, for being obtained by real-time online feature extraction algorithm or by various features extraction algorithm point
Do not obtain current question and answer to the problems in sentence and answer statement feature vector;Wherein, the various features extraction algorithm packet
Include real-time online feature extraction algorithm;
Feature vector average calculation unit, the feature vector mean value for computational problem sentence and answer statement;
Weight coefficient is arranged for the complexity based on current question and answer pair in foundation characteristic vector acquiring unit, and by problem
Sentence feature vector mean value and answer statement feature vector mean value are weighted and sum, and obtain and feature extraction algorithm used
Corresponding single or multiple foundation characteristic vectors;
Text eigenvector determination unit, the single foundation characteristic vector for obtaining real-time online feature extraction algorithm
Be determined as the Text eigenvector of current question and answer pair, or by the multiple foundation characteristics obtained based on various features extraction algorithm to
Amount is merged, then fusion results are determined as to the Text eigenvector of current question and answer pair.
Optionally,
The various features extraction algorithm further includes that professional knowledge planting modes on sink characteristic extracts model;
The system also includes:
Knowledge base dependency degree computing module calculates work as in real time for the keyword according to question and answer pair in the question and answer text
The knowledge base dependency degree of preceding question and answer pair;Wherein, the keyword is the business word in question and answer centering and preset professional knowledge library
The word that remittance matches;
The Text eigenvector determination unit specifically includes:
Subelement is weighted, is used for using the knowledge base dependency degree as weight coefficient, to base relevant to professional knowledge library
Plinth feature vector is weighted;
Subelement is merged, for the foundation characteristic vector relevant to professional knowledge library and other foundation characteristics after weighting
Vector is merged.
Optionally, the knowledge base dependency degree computing module specifically includes:
Keyword coverage rate computing unit, for calculating the keyword number and the question and answer centering total words of each question and answer pair
Ratio, obtain the keyword coverage rate of each question and answer pair;
Knowledge base dependency degree computing unit, for calculating in this inquiry from history question and answer to the key to current question and answer pair
The weighted sum of word coverage rate obtains the knowledge base dependency degree of current question and answer pair.
Optionally, the display module specifically includes:
Question and answer are to display unit, for by all question and answer pair of the current question and answer pair of target response people and other respondents
Comparison result be ranked up, therefrom choose and show the question and answer pair for meeting other respondents of preset standard;And/or
Similarity display unit, for directly show the current question and answer pair of target response people and other respondents based on
The similarity comparison result of the question and answer pair of same or similar problem.
Optionally, the question and answer text generation module specifically includes:
Voice collecting unit, for acquiring the voice data in this inquiry;
Speaker Identification unit, for dividing the voice data boundary of different speakers, wherein institute according to acoustic feature
Stating different speakers includes target response people and enquirement people;
Voice segments generation unit, for based between different speakers voice switching or mute duration, by voice data
Labeled as the voice segments for being directed to different speakers, wherein institute's speech segment characterizes a speaker once complete voice
Data;
Question and answer are to construction unit, for the time sequencing according to each institute's speech segment or according to the semanteme to institute's speech segment
Analysis is as a result, generate the question and answer based on a question and answer to voice data;
Transcription unit, in real time by the question and answer to voice data transcription at corresponding text data;
Question and answer text generation unit, for the text data of question and answer pair to be aggregated into the question and answer text of this inquiry.
The present invention is by proposing a kind of inquiry processing scheme cooperateed with online, by speech recognition technology automatically by inquiry language
Phonemic transcription is the question and answer text as unit of question and answer pair, and by way of calculating question and answer to complexity, generates and return for target
The feature vector of the current question and answer pair of people is answered, for carrying out comparing online for similarity with the question and answer situation of other staff, finally will
Comparison result is shown on demand, so as to realize online collaboration processing, promotes the efficiency handled a case, handled official business;And due to using
Mature voice processing technology, the present invention can not only substantially save manpower and time cost, additionally it is possible to ensure the accuracy handled,
Reduce False Rate.
Detailed description of the invention
To make the object, technical solutions and advantages of the present invention clearer, the present invention is made into one below in conjunction with attached drawing
Step description, in which:
Fig. 1 is the flow chart of the embodiment of inquiry processing method provided by the invention;
Fig. 2 is the flow chart of the specific embodiment provided by the invention about step S1;
Fig. 3 is the flow chart of the specific embodiment provided by the invention about step S2;
Fig. 4 is the flow chart of the specific embodiment provided by the invention about step S3;
Fig. 5 is the flow chart of the embodiment of calculation knowledge library provided by the invention dependency degree;
Fig. 6 is the block diagram of the embodiment of inquiry processing system provided by the invention.
Specific embodiment
The embodiment of the present invention is described below in detail, the example of embodiment is shown in the accompanying drawings, wherein identical from beginning to end
Or similar label indicates same or similar element or element with the same or similar functions.It is retouched below with reference to attached drawing
The embodiment stated is exemplary, and for explaining only the invention, and is not construed as limiting the claims.
The present invention is by proposing a kind of inquiry processing scheme cooperateed with online, by speech recognition technology automatically by inquiry
Phonetic transcription is the question and answer text as unit of question and answer pair, and by way of calculating question and answer to complexity, generates and be directed to target
The feature vector of the current question and answer pair of respondent, for carrying out comparing online for similarity with the question and answer situation of other staff, finally
Comparison result will be shown on demand, so as to realize online collaboration processing, promote the efficiency handled a case, handled official business.Accordingly it is found that this hair
Bright application field is not limited to the police service application illustrated above, as long as similar inquiry scene and presence and need mentioned above
Other demands being consistent are sought, those skilled in the art can give in real time using technical solution of the present invention.But for reality
Apply the explanation of mode just and also to the design for keeping those skilled in the art's intuitivism apprehension of the invention, scheme and effect
Fruit, the present invention borrow police service to apply and be specifically unfolded in detail to implementation process in this.Therefore, to embodiment party of the invention
Before formula is specifically described, first to involved in various embodiments of the present invention and preferred embodiment or the related notion that may relate to or
Background carries out description below:
It is related to one of the trial scene of more people's cases:
Theft gang member has P1 people, arranges P2 policemen simultaneously and separately tries this P1 people, and is directed to each suspicion
It doubts people and establishes a notes comprising question and answer to sequence and relevant information.
It is related to the two of the trial scene of more people's cases:
P1 people of personnel concerning the case, due to bringing a prisoner before the court or arresting time difference of appearing in court, P2 policemen may then try this at times
P1 people, and a notes comprising question and answer to sequence and relevant information are established for each personnel concerning the case.
Question and answer pair: police service hearing when, a question and answer to refer to by personnel in charge of the case it is primary put question to and a personnel concerning the case
Answer formed;Wherein, it puts question to and answer may each be one or more words.For example, asking when clique's burglary is tried
Answer questions record sample:
Question and answer are to one: puing question to 1, answer 1;
Did ask: you implement theft?
It answers: not implementing, but participated in.
Question and answer are to two: puing question to 1, answer more;
Ask: are you clear by handing over?
Answer: probably when XXXX XX o'clock morning XX month XX day, I and A from XXX come out along XXX it is overhead walk.
Get off from one outlet just into a village, because the public lavatory in my village Yao Qu is convenient.A has found the kilocalorie that is at a stop by public lavatory
Vehicle.I from lavatory come out after, A proposes that I keeps watch, he hold with a pocket knife go sled car door.After car door is removed, he
Thing is inside turned over, Cong Cheli has found several hundred yuan.After he has taken money, as soon as having divided me hundred pieces, A leaves separately with me
?.
Question and answer are to three: puing question to more, answer 1;
Ask that: XXX now informs that you steal because being accused of to you, ratify through XXX public security subbureau that decision detains three for criminal act to you
Day.Time limit from XXXX XX month XX day to XXXX XX month XX day, detainment place are the detention house XXX.Do you understand?
It answers: understanding.
Question and answer are to sequence: when referring to by trying, the question and answer for more wheel question and answer generation that personnel in charge of the case and current personnel concerning the case carry out
To sequence.
History question and answer pair: during an inquiry, in current question and answer to question and answer before to being referred to as history question and answer
It is right.
Based on above-mentioned, the present invention provides a kind of embodiments of inquiry processing method, as shown in Figure 1, this method can wrap
Include following steps:
Step S1, question and answer voice is subjected to transcription in real time, obtains the question and answer text as unit of question and answer pair;
Target response people designated herein can refer to a personnel concerning the case in aforementioned sample, to target response people
When carrying out inquiry, by mature and intelligence voice processing technology, personnel in charge of the case and case-involving people in Interrogation Procedure are realized
Question and answer voice collecting, the identifying processing of member, and so as to form the question and answer text process of this inquiry based on target response people,
It can use for reference a variety of existing speech processes solutions, it is for reference that the present invention is provided below a kind of embodiment, this
It will not go into details at place.But it should be noted that skilled person will appreciate that, can also be according to needed for scene when carrying out speech transcription
It obtains and stores to question and answer to relevant information, such as store the original question and answer voice number of this hearing in hearing database
According to, then can store in aforementioned question and answer text question and answer participant, hearing the time, case by item and each question and answer to bases such as used times
This information, without limitation to this present invention.
Step S2, according to the word number of question and answer pair in question and answer text and sentence number, the complexity of current question and answer pair is calculated in real time
Degree;
There can be different angles to the regulation direction of the feature vector of question and answer pair according to different requirements, such as to violate
Guilty psychoanalysis is foothold, then can be from exhaling when even speaking the tone in question answering process, word speed, the rhythm, dead time
Rhythm etc. is inhaled to stress to investigate.But in embodiment proposed by the present invention, combine it is related be applicable in the application experience of scene with
And the expectation to the present embodiment treatment effect, the present invention focus on to consider shadow of the complexity to subsequent comparison result of question and answer pair
It rings.Also, the complexity of question answer dialog can also reflect from a variety of dimensions, such as semantic content, key message amount and table
Logicality etc. is stated, by the feedback of a large amount of reality scenes, the word number and sentence number of discovery question and answer centering and aforementioned each complicated journey
The dimension strong correlation of degree that is to say the statement for giving expression to a large amount of semantic content or a large amount of key messages or logic complexity,
It is almost that a large amount of word number and sentence number are contained in question answering process;Accordingly, the present invention based on practical experience, selection with
The word number and sentence number of question and answer pair be emphasis, investigate question and answer pair complexity, thus it will be appreciated by those skilled in the art that
, when implementing this step, can also specifically include participle, mark and filtering stop words etc. it is conventional to speech recognition text
Pretreatment operation.
Additionally it should be noted that, since existing voice processing technology is enough to handle voice data in real time, thus this
In step, the object of computation complexity proposed by the invention is " current question and answer to ", that is, to just having sent out in this inquiry
Raw question answering process carries out the determination of complexity;Furthermore refer to that computation complexity is according to question and answer in question and answer text in this step
Pair word number and sentence number either referring to the word number and sentence number of current question and answer pair, be also possible in actual operation
Refer to entire question and answer to the word number and sentence number of sequence (from history to currently).So being calculated by word number and sentence number complicated
The process of degree can there are many modes, such as both word number and sentence number are summed;Alternatively, by all sentences by respectively containing
Word number be ranked up, select standard compliant sentence number by predetermined word number standard;Again alternatively, according to every in question and answer text
The word number of a question and answer pair calculates current word complexity relevant to the word complexity of history question and answer pair, similarly, root
According to the sentence number of each question and answer pair in question and answer text, it is multiple to calculate current sentence relevant to the sentence complexity of history question and answer pair
Miscellaneous degree calculates the complexity of current question and answer pair further according to the current word complexity and current sentence complexity in real time.This hair
It is bright to be given in embodiments below for, to relevant complexity calculating method, it will not go into details herein to history question and answer.
Step S3, in conjunction with the complexity of current question and answer pair, the Text eigenvector of current question and answer pair is obtained;
Knownly, it when carrying out text comparison operation in voice processing technology field, commonly uses and effective mode is pair
Text eigenvector carries out the comparison of special card distance, thus before carrying out subsequent comparison, it is proposed to obtain current question and answer by this step
Pair Text eigenvector, and extracting mode can then be chosen there are many prior art, such as online extract real-time or pass through
Trained Feature Selection Model etc. carries out the extraction of feature vector in advance, and a kind of preferred extraction is provided below in the present invention
The mode of feature vector, it will not go into details herein;But it is noted that the text for focusing on emphasizing the current question and answer pair of this step
Eigen vector be it is relevant to the complexity for the current question and answer pair being calculated in abovementioned steps, that is to say as illustrated above,
The present invention stresses the regulation direction by question and answer to complexity as feature representation, thus correlation designated herein, can refer to by working as
The complexity of preceding question and answer pair, specifically will be under as the weight in the Text eigenvector expression for calculating current question and answer pair
Text provides implementation reference.
Step S4, the current question and answer pair of target response people and the question and answer of other respondents are compared to similarity is carried out;
Other respondents designated herein, refer to be different from target response people described in this inquiry other by inquiry pair
As.By the explanation of above-mentioned steps, comparison process obviously refer between question and answer pair by the similarity of Text eigenvector into
Row is compared, is matched.Specifically in actual operation, it can be the Text eigenvector of the current question and answer pair based on target response people,
Loop through the question and answer of other respondents (one or more or whole, but do not include target response people) relevant to this inquiry
All or specific question and answer carry out dynamic similarity calculation to Text eigenvector in text.Here need to point out at 3 points, one,
Due to the calculating based on Text eigenvector be it will be apparent that therefore, be compared it is previous as also need to having stored
Question and answer in the question and answer text (question and answer comprising other respondents are to characteristic sequence) of other respondents are mentioned to feature vector is carried out
Extract operation, it is of course also possible to consider separately to establish the question and answer of its people to feature database, without carrying out every time for each respondent
The question and answer of other respondents are repeatedly extracted when dynamic comparison to Text eigenvector, i.e., directly by the current question and answer of target response people
Pair Text eigenvector the Text eigenvector in feature database is compared with the question and answer of other respondents.It is designated herein to ask
It answers questions feature database can be integrated with aforementioned question and answer text or to establish and be associated with, so that question and answer text all is asked with this as having added
Relevant question and answer are ask to the information recording carrier of data (the question and answer text can be considered as to a information completely comprehensive notes);
Secondly, in comparison process, can both be compared one by one, can also be based on being asked with one or more of the current question and answer to strong correlation
It answers questions and carries out specific aim comparison, such as put question to template according to preset hearing, by the corresponding question and answer of order to progress specific aim
It compares;Thirdly, due to the maturing of existing voice processing technique and network communications technology so that comparison process is enough to be useful in point
Period inquiry or synchronous inquiry.For example, since suspect appears in court, the time is different, and the hearing timing to suspect A and B is
After A elder generation B, then the current question and answer pair of B and the question and answer of the A formerly stored can be carried out sequence during inquesting B
It compares one by one or specific aim compares.If A uses synchronous inquiry with B, can be carried out efficient during inquesting A and B
It is online to intersect comparison, i.e., it can be by the history question and answer pair or current question and answer pair of the current question and answer pair of A and B during inquesting A
Real-time online comparison is carried out, similarly, the current question and answer of B are to can also be with the history of A or current question and answer to being compared;Furthermore
In intersecting comparison process if there is because progress difference or question order change, cause once to compare fail to obtain it is expected
When comparison result, it can also be repeated aforementioned by puing question to the technological means such as artificial cooperation or setting delay, timing between people
Comparison process updates comparison result with this.Specifically alignments in practical applications and used similarity calculation side
Method etc., all can be by being adjusted needed for inquiry and choosing, and this is not limited by the present invention.
Step S5, comparison result is shown by preset demand strategy.
Finally, being to be shown the comparison result in abovementioned steps to related personnel.Designated herein presses preset need
Strategy is sought, is to show for different inquiry scenes or different enquirement people, it is understood that there may be there is different expectations to comparison result,
Thus show that the mode of result can have but be not limited to following strategy:
The current question and answer pair of target response people and the comparison result of all question and answer pair of other respondents are ranked up,
It therefrom chooses and shows the question and answer pair for meeting other respondents of preset standard.In the strategy it is emphasised that sequence after selection with
And it directly shows standard compliant question and answer (can be question and answer can also be with the speech recognitions of question and answer pair text to raw tone to content
This), therefore sequence is not limited using inverted order or positive sequence;And preset standard described herein can then refer to preset different phase
Like degree thresholding, specific standards depend on different inquiry demands;And show standard compliant question and answer to being then in order to directly make phase
Pass personnel recognize the practical problem of question and answer pair and voice (or text) content of answer, so as to promoted handle a case, office efficiency.
It can be with aforementioned another exhibition strategy simultaneously or separately realized: directly showing that target response people's is current
The similarity comparison result of question and answer pair and the question and answer pair based on same or similar problem of other respondents.It is emphasized in the strategy
It is that can obtain semantic content in real time and accurately using existing voice processing technology, therefore can be by current question and answer centering
It asks a question and is carried out leading " positioning " with asking a question for the question and answer pair of other respondents, i.e., it is similar right between first decision problem
Lock onto target problem, then the similarity of the entire question and answer pair of comprehensive descision can also be related using problem certainly in other strategies
Rather than the locking of similar carry out problem, without limitation to this present invention;And the result finally shown can simplify as based on the phase
Same or Similar Problems similarity situations, such as the matching score of the two, in this way, related personnel can be rapid by matching score
Know comparison result, and specific question and answer can then carry out content information to transfer analysis in the later period, to save this inquiry
Time.
It all has his own strong points as it can be seen which kind of above-mentioned strategy no matter is taken to be shown, thus can be according to actually required selection
Its one or more progress result displaying.And it is also based on the embodiment to be expanded, such as after this inquiry,
Automatically put on record comprising question and answer text generation structured document of the question and answer to information and comparison result as filing.
The present invention proposes a kind of inquiry processing scheme cooperateed with online through the foregoing embodiment, by speech recognition technology
Automatically raw by the question and answer text that inquiry phonetic transcription is as unit of question and answer pair, and by way of calculating question and answer to complexity
At the feature vector for being directed to the current question and answer pair of target response people, for carrying out the online of similarity with the question and answer situation of other staff
It compares, will finally show comparison result on demand, so as to realize online collaboration processing, promote the efficiency handled a case, handled official business;And
Due to using mature voice processing technology, the present invention can not only substantially save manpower and time cost, additionally it is possible to ensure to locate
The accuracy of reason reduces False Rate.
About step S1, the present invention provides a kind of question and answer obtained as unit of question and answer pair preferably through speech transcription
The implementation example of text, as shown in Fig. 2, can specifically include following steps:
Step S11, the voice data in this inquiry is acquired;
Step S12, according to acoustic feature, the voice data boundary of different speakers is divided;
The voice data boundary of different speakers can be specifically identified using speaker's separation method, wherein difference is said
Words people can be feeling the pulse with the finger-tip mark respondent and put question to people.
Step S13, based on the voice switching or mute duration between different speakers, voice data is labeled as not
With the voice segments of speaker;
Here it should be noted that, is used in the present embodiment using voice switching and mute duration as the item of delimitation voice segments
Part is because the step of subsequent builds question and answer pair determines that such as computation complexity and similarity are more accurate than peering
Property.Voice switching designated herein refers to that voice source converts between enquirement people and target response people, when there are different theorys
Show that the voice of speaker before switching terminates when words human speech sound switching, can include among these the people speak terminate or by
Situations such as interrupting;Mute duration refers to after the voice of a speaker mute interval occur and be more than certain time length
Situation, such as put question to people after proposition problem when inquiry, respondent's silencing is not answered and silencing duration has been more than one default
Time threshold.Such situation puts question to people that would generally reintroduce identical or different problem, from speaker identification's angle due to two
Secondary ask several questions continuously all comes from same sound source, this voice that may be considered the same speaker is not over, so that subsequent
The building operation of question and answer pair occurs delimiting mistake, but actually previous question and answer to that may be over, (return by previous question and answer centering
Content is answered as sky), therefore the present invention is able to ascend the building accuracy of question and answer pair by this method.In this, by above two situation
One of as determine voice segments foundation, be by the voice data before switching or more than the voice number before mute duration
According to being used to characterize the voice segments of the primary complete voice data of a speaker labeled as one.It is noted that in this step
In two conditions indicated with "or" relationship, only for both illustrating that voice segments can be marked by meeting one, not represented
During complete inquiry voice segment mark can only be carried out with alternative one.
Step S14, it according to the time sequencing of each voice segments or according to the semantic analysis result to voice segments, generates and is based on one
The question and answer of secondary question and answer are to voice data;
Can be determined by the sequencing of voice segments and form asking and answering for question and answer pair, for example, chronologically have A0, Q1 and
Tri- voice segments of A1, since for time angle, answer is usually located at after enquirement, if A0 indicates previous moment respondent
Voice segments, Q1 then indicates that current time puts question to the voice segments of people, and A1 then indicates the voice segments of the respondent after Q1, therefore can
Delimiting Q1 and A1 for a question and answer pair;Alternatively, carrying out semantic analysis to voice segments, searched and problem phase according to semantic results
The two is configured to a question and answer pair accordingly by the answer of pass, and there are many prior arts to support for the specific implementation of the process, this hair
It is bright that therefore not to repeat here.
Step S15, in real time by question and answer to voice data transcription at corresponding text data;
Above-mentioned building is still the question and answer pair based on voice data, therefore in this step carries out question and answer to voice data
Transcription obtains the text data corresponding to question and answer pair, and there are many prior arts to support for specific transfer method, and the present invention is herein
It does not repeat.
Step S16, the text data of question and answer pair is aggregated into the question and answer text of this inquiry.It that is to say expression, this is asked
The question and answer text of inquiry is by multiple question and answer of target response people to being formed.
About abovementioned steps S2, the present invention provides a kind of implementations to history question and answer to relevant complexity calculating method
With reference to as shown in figure 3, the calculation method may include steps of:
Step S21, the ratio for answering word number and problem word number for calculating each question and answer centering, obtains each question and answer
Pair word number complexity;
Step S22, calculate in this inquiry from history question and answer to all word number complexities to current question and answer pair plus
Quan He obtains current word complexity;
Step S23, the ratio for answering sentence number and problem sentence number for calculating each question and answer centering, obtains each question and answer
Pair sentence number complexity;
Step S24, calculate in this inquiry from history question and answer to all sentence number complexities to current question and answer pair plus
Quan He obtains current sentence complexity;
Step S25, current word complexity is added with current sentence complexity, obtains the words and phrases of current question and answer pair
Complexity;
Step S26, the ratio for calculating words and phrases complexity and preset complexity upper limit value, obtains the complexity of current question and answer pair
Degree.
The embodiment can be exemplified below: the text of a question and answer pair be segmented and filtered the pretreatment such as stop words
Afterwards, obtaining problem word number is QW, and problem sentence number is QC, similarly, obtains answering word number being AW, answers sentence number
For AC.Thus aforementioned word number complexity V=answers word number/problem word number, the i.e. theoretical value range of V=AW/QW, V
(0, ∞), but an empirical value is preset for it in combination with practical application scene, it that is to say for a question and answer pair under the scene
Word complexity upper limit value D1.Preceding sentence number complexity W=answer sentence number/problem sentence number, i.e. W=AC/QC, W's
Theoretical value range (0, ∞), but same combinable practical application scene presets an empirical value for it, that is to say for this
The sentence complexity upper limit value D of a question and answer pair under scape2。
In the present embodiment, it is contemplated that the complexity of current question and answer pair and the history question and answer of this inquiry to related,
Be current question and answer pair complexity may be subjected to before history question and answer pair complexity influence, therefore propose calculate separately
From history question and answer to the respective weighting of all word number complexities/sentence number complexity to current question and answer pair in this inquiry
With obtain current word complexityAnd current sentence complexityWherein, i expression is asked
Answer questions sequence index (i.e. which question and answer to), i ∈ { 1,2,3 ..., N };N indicates current question and answer pair, that is to say i=1,2,
3 ..., N-1 indicates history question and answer pair;ViIndicate the word number complexity of i-th of question and answer pair;WiIndicate the sentence of i-th of question and answer pair
Subnumber complexity;Weight a and weight β is the empirical value obtained according to a large amount of practices, then by a in formulaN-i、βN-iIt is found that more
Close to current question and answer on history question and answer it is influenced it is bigger, this allow for would generally be used during inquiry it is incremental
Progressive question formulation, thus it is bigger to shared weight with history question and answer of the current question and answer to relative close.In other implementations
It can also be by the way that the different coefficient of weight be arranged, i.e., so that the weight closer to the history question and answer pair of current question and answer pair is got in example
Greatly.
Then, due to having separately designed experience upper limit value for word complexity and sentence complexity, thus it is current calculating
Following formula can be used when the complexity SC of question and answer pair:
Denominator part in formula is to indicate the complexity upper limit value of word and sentence according to same weight coefficient
Make weighted sum processing, convenient for acquiring the current word in this inquiry/sentence complexity and experience upper limit accounting, and the accounting
SC is the complexity of current question and answer pair described in the present embodiment.It can also remark additionally for above-mentioned calculation: a side
Face can also be normalized above-mentioned SC using such as tanh curve characteristic for the convenience of mathematical expression and subsequent calculating
Processing, is not construed as limiting this present invention;On the other hand, calculating step shown in Fig. 3 is practical is divided into two parts, i.e. a part is comprehensive
Having closed question and answer is multiple using current word/sentence to current word/sentence complexity, another part is acquired after the complexity of sequence
Miscellaneous degree and preset complexity upper limit value calculate the complexity of final current question and answer pair, but this two parts calculation is not only
There is above-mentioned algorithm, the calculation method based on word number and sentence number can be selected according to demand actually, such as current calculating
It may include seeking mean value when word/sentence complexity, current word complexity and word can be calculated separately when calculating SC
Upper limit ratio and current sentence complexity and sentence upper limit ratio, then again merge the two.Using different calculations
It is likely to be obtained different complexity numerical value, this depends on the actually required of scene, and aforesaid way is only to cast a brick to attract jade, this field skill
Art personnel can expand on the basis of above-mentioned implementation reference.
Need to make to illustrate about above mentioned step S3, one, because sentence compares isolated word for, can be compared with
Clearly to give expression to semanteme.Therefore, when to current question and answer to Text eigenvector extraction is carried out, preferably with current question and answer pair
The problems in sentence and answer statement be unit, carry out Text eigenvector extraction.Secondly, unquestionable, this inquiry
Content and the thing of this inquiry are by strong correlation, therefore the content of this inquiry is the critical data of core the most, but by
It is unsatisfactory for model training condition in the data sample of this inquiry, therefore preferred consideration should at least use a kind of real-time online feature
Extraction algorithm, to feature extraction operation is carried out, that is to say no matter take one or more extraction algorithms, include to current question and answer
Real-time online feature extraction algorithm, real time extracting method can be, but not limited to the word of all words by calculating each sentence
The mean value of vector obtains the Text eigenvector of each sentence.
Based on accuracy described previously and for lifting feature extraction, the present invention provides a kind of for step S3's
The concrete scheme for obtaining Text eigenvector, as shown in figure 4, may include steps of:
Step S31, by various features extraction algorithm obtain respectively current question and answer to the problems in sentence and answer statement
Feature vector;
In the present embodiment the present invention propose to extract respectively using many algorithms the feature of current each sentence of question and answer centering to
Amount, many of feature extraction algorithm not only includes aforementioned real-time online feature extraction algorithm, can also include professional knowledge
Planting modes on sink characteristic extracts model, offline big data Feature Selection Model and other feature extraction tools, and the present invention is for feature extraction work
The sum of tool is not restricted, can be two kinds, three kinds even four kinds or more in the present embodiment, but in actual operation
It also needs to consider the balance of the feature quantity extracted and operational data amount, thus preferably can be used including real-time online feature extraction
Algorithm, professional knowledge planting modes on sink characteristic extract model and offline big data Feature Selection Model, this three kinds of feature extraction tools.Its
In, the problem of being extracted by real-time online feature extraction algorithm/answer statement feature vector is represented by VECRT;The business
Knowledge base Feature Selection Model, which can be, according to each professional knowledge library (by taking police service as an example, can be such as gambling case early period
Library, larceny case library, false invoice case library etc.) data sample training the problem of obtaining, being extracted by it/answer statement spy
Sign vector is represented by VECKB;The offline big data Feature Selection Model can be used as unit of sentence, be based on word, word, word
The deep neural network model of the information such as property, the problem of being extracted by it/answer statement feature vector are represented by VECOM.Have
Closing the offline big data Feature Selection Model specifically can be such that
As unit of sentence, fusion word, word, the information such as part of speech, by the end the Encoder network of transformer+two-way
The depth network model of LSTM network+Attention Layer obtains the big of off-line training using LDA case subject classification
Data model extracts the Text eigenvector of every words;The structure of the model can be consisted of three parts mainly:
First part is embeding layer (input layer), CNN network+pre-training term vector+word of the structure based on word vector
Converge position encoded+part of speech coding etc., function is that the vocabulary vectorization of abundant sentence indicates.It realizes one side to eliminate
Influence of the unk vocabulary (non-dictionary vocabulary) to semantic meaning representation carries out unk vocabulary using the CNN network structure based on word vector
Vectorization indicates that (since word set is determining finite aggregate, and vocabulary is an open set, any one unk vocabulary
The word for being included is all in word set, it is possible to eliminate influence of the unk vocabulary to semantic meaning representation by word vector);Another party
Face extraly increases position vector based on vocabulary and part-of-speech information to word to reinforce the characterization to lexical semantic in sentence
Remittance is characterized.
The second part is sentence vector expression layer (hidden layer), and structure is mainly transformer (in machine translation
A kind of deep learning network structure) the end Encoder network+two-way LSTM network+Attention Layer (multiple groups
Attention reinforces semantic), carrying out profound semantic vectorization to sentence by this hidden layer network structure indicates.
Third part is the application (output layer) of LDA case subject classification model, and mainly the sentence vector of generation exists
It is applied in practical case data.LDA model is to carry out subject content modeling based on the case major class in existing case library,
Generate the topic model in case library.For the semantic vector that the second part generates, the space for carrying out subject content is affine, generates
The problem of theme feature vector (feature vector of i.e. each sentence) of each sentence, which makes different case classes, is generated
Semantic vector have space good discrimination.
About the training of above-mentioned model, all general hearing scenes (including clique's case, non-clique's case can be obtained
Deng) question and answer to data as data source, then by cross entropy loss function training criterion, respectively obtain hearing ask, answer to
Amount, judges whether question and answer match using fully-connected network.Training and structure composition about the model the present invention is not limited to above-mentioned,
But it may be mentioned that aforementioned professional knowledge planting modes on sink characteristic extracts model structure can adopt with the offline big data Feature Selection Model
Use approximate construction.
Step S32, the feature vector mean value of computational problem sentence and answer statement;
For the expression and subsequent calculating process convenient for holding feature vector on the whole, to said extracted to the problem of/return
The feature vector for answering sentence carries out mean value computation, need to point out if that is to say only with a kind of feature extraction tools by existing in real time
The feature vector that line feature extraction algorithm extracts, it is only necessary to which mean value is sought to the corresponding feature vector of the algorithm;It connects above, it is right
It can be respectively indicated by the feature vector mean value that aforementioned three kinds of feature extraction tools obtain as follows:
1) it is based on real-time online feature extraction algorithm:Wherein QC table
Show current question and answer to the problems in sentence sum, AC indicate the answer of current question and answer centering sentence sum.
2) model is extracted based on professional knowledge planting modes on sink characteristic:
3) it is based on offline big data Feature Selection Model:
Step S33, complexity based on current question and answer pair is arranged weight coefficient, and by problem sentence feature vector mean value and
Answer statement feature vector mean value is weighted and sums, and it is special to obtain multiple bases corresponding with feature extraction algorithm used
Levy vector;
This step combines the complexity SC for the current question and answer pair that abovementioned steps obtain, and seeks mentioning based on different characteristic respectively
The foundation characteristic vector of tool is taken, " basis " referred to herein is the differentiation table of complete Text eigenvector in opposite subsequent step
It states;It is also pointed out that the present embodiment is to highlight main function of the answer content of target response people during inquiry, for spy
The weighting scheme of sign vector mean value can be the feature vector mean value that complexity SC is assigned to answer statement, and (1-SC) is assigned
The feature vector mean value of problem sentence, but it is not limited to this weighting scheme in practical applications.It connects above, to based on aforementioned three
The foundation characteristic vector that kind feature extraction tools obtain can respectively indicate as follows:
1) real-time online feature extraction algorithm:
Wherein use VECAMP1Indicate ongoing basis feature vector.
2) professional knowledge planting modes on sink characteristic extracts model:
Wherein use VECAMP2Indicate knowledge base foundation characteristic vector.
3) offline big data Feature Selection Model:
Wherein use VECAMP3Indicate offline big data foundation characteristic vector.
Step S34, the multiple foundation characteristic vectors obtained based on various features extraction algorithm are merged, then will fusion
As a result it is determined as the Text eigenvector of current question and answer pair.
Complete Text eigenvector is expressed in order to obtain, using fusion above three foundation characteristic vector in this step
Mode obtains the Text eigenvector of current question and answer pair of the present invention:
VECAMP=VECAMP1+VECAMP2+VECAMP3
Wherein use VECAMPIndicate the Text eigenvector of current question and answer pair.Also, specific amalgamation mode can both use
Vector weighting splicing (each foundation characteristic vector dimension is identical) is also possible to vector weighted accumulation averaging etc., to this this hair
It is bright to be not construed as limiting.
But it should be noted that if that is to say only with a kind of feature extraction tools by real-time online feature extraction algorithm
Feature vector is extracted, then the Text eigenvector of current question and answer pair can be expressed as:
VECAMP=VECAMP1
Due to having used professional knowledge planting modes on sink characteristic to extract model in above-described embodiment, of the invention another preferably
In embodiment, the complexity in addition to calculating current question and answer pair is proposed, can also further calculate current question and answer in real time to knowing
Knowing library dependency degree KDD can also be to VEC so during carrying out the fusion of aforementioned base feature vectorAMP2It carries out being based on knowing
Know the control of library dependency degree.Specifically, can using the knowledge base dependency degree KDD as weight coefficient, to professional knowledge
The relevant foundation characteristic vector VEC in libraryAMP2It is weighted;Later again by the foundation characteristic relevant to professional knowledge library after weighting
Vector is merged with other foundation characteristic vectors, obtains the another kind of the Text eigenvector of current question and answer pair of the present invention
Expression-form:
VECAMP=VECAMP1+KDD×VECAMP2+VECAMP3
The calculation method of knowledge base dependency degree described in the better embodiment, the present invention provides a kind of and question and answer
The relevant reference example of the keyword of question and answer pair in text, as shown in Figure 5
Step S300, the keyword number of each question and answer pair and the ratio of the question and answer centering total words are calculated, is obtained each
The keyword coverage rate of question and answer pair;
Wherein alleged keyword is the word that the business vocabulary in question and answer centering and preset professional knowledge library matches;Tool
It, equally can be first to current question and answer to being segmented, filters the pretreatment operations such as stop words for body, and then obtain question and answer pair
Total words AQW=QW+AW.Wherein, QW is problem word number, and AW is to answer word number, and AQW is current question and answer centering list
Word sum.Then, according to existing professional knowledge library (by taking police service as an example, can be such as gambling case library, larceny case library,
False invoice case library etc.), a lists of keywords relevant to the business can be generated offline;Again by current question and answer centering AQW
A word and the lists of keywords carry out character string comparison, if character string is equal to be denoted as successful match one, finally count
All keyword numbers to match out are indicated with BW in this example.And it is directed to the keyword coverage rate P=BW/AQW, P of the business
Value range be (0,1].
Step S301, the weighting in this inquiry from history question and answer to the keyword coverage rate to current question and answer pair is calculated
With obtain the knowledge base dependency degree of current question and answer pair.
Similarly with computation complexity described previously, the present invention by entire question and answer to question and answer all in sequence to being considered in
It is interior, one is obtained to history question and answer to relevant knowledge base dependency degree, and is asked closer to the history of current question and answer pair
Answer questions its on current question and answer on knowledge base dependency degree influence it is bigger, finally obtain current question and answer pair knowledge base dependency degree indicate
ForWherein, i indicate question and answer to sequence index (i.e. which question and answer to), i ∈ 1,2,3 ...,
N};N indicates current question and answer pair, that is to say that i=1,2,3 ..., N-1 indicate history question and answer pair;PiIndicate i-th of question and answer to knowing
Know library dependency degree;Weight δ is also the empirical value obtained according to a large amount of practices.It can also remark additionally for above-mentioned calculation
It is: on the one hand, equally convenience can be provided for mathematical expression and subsequent calculate, using such as tanh curve characteristic by above-mentioned KDD
Normalized is done, this present invention is not construed as limiting;On the other hand, the side of the knowledge base dependency degree of the current question and answer pair of above-mentioned calculating
Formula is simultaneously not exclusive, can obtain the knowledge base dependency degree by a variety of calculation methods in conjunction with actually required.
Based on above content, supplementary explanation, the ratio of so-called Text eigenvector can be remake about abovementioned steps S4
To being by the VEC of the target response people of above-mentioned acquisitionAMPWith the question and answer of other respondents to the text of each question and answer pair in sequence
Feature vector VEC1, VEC2, VEC3 ... wait the calculating of progress such as COS distance or Euclidean distance;Feature comparison process non-
The emphasis of invention, when implementation, can use for reference a variety of existing alignment schemes, and the present invention does not repeat this.But it can supplement,
When the foundation characteristic vector of this question and answer pair of hind computation, including above-mentioned offline big data model foundation characteristic vector, professional knowledge
Library similarity calculation foundation characteristic vector, online question and answer are to similarity foundation characteristic vector in real time, the basic letter of this question and answer pair
Breath and foundation characteristic vector are appended to some preset database, an inquiry integrated data set as target response people;
Finally, structured document can also be accordingly generated, as the notes folder of inquiry, this folder may include each separately inquiry
Notes, the answered situation table of identical/Similar Problems difference respondent etc..
The present invention is by proposing a kind of inquiry processing scheme cooperateed with online, by speech recognition technology automatically by inquiry language
Phonemic transcription is the question and answer text as unit of question and answer pair, and by way of calculating question and answer to complexity, generates and return for target
The feature vector of the current question and answer pair of people is answered, for carrying out comparing online for similarity with the question and answer situation of other staff, finally will
Comparison result is shown on demand, so as to realize online collaboration processing, promotes the efficiency handled a case, handled official business;And due to using
Mature voice processing technology, the present invention can not only substantially save manpower and time cost, additionally it is possible to ensure the accuracy handled,
Reduce False Rate.
Corresponding to foregoing embodiments and its preferred embodiment, the present invention also provides a kind of implementation of inquiry processing system ginsengs
Examine, as shown in fig. 6, the system may include at least one be used to store dependent instruction memory and at least one for holding
The processor of the following each modules of row:
Question and answer text generation module, for question and answer voice to be carried out transcription in real time when carrying out inquiry to target response people,
Obtain the question and answer text as unit of question and answer pair;
Complicated dynamic behaviour module, for calculating in real time current according to the word number of question and answer pair in question and answer text and sentence number
The complexity of question and answer pair;
Feature vector obtains module, for combining the complexity of current question and answer pair, obtains the text feature of current question and answer pair
Vector;
Comparison module, for by the question and answer of the current question and answer pair of target response people and other respondents to carrying out similarity ratio
It is right;
Display module, for comparison result to be shown by preset demand strategy.
Optionally, complicated dynamic behaviour module specifically includes:
Current word complicated dynamic behaviour unit is calculated and is gone through for the word number according to question and answer pair each in question and answer text
The relevant current word complexity of the word complexity of history question and answer pair;
Current sentence complicated dynamic behaviour unit is calculated and is gone through for the sentence number according to question and answer pair each in question and answer text
The relevant current sentence complexity of the sentence complexity of history question and answer pair;
Question and answer are to complicated dynamic behaviour unit, for being counted in real time according to current word complexity and current sentence complexity
Calculate the complexity of current question and answer pair.
Optionally,
Current word complicated dynamic behaviour unit specifically includes:
First computation subunit, for calculating the ratio for answering word number and problem word number of each question and answer centering,
Obtain the word number complexity of each question and answer pair;
Second computation subunit, for calculating in this inquiry from history question and answer to all word numbers to current question and answer pair
The weighted sum of complexity obtains current word complexity;
Current sentence complicated dynamic behaviour unit specifically includes:
Third computation subunit, for calculating the ratio for answering sentence number and problem sentence number of each question and answer centering,
Obtain the sentence number complexity of each question and answer pair;
4th computation subunit, for calculating in this inquiry from history question and answer to all sentence numbers to current question and answer pair
The weighted sum of complexity obtains current sentence complexity.
Optionally, question and answer specifically include complicated dynamic behaviour unit:
Words and phrases complicated dynamic behaviour subelement is obtained for current word complexity to be added with current sentence complexity
To the words and phrases complexity of current question and answer pair;
Current question and answer are to complicated dynamic behaviour subelement, for calculating the ratio of words and phrases complexity and preset complexity upper limit value
Value, obtains the complexity of current question and answer pair.
Optionally, feature vector obtains module and specifically includes:
Feature acquiring unit, for being obtained by real-time online feature extraction algorithm or by various features extraction algorithm point
Do not obtain current question and answer to the problems in sentence and answer statement feature vector;Wherein, various features extraction algorithm includes real
When online feature extraction algorithm;
Feature vector average calculation unit, the feature vector mean value for computational problem sentence and answer statement;
Weight coefficient is arranged for the complexity based on current question and answer pair in foundation characteristic vector acquiring unit, and by problem
Sentence feature vector mean value and answer statement feature vector mean value are weighted and sum, and obtain and feature extraction algorithm used
Corresponding single or multiple foundation characteristic vectors;
Text eigenvector determination unit, the single foundation characteristic vector for obtaining real-time online feature extraction algorithm
Be determined as the Text eigenvector of current question and answer pair, or by the multiple foundation characteristics obtained based on various features extraction algorithm to
Amount is merged, then fusion results are determined as to the Text eigenvector of current question and answer pair.
Optionally,
Various features extraction algorithm further includes that professional knowledge planting modes on sink characteristic extracts model;
System further include:
Knowledge base dependency degree computing module calculates currently ask in real time for the keyword according to question and answer pair in question and answer text
The knowledge base dependency degree answered questions;Wherein, keyword is that question and answer centering matches with the business vocabulary in preset professional knowledge library
Word;
Text eigenvector determination unit specifically includes:
Subelement is weighted, is used for using knowledge base dependency degree as weight coefficient, it is special to basis relevant to professional knowledge library
Sign vector is weighted;
Subelement is merged, for the foundation characteristic vector relevant to professional knowledge library and other foundation characteristics after weighting
Vector is merged.
Optionally, knowledge base dependency degree computing module specifically includes:
Keyword coverage rate computing unit, for calculating the keyword number and the question and answer centering total words of each question and answer pair
Ratio, obtain the keyword coverage rate of each question and answer pair;
Knowledge base dependency degree computing unit, for calculating in this inquiry from history question and answer to the key to current question and answer pair
The weighted sum of word coverage rate obtains the knowledge base dependency degree of current question and answer pair.
Optionally, display module specifically includes:
Question and answer are to display unit, for by all question and answer pair of the current question and answer pair of target response people and other respondents
Comparison result be ranked up, therefrom choose and show the question and answer pair for meeting other respondents of preset standard;And/or
Similarity display unit, for directly show the current question and answer pair of target response people and other respondents based on
The similarity comparison result of the question and answer pair of same or similar problem.
Optionally, question and answer text generation module specifically includes:
Voice collecting unit, for acquiring the voice data in this inquiry;
Speaker Identification unit, for dividing the voice data boundary of different speakers, wherein no according to acoustic feature
It include target response people and enquirement people with speaker;
Voice segments generation unit, for based between different speakers voice switching or mute duration, by voice data
Labeled as the voice segments for being directed to different speakers, wherein a voice segments characterize a speaker once complete voice data;
Question and answer are to construction unit, for the time sequencing according to each voice segments or according to the semantic analysis knot to voice segments
Fruit generates the question and answer based on a question and answer to voice data;
Transcription unit, in real time by question and answer to voice data transcription at corresponding text data;
Question and answer text generation unit, for the text data of question and answer pair to be aggregated into the question and answer text of this inquiry.
Although the working method and technical principle of the above system embodiment and preferred embodiment are all recorded in above, still need to
, it is noted that various component embodiments of the invention can be implemented in hardware, or to transport on one or more processors
Capable software module is realized, or is implemented in a combination thereof.Can in embodiment module or unit be combined into a mould
Block or unit, also they can be divided into multiple submodules or subelement to be practiced.
And all the embodiments in this specification are described in a progressive manner, identical phase between each embodiment
As partially may refer to each other, each embodiment focuses on the differences from other embodiments.Especially for
For system embodiment, since it is substantially similar to the method embodiment, so describing fairly simple, related place is referring to method
The part of embodiment illustrates.System embodiment described above is only schematical, wherein saying as separation unit
Bright unit may or may not be physically separated, and component shown as a unit can be or can not also
It is physical unit, it can it is in one place, or may be distributed over multiple network units.It can be according to actual need
Some or all of the modules therein is selected to achieve the purpose of the solution of this embodiment.Those of ordinary skill in the art are not paying
Out in the case where creative work, it can understand and implement.
It is described in detail structure, feature and effect of the invention based on the embodiments shown in the drawings, but more than
Only presently preferred embodiments of the present invention needs to explain, technical characteristic involved in above-described embodiment and its preferred embodiment, this
Field technical staff can be under the premise of not departing from, not changing mentality of designing and technical effect of the invention, reasonably group
Conjunction mixes into a variety of equivalent schemes;Therefore, the present invention does not limit the scope of implementation as shown in the drawings, all according to conception of the invention
Made change or equivalent example modified to equivalent change, when not going beyond the spirit of the description and the drawings,
It should be within the scope of the present invention.
Claims (12)
1. a kind of inquiry processing method characterized by comprising
When carrying out inquiry to target response people, question and answer voice is subjected to transcription in real time, obtains the question and answer as unit of question and answer pair
Text;
According to the word number and sentence number of question and answer pair in the question and answer text, the complexity of current question and answer pair is calculated in real time;
In conjunction with the complexity of current question and answer pair, the Text eigenvector of current question and answer pair is obtained;
By the Text eigenvector of the Text eigenvector of the current question and answer pair of target response people and the question and answer pair of other respondents
Carry out similarity comparison;
Comparison result is shown by preset demand strategy.
2. inquiry processing method according to claim 1, which is characterized in that described according to question and answer pair in the question and answer text
Word number and sentence number, the complexity for calculating current question and answer pair in real time includes:
According to the word number of each question and answer pair in the question and answer text, calculate relevant to the word complexity of history question and answer pair
Current word complexity;
According to the sentence number of each question and answer pair in the question and answer text, calculate relevant to the sentence complexity of history question and answer pair
Current sentence complexity;
According to the current word complexity and the current sentence complexity, the complexity of current question and answer pair is calculated in real time.
3. inquiry processing method according to claim 2, which is characterized in that
The word number according to each question and answer pair in the question and answer text calculates the word complexity phase with history question and answer pair
The current word complexity of pass includes:
The ratio for answering word number and problem word number for calculating each question and answer centering, the word number for obtaining each question and answer pair are multiple
Miscellaneous degree;
The weighted sum in this inquiry from history question and answer to all word number complexities to current question and answer pair is calculated, is obtained described
Current word complexity;
The sentence number according to each question and answer pair in the question and answer text calculates the sentence complexity phase with history question and answer pair
The current sentence complexity of pass includes:
The ratio for answering sentence number and problem sentence number for calculating each question and answer centering, the sentence number for obtaining each question and answer pair are multiple
Miscellaneous degree;
The weighted sum in this inquiry from history question and answer to all sentence number complexities to current question and answer pair is calculated, is obtained described
Current sentence complexity.
4. inquiry processing method according to claim 2, which is characterized in that it is described according to the current word complexity with
And the current sentence complexity, the complexity for calculating current question and answer pair in real time include:
The current word complexity is added with the current sentence complexity, the words and phrases for obtaining current question and answer pair are complicated
Degree;
The ratio for calculating the words and phrases complexity Yu preset complexity upper limit value, obtains the complexity of current question and answer pair.
5. inquiry processing method according to claim 1, which is characterized in that the complexity of the current question and answer pair of combination,
The Text eigenvector for obtaining current question and answer pair includes:
It is obtained by real-time online feature extraction algorithm or current question and answer centering is obtained by various features extraction algorithm respectively
The feature vector of problem sentence and answer statement;Wherein, the various features extraction algorithm includes that real-time online feature extraction is calculated
Method;
The feature vector mean value of computational problem sentence and answer statement;
Complexity based on current question and answer pair is arranged weight coefficient, and by problem sentence feature vector mean value and answer statement feature
Vector mean value is weighted and sums, obtain single or multiple foundation characteristics corresponding with feature extraction algorithm used to
Amount;
The single foundation characteristic vector that real-time online feature extraction algorithm is obtained be determined as the text features of current question and answer pair to
Amount, or the multiple foundation characteristic vectors obtained based on various features extraction algorithm are merged, then fusion results are determined
For the Text eigenvector of current question and answer pair.
6. inquiry processing method according to claim 5, which is characterized in that
The various features extraction algorithm further includes that professional knowledge planting modes on sink characteristic extracts model;
The method also includes:
According to the keyword of question and answer pair in the question and answer text, the knowledge base dependency degree of current question and answer pair is calculated in real time;Wherein, institute
Stating keyword is the word that the business vocabulary in question and answer centering and preset professional knowledge library matches;
The multiple foundation characteristic vectors obtained based on various features extraction algorithm carry out fusion
Using the knowledge base dependency degree as weight coefficient, foundation characteristic vector relevant to professional knowledge library is weighted;
Foundation characteristic vector relevant to professional knowledge library after weighting is merged with other foundation characteristic vectors.
7. inquiry processing method according to claim 6, which is characterized in that described according to question and answer pair in the question and answer text
Keyword, the knowledge base dependency degree for calculating current question and answer pair in real time includes:
The keyword number of each question and answer pair and the ratio of the question and answer centering total words are calculated, the keyword of each question and answer pair is obtained
Coverage rate;
The weighted sum in this inquiry from history question and answer to the keyword coverage rate to current question and answer pair is calculated, current question and answer are obtained
Pair knowledge base dependency degree.
8. inquiry processing method according to claim 1, which is characterized in that described that comparison result is pressed preset demand plan
It is slightly shown and includes:
The current question and answer pair of target response people and the comparison result of all question and answer pair of other respondents are ranked up, therefrom
It chooses and shows the question and answer pair for meeting other respondents of preset standard;And/or
Directly show the current question and answer pair of target response people and the question and answer pair based on same or similar problem of other respondents
Similarity comparison result.
9. a kind of inquiry processing system characterized by comprising
Question and answer text generation module, for question and answer voice being carried out transcription in real time, is obtained when carrying out inquiry to target response people
Question and answer text as unit of question and answer pair;
Complicated dynamic behaviour module calculates current in real time for the word number and sentence number according to question and answer pair in the question and answer text
The complexity of question and answer pair;
Feature vector obtains module, for combining the complexity of current question and answer pair, obtains the Text eigenvector of current question and answer pair;
Comparison module, for by the question and answer pair of the Text eigenvector of the current question and answer pair of target response people and other respondents
Text eigenvector carries out similarity comparison;
Display module, for comparison result to be shown by preset demand strategy.
10. inquiry processing system according to claim 9, which is characterized in that the complicated dynamic behaviour module specifically includes:
Current word complicated dynamic behaviour unit is calculated and is gone through for the word number according to each question and answer pair in the question and answer text
The relevant current word complexity of the word complexity of history question and answer pair;
Current sentence complicated dynamic behaviour unit is calculated and is gone through for the sentence number according to each question and answer pair in the question and answer text
The relevant current sentence complexity of the sentence complexity of history question and answer pair;
Question and answer are used for according to the current word complexity and the current sentence complexity complicated dynamic behaviour unit, real
When calculate the complexities of current question and answer pair.
11. inquiry processing system according to claim 9, which is characterized in that described eigenvector obtains module and specifically wraps
It includes:
Feature acquiring unit, for being obtained by real-time online feature extraction algorithm or being obtained respectively by various features extraction algorithm
Take current question and answer to the problems in sentence and answer statement feature vector;Wherein, the various features extraction algorithm includes real
When online feature extraction algorithm;
Feature vector average calculation unit, the feature vector mean value for computational problem sentence and answer statement;
Weight coefficient is arranged for the complexity based on current question and answer pair in foundation characteristic vector acquiring unit, and by problem sentence
Feature vector mean value and answer statement feature vector mean value are weighted and sum, and obtain opposite with feature extraction algorithm used
The single or multiple foundation characteristic vectors answered;
Text eigenvector determination unit, the single foundation characteristic vector for obtaining real-time online feature extraction algorithm determine
For the Text eigenvector of current question and answer pair, or by the multiple foundation characteristic vectors obtained based on various features extraction algorithm into
Row merges, then fusion results are determined as to the Text eigenvector of current question and answer pair.
12. inquiry processing system according to claim 9, which is characterized in that the display module specifically includes:
Question and answer are to display unit, for by the ratio of the current question and answer pair of target response people and all question and answer pair of other respondents
Result is ranked up, therefrom choose and shows the question and answer pair for meeting other respondents of preset standard;And/or
Similarity display unit, for directly showing the current question and answer pair of target response people with other respondents based on identical
Or the similarity comparison result of the question and answer pair of Similar Problems.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811426774.5A CN109543020B (en) | 2018-11-27 | 2018-11-27 | Query processing method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811426774.5A CN109543020B (en) | 2018-11-27 | 2018-11-27 | Query processing method and system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109543020A true CN109543020A (en) | 2019-03-29 |
CN109543020B CN109543020B (en) | 2022-11-04 |
Family
ID=65850424
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811426774.5A Active CN109543020B (en) | 2018-11-27 | 2018-11-27 | Query processing method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109543020B (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109947927A (en) * | 2019-03-31 | 2019-06-28 | 联想(北京)有限公司 | A kind of information processing method and device |
CN110825852A (en) * | 2019-11-07 | 2020-02-21 | 四川长虹电器股份有限公司 | Long text-oriented semantic matching method and system |
CN110866103A (en) * | 2019-11-08 | 2020-03-06 | 中山大学 | Sentence diversity generation method and system in dialog system |
CN111128179A (en) * | 2019-09-12 | 2020-05-08 | 恒锋信息科技股份有限公司 | Intelligent supervision method and system based on voice recognition technology |
CN111931498A (en) * | 2020-08-14 | 2020-11-13 | 工银科技有限公司 | User online question processing method and system based on complexity analysis |
US11475068B2 (en) * | 2019-07-26 | 2022-10-18 | Beijing Boe Technology Development Co., Ltd. | Automatic question answering method and apparatus, storage medium and server |
US20230051306A1 (en) * | 2021-08-11 | 2023-02-16 | International Business Machines Corporation | Multi-user voice assistant with disambiguation |
CN117093464A (en) * | 2023-10-17 | 2023-11-21 | 青岛海尔乐信云科技有限公司 | User incoming line real-time monitoring and early warning method and system |
CN118132731A (en) * | 2024-05-06 | 2024-06-04 | 杭州数云信息技术有限公司 | Dialogue method and device, storage medium, terminal and computer program product |
Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20150120718A1 (en) * | 2013-10-24 | 2015-04-30 | International Business Machines Corporation | Smart question routing and answerer growing for online community question-answer services |
CN104679910A (en) * | 2015-03-25 | 2015-06-03 | 北京智齿博创科技有限公司 | Intelligent answering method and system |
CN105488039A (en) * | 2014-09-15 | 2016-04-13 | 华为技术有限公司 | Query method and device |
CN105824933A (en) * | 2016-03-18 | 2016-08-03 | 苏州大学 | Automatic question answering system based on main statement position and implementation method thereof |
CN106339856A (en) * | 2016-08-23 | 2017-01-18 | 王志强 | Method and system of associated generation of written record and recording terminal |
CN106528759A (en) * | 2016-11-04 | 2017-03-22 | 上海智臻智能网络科技股份有限公司 | Information processing method and device for intelligent question-answering system |
US20170249311A1 (en) * | 2016-02-26 | 2017-08-31 | Yahoo! Inc. | Quality-based scoring and inhibiting of user-generated content |
CN107885844A (en) * | 2017-11-10 | 2018-04-06 | 南京大学 | Automatic question-answering method and system based on systematic searching |
CN107895037A (en) * | 2017-11-28 | 2018-04-10 | 北京百度网讯科技有限公司 | A kind of question and answer data processing method, device, equipment and computer-readable medium |
CN107980130A (en) * | 2017-11-02 | 2018-05-01 | 深圳前海达闼云端智能科技有限公司 | It is automatic to answer method, apparatus, storage medium and electronic equipment |
CN108415980A (en) * | 2018-02-09 | 2018-08-17 | 平安科技(深圳)有限公司 | Question and answer data processing method, electronic device and storage medium |
CN108763529A (en) * | 2018-05-31 | 2018-11-06 | 苏州大学 | A kind of intelligent search method, device and computer readable storage medium |
-
2018
- 2018-11-27 CN CN201811426774.5A patent/CN109543020B/en active Active
Patent Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20150120718A1 (en) * | 2013-10-24 | 2015-04-30 | International Business Machines Corporation | Smart question routing and answerer growing for online community question-answer services |
CN105488039A (en) * | 2014-09-15 | 2016-04-13 | 华为技术有限公司 | Query method and device |
CN104679910A (en) * | 2015-03-25 | 2015-06-03 | 北京智齿博创科技有限公司 | Intelligent answering method and system |
US20170249311A1 (en) * | 2016-02-26 | 2017-08-31 | Yahoo! Inc. | Quality-based scoring and inhibiting of user-generated content |
CN105824933A (en) * | 2016-03-18 | 2016-08-03 | 苏州大学 | Automatic question answering system based on main statement position and implementation method thereof |
CN106339856A (en) * | 2016-08-23 | 2017-01-18 | 王志强 | Method and system of associated generation of written record and recording terminal |
CN106528759A (en) * | 2016-11-04 | 2017-03-22 | 上海智臻智能网络科技股份有限公司 | Information processing method and device for intelligent question-answering system |
CN107980130A (en) * | 2017-11-02 | 2018-05-01 | 深圳前海达闼云端智能科技有限公司 | It is automatic to answer method, apparatus, storage medium and electronic equipment |
CN107885844A (en) * | 2017-11-10 | 2018-04-06 | 南京大学 | Automatic question-answering method and system based on systematic searching |
CN107895037A (en) * | 2017-11-28 | 2018-04-10 | 北京百度网讯科技有限公司 | A kind of question and answer data processing method, device, equipment and computer-readable medium |
CN108415980A (en) * | 2018-02-09 | 2018-08-17 | 平安科技(深圳)有限公司 | Question and answer data processing method, electronic device and storage medium |
CN108763529A (en) * | 2018-05-31 | 2018-11-06 | 苏州大学 | A kind of intelligent search method, device and computer readable storage medium |
Non-Patent Citations (6)
Title |
---|
CHASE GEIGLE 等: "Scaling up Online Question Answering via Similar Question Retrieval", 《ACM》 * |
王益尧: "基于多特征融合问句理解问答系统研究", 《数据通信》 * |
胡云鹏等: "基于支持向量机与朴素贝叶斯的犯罪度理论研究", 《吉林大学学报(信息科学版)》 * |
许世华: "社区问答系统中问题理解的关键技术研究", 《中国优秀博硕士学位论文全文数据库(硕士)信息科技辑》 * |
许笛等: "基于用户兴趣和需求的问句推荐新方法", 《计算机应用与软件》 * |
谭红叶等: "面向阅读理解复杂问题的句子融合", 《中文信息学报》 * |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109947927A (en) * | 2019-03-31 | 2019-06-28 | 联想(北京)有限公司 | A kind of information processing method and device |
US11475068B2 (en) * | 2019-07-26 | 2022-10-18 | Beijing Boe Technology Development Co., Ltd. | Automatic question answering method and apparatus, storage medium and server |
CN111128179A (en) * | 2019-09-12 | 2020-05-08 | 恒锋信息科技股份有限公司 | Intelligent supervision method and system based on voice recognition technology |
CN110825852A (en) * | 2019-11-07 | 2020-02-21 | 四川长虹电器股份有限公司 | Long text-oriented semantic matching method and system |
CN110866103A (en) * | 2019-11-08 | 2020-03-06 | 中山大学 | Sentence diversity generation method and system in dialog system |
CN110866103B (en) * | 2019-11-08 | 2023-07-07 | 中山大学 | Sentence diversity generation method and system in dialogue system |
CN111931498B (en) * | 2020-08-14 | 2024-03-01 | 中国工商银行股份有限公司 | User online question processing method and system based on complexity analysis |
CN111931498A (en) * | 2020-08-14 | 2020-11-13 | 工银科技有限公司 | User online question processing method and system based on complexity analysis |
US20230051306A1 (en) * | 2021-08-11 | 2023-02-16 | International Business Machines Corporation | Multi-user voice assistant with disambiguation |
US11830490B2 (en) * | 2021-08-11 | 2023-11-28 | International Business Machines Corporation | Multi-user voice assistant with disambiguation |
CN117093464B (en) * | 2023-10-17 | 2024-01-26 | 青岛海尔乐信云科技有限公司 | User incoming line real-time monitoring and early warning method and system |
CN117093464A (en) * | 2023-10-17 | 2023-11-21 | 青岛海尔乐信云科技有限公司 | User incoming line real-time monitoring and early warning method and system |
CN118132731A (en) * | 2024-05-06 | 2024-06-04 | 杭州数云信息技术有限公司 | Dialogue method and device, storage medium, terminal and computer program product |
Also Published As
Publication number | Publication date |
---|---|
CN109543020B (en) | 2022-11-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109543020A (en) | Inquiry handles method and system | |
US10074363B2 (en) | Method and apparatus for keyword speech recognition | |
Tang et al. | Collaborative joint training with multitask recurrent model for speech and speaker recognition | |
US20180308487A1 (en) | Dialogue System Incorporating Unique Speech to Text Conversion Method for Meaningful Dialogue Response | |
CA2085842C (en) | Neural network-based speech token recognition system and method | |
CN112233680B (en) | Speaker character recognition method, speaker character recognition device, electronic equipment and storage medium | |
CN110347787A (en) | A kind of interview method, apparatus and terminal device based on AI secondary surface examination hall scape | |
CN113918813A (en) | Method and device for recommending posts based on external knowledge in chat record form | |
Elshaer et al. | Transfer learning from sound representations for anger detection in speech | |
Pawar et al. | Speaker Identification using Neural Networks. | |
CN109104534A (en) | A kind of system for improving outgoing call robot and being intended to Detection accuracy, recall rate | |
CN114898779A (en) | Multi-mode fused speech emotion recognition method and system | |
Ananthi et al. | Speech recognition system and isolated word recognition based on Hidden Markov model (HMM) for Hearing Impaired | |
Zavaliagkos et al. | A hybrid continuous speech recognition system using segmental neural nets with hidden Markov models | |
JP2018045364A (en) | Conversation member optimization device, conversation member optimization method, and program | |
Al-Shayea et al. | Speaker identification: A novel fusion samples approach | |
Johar | Paralinguistic profiling using speech recognition | |
Tasnia et al. | An overview of bengali speech recognition: Methods, challenges, and future direction | |
Rasipuram et al. | Improving articulatory feature and phoneme recognition using multitask learning | |
CN113724693B (en) | Voice judging method and device, electronic equipment and storage medium | |
Banjara et al. | Nepali speech recognition using cnn and sequence models | |
Chen et al. | Topic and Keyword Identification for Low-resourced Speech Using Cross-Language Transfer Learning. | |
Watada | Speech recognition in a multi-speaker environment by using hidden markov model and mel-frequency approach | |
US20220270636A1 (en) | Dialogue emotion correction method based on graph neural network | |
Kumar et al. | A novel pitch based voice recognition model (PVRM) |
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 |