CN106649561A - Intelligent question-answering system for tax consultation service - Google Patents
Intelligent question-answering system for tax consultation service Download PDFInfo
- Publication number
- CN106649561A CN106649561A CN201610990193.9A CN201610990193A CN106649561A CN 106649561 A CN106649561 A CN 106649561A CN 201610990193 A CN201610990193 A CN 201610990193A CN 106649561 A CN106649561 A CN 106649561A
- Authority
- CN
- China
- Prior art keywords
- text
- module
- word
- question
- wmd
- 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
- 238000006243 chemical reaction Methods 0.000 claims abstract description 16
- 238000000034 method Methods 0.000 claims abstract description 16
- 238000004458 analytical method Methods 0.000 claims abstract description 4
- 230000015654 memory Effects 0.000 claims description 17
- 239000013598 vector Substances 0.000 claims description 17
- 238000012549 training Methods 0.000 claims description 16
- 230000008569 process Effects 0.000 claims description 13
- 230000006870 function Effects 0.000 claims description 10
- 230000011218 segmentation Effects 0.000 claims description 8
- 238000011524 similarity measure Methods 0.000 claims description 8
- 239000011159 matrix material Substances 0.000 claims description 7
- 230000007935 neutral effect Effects 0.000 claims description 7
- 238000009826 distribution Methods 0.000 claims description 5
- 230000005540 biological transmission Effects 0.000 claims description 4
- 238000006073 displacement reaction Methods 0.000 claims description 4
- 238000007477 logistic regression Methods 0.000 claims description 4
- 230000008901 benefit Effects 0.000 claims description 3
- 230000009849 deactivation Effects 0.000 claims description 3
- 238000013461 design Methods 0.000 claims description 3
- 238000009434 installation Methods 0.000 claims description 3
- 230000001174 ascending effect Effects 0.000 claims description 2
- 238000000605 extraction Methods 0.000 claims description 2
- 230000033001 locomotion Effects 0.000 claims description 2
- 238000012216 screening Methods 0.000 claims description 2
- 230000004913 activation Effects 0.000 claims 1
- 210000005036 nerve Anatomy 0.000 claims 1
- 238000004321 preservation Methods 0.000 claims 1
- 238000005516 engineering process Methods 0.000 abstract description 6
- 238000004364 calculation method Methods 0.000 abstract description 3
- 238000013473 artificial intelligence Methods 0.000 abstract description 2
- 238000012545 processing Methods 0.000 abstract description 2
- 238000004422 calculation algorithm Methods 0.000 description 7
- 230000008859 change Effects 0.000 description 4
- 238000005457 optimization Methods 0.000 description 4
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 230000029305 taxis Effects 0.000 description 2
- 239000012141 concentrate Substances 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000009223 counseling Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/903—Querying
- G06F16/9032—Query formulation
- G06F16/90332—Natural language query formulation or dialogue systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Databases & Information Systems (AREA)
- Mathematical Physics (AREA)
- General Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Business, Economics & Management (AREA)
- Technology Law (AREA)
- Strategic Management (AREA)
- Marketing (AREA)
- Finance (AREA)
- Economics (AREA)
- Artificial Intelligence (AREA)
- Development Economics (AREA)
- Computational Linguistics (AREA)
- Accounting & Taxation (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention belongs to the field of artificial intelligence technology and particularly discloses an intelligent question-answering system for tax consultation service. The system comprises a terminal device where an Android operating system is installed and a computer; an application software program is installed in the terminal, wherein the application software comprises a voice conversion module and a question return module; a service software system is installed on the computer, wherein the service software system comprises a question understanding module and a question retrieval module; when the system operates, the voice conversion module converts voice data output by a user into text data, semantic understanding is performed through the question understanding module, the question retrieval module is used for retrieving answers, and a processing result is transmitted to the terminal user through the question return module. According to the system, voice recognition, text classification, similarity calculation and other technologies are used comprehensively to form a method for performing text similarity matching on an incomplete data set in the area of expertise, profound semantic analysis can be performed on questions raised by taxpayers, and meanwhile ceaseless accurate consultation service is provided for mass users to meet the actual requirement for tax consultation.
Description
Technical field
The invention belongs to field of artificial intelligence, and in particular to towards the intelligent Answer System of advisory tax business.
Background technology
With the fast development of 12366 service hotline more than ten years, advisory tax has become taxpayer and has understood tax law and expression
The important way of demand, but present taxation people consulting total amount rapid growth, problem complexity is constantly deepened, hot issue phase
To concentrating, only relying on original consultation way can not meet Man's Demands of paying taxes.
Tax intelligent consulting can be provided for magnanimity taxpayer and taken online incessantly as one kind application of question answering system
Business, has been increasingly becoming a kind of pattern of following consultancy service.It disclosure satisfy that government's in-depth reform, changes role, lifts service
The demand of quality, is that a kind of raising is answered accuracy and user satisfaction, lifting advisory tax efficiency, reducing tax cost and had
Effect means.At present tax intelligent consulting mainly adopts Keywords matching, lacks to the semantic understanding of problem, it is impossible to meet user's
Demand.
The content of the invention
To overcome the problem of current traditional method presence, the present invention to propose a kind of intelligent answer system of advisory tax business
System.
The intelligent Answer System of advisory tax business proposed by the present invention, using this is similar in the enterprising style of writing of Universal Database
The method of degree matching, generalization ability is excessively poor and the private database scale of construction is excessive on incomplete database to solve conventional model
Problem.
Present invention uses word displacement (Word Mover ' s Distance, a WMD) model, can pass through
Calculate the distance between two texts to calculate similarity, effectively raise to the semantic understanding of problem.Present invention uses
One long neutral net in short-term (Long-Short Term Memory, LSTM) model, first calculates again similarity to Question Classification
Method, so effectively can shorten the calculating time while Similarity Measure accuracy rate is improved.In addition to WMD models and
LSTM networks are also carried out the optimization of algorithm, greatly reduce the time complexity of algorithm.
The intelligent Answer System of the advisory tax business that the present invention is provided, including：
One installation Android operation system terminal unit is for gathering user speech problem and speech data is real-time
It is converted into the incoming computer of text data；It is additionally operable to for the answer of final matching to return to user and is shown；
One computer, for carrying out understanding retrieval to text question incoming in real time；
The installing terminal equipment has Application Software Program, and the application software includes that voice conversion module 1, problem returns mould
Block 4, for the interface for gathering user voice data, provide the user accurately problem answers, provide the user close friend；
Service software system is installed, the process software system is examined including problem Understanding Module 2, problem on the computer
Rope module 3, for carrying out semantic analysis, to understand problem, for retrieving similarity highest problem；
The speech data that user exports is converted into text data by voice conversion module 1 when system works, and is managed by problem
Solution module 2 carries out semantic understanding, using the problem retrieval retrieval answer of module 3, and is passed result by problem return module 4
It is defeated by terminal use.
Described voice conversion module 1, for completing the function that user voice signal is converted into the information of correspondence text, bag
Include by the collection to user speech, feature extraction is carried out to voice messaging, form model to be identified, and carry out with reference model
Matching, finds similarity highest model and final output recognition result；Its input is voice messaging, is output as text envelope
Breath.
Described problem Understanding Module 2, for completing the process to text message, including to be input into text carry out participle,
Text is classified, stop words for including in text etc. is deleted；Its input is question text, is output as Feature Words.
Described problem retrieval module 3, for completing to match user input problem with question and answer in tax corpus, bag
The comparison by the matching degree between two given problems is included, two some most problems of problem common trait are retrieved
Come；Its input is problem characteristic, and output is multiple similarity highest question numbers.
Described problem returns module 4, for completing the displaying to matching problem, presets return problematic amount, will
Similarity highest problem answers return to user；Its input is the question number that need to be returned, and is output as corresponding problem and answers
Case.
It is of the invention comprehensive using speech recognition technology, Text Classification, Similarity Measure technology, one kind is defined special
The method of enterprising this Similarity matching of composing a piece of writing of industry field Incomplete data set, can propose that problem carries out the semanteme of profound level to taxpayer
Analysis, and can simultaneously tackle the user of magnanimity, there is provided continual accurate consultancy service, to meet the actual need of advisory tax
Will.
In the present invention, described problem Understanding Module 2 includes Chinese word segmentation module 21, text classification module 22, goes to disable
Word module 23.Wherein, Chinese word segmentation module 21, for carrying out participle to being input into text, to determine text in the Feature Words that include；
Text classification module 22, for according to categorical data is trained, classifying to text, to improve rate of accurateness and efficiency；
Go stop words module, for deleting text in the stop words that includes, to improve system effectiveness.Chinese word segmentation module 21 determines text
In the Feature Words that include be transferred to text classification module 22；Text classification module 22 is classified to problem；Feature Words are transmitted
Processed to stop words module 23 is gone.
Described Chinese word segmentation module 21, is analyzed to sentence, should according to certain which word of understanding rule judgment
Word is grouped together into, each word of whole sentence is separated.
Described text classification module 22, by predefined classification specific feature is extracted, and sets up corresponding judgement
Rule, then treats classifying text and is classified automatically；
Described goes stop words module 23, using deactivation vocabulary set in advance, drawings word in problem is removed.
In the present invention, described text classification module 22, using long neutral net (Long-Short Term in short-term
Memory, LSTM) model.
In the present invention, described problem retrieval module 3, using word displacement (Word Mover ' s Distance,
WMD) model carries out Similarity Measure.
In the present invention, system needs to carry out the collection and training of data before input automatic running.Problem retrieves module 3
Tax question and answer corpus need to be used, it is the question and answer storehouse of the advisory tax system core, be the Data Source for answering customer problem.Ask
Topic retrieval module 3 also needs to use term vector to train storehouse, mainly completes the training to different terms feature, is follow-up similarity
Calculating is used.Text classification module 22 need to use the training set through manual sort, the feature of each classification be extracted, for complete
The classification task of paired strange problem.Stop words module 23 is gone to gather deactivation vocabulary.Training does not need extra equipment, only needs
To complete in the calculation.After completing above-mentioned initial work, system just can bring into operation.
Advantages of the present invention
The present invention effectively can be understood the problem that user proposes, and provides accurate answer in real time, with wide
Application prospect.Advisory tax demand is acutely increasing, and the present invention can simultaneously tackle the user of magnanimity, can effectively help
Work of helping others customer service shunts user.Complexity can be carried out and comprehensively be designed, overcome artificial customer service state's land tax scope of business point
Erect and put, the drawbacks of tackling is difficult to challenge.Setting up for new problem can be in advance carried out for hot issue, be prevented ahead of time
Consulting peak.The present invention uses phonetic entry pattern, is the mode that taxpayer is more happy to use.
Description of the drawings
Fig. 1 is the overall system architecture figure of the present invention.
Fig. 2 is the flow chart of the system of the present invention.
Fig. 3 is the text classification modular model figure of the present invention.
Fig. 4 is the problem retrieval modular model figure of the present invention.
Specific embodiment
The preferable embodiment of the present invention is given below according to Fig. 1-Fig. 4, and is described in detail, facilitated a better understanding of
The present invention rather than the use range for limiting the present invention.
Such as Fig. 1, to complete to propose user the demand that problem carries out effective answer, information between modules of the present invention
Transmission is that the order for returning module 4 according to voice conversion module 1, problem Understanding Module 2, problem retrieval module 3, problem is carried out
's.Wherein problem Understanding Module 2, problem retrieval module 3 is the core of whole system, by background service running software completing
Substantial amounts of evaluation work, voice conversion module 1 and problem return module 4 mainly by the operation of foreground application software to complete
Exchange with the information of user.
As Fig. 2, the idiographic flow of system are designed, main flow process is as follows：Speech data conversion to user input is written
Notebook data.Participle is carried out to input problem using participle component.The present invention employs a base class grader in classification,
Some problems for containing particular words are directly classified, its classification is directly may determine that, not requiring the use of sorting algorithm is carried out
Computing, can so accelerate classification speed and improve classification accuracy.By whether the judgement containing key word carries out split flow
Process, if do not find key word by participle data-pushing to LSTM graders, and using early stage we design LSTM models
The data result for training is classified to participle problem.Stop words is removed to participle data using vocabulary is disabled, is mainly
Improve the speed and accuracy rate of Similarity Measure.WMD Similarity Measure devices use the term vector of the training of word2Vec
Data, and the Feature Words data to removing stop words carry out Similarity Measure with the problem language material for processing, and export similarity
The ID of highest series of problems, the ID of problem are the unique numbers of the question and answer in tax corpus, and the purpose of numbering is easy for soon
It is indexed fastly.Corresponding question and answer are inquired in untreated question and answer language material according to problem ID and returns to user.
The groundwork of the present invention concentrates on text classification module 22, problem retrieval module 3, is to complete text similarity meter
The nucleus module of calculation.
Text classification module 22 has used LSTM networks, and network design memory module is used to completing to historical data
Preserve, and memory module is made up of memory cell, information can be transmitted freely in each mnemon and not receive ladder
The impact that degree disappears, these units can also be increased or remove, and memory cell is mainly by input gate, forgetting door, out gate
Several major parts are constituted.The setting of various doors is mainly used for adjusting relation of the memory cell itself and external environment condition between,
Wherein input gate mainly decides whether that the data to receiving are changed, when forgetting door mainly determines that memory cell itself is previous
Whether the state at quarter is deleted, and what out gate then affected is other neurocytes.
Mainly introduce how neurocyte i.e. memory cell is updated at each moment below, it is assumed that h is
The output of LSTM units, C is the value of LSTM mnemons, and x is input, and W is corresponding weight matrix, and σ, tanh are to activate letter
Number, b is vectorial from rate (BIAS) to take advantage of.Renewal process is described with equation below：
(1) value of moment t neurocytes is designated as
(2) value for calculating input gate is designated as it：
it=σ (Wxixt+Wcict-1+bi) (2)
(3) calculate forgetting door and be designated as f in the value of moment tt：
ft=σ (Wxfxt+Whfht-1+bf) (3)
(4) neurocyte is updated, the god after renewal in the value of moment t and the value of forgetting door with reference to neurocyte
Jing cell values are designated as Ct：
Any (5) remembered partial information calculated and is exported by the hidden layer that sigmoid functions are activated with reference to the new value of neurocyte
For ot：
ot=σ (Wx0xt+Whoht-1+Wcoct-1+bo) (5)
(6) with a tanh function to filter to final renewal determining to want final output ht：
ht=ot*tanh(Ct) (6)
Arranged by the above, LSTM networks not only overcome the problem of gradient disappearance, while it is special to be also equipped with some
Distinguished service energy, for example, preserve and read, reset and update the ability of different time historical information.
As shown in figure 3, the present invention uses a kind of change of traditional LSTM networks, in memory cell, out gate is not
Depend on memory cell CtState, this is conducive to quickly training neutral net.The calculating of the out gate that the present invention is used is public
Without W in formulahoct-1, and formula 5 is become into equation below：
ot=σ (Wx0xt+Whoht-1+bo) (7)
Different from traditional LSTM networks, the network that the present invention is used is that an average pond is increased in traditional LSTM
Change function and a logistic regression layer.The input of whole neutral net is a sentence, is made up of term vector and is designated as x0, x1, ...,
xn, these term vectors become abstract representation h after the process of LSTM networks0, h1..., hn, these abstract representations Jing again
Average pond is crossed, a vector representation for obtaining whole sentence is：
The vector representation h of final sentence is calculated final classification through logistic regression layer by softmax functions, and
Belong to the probability of each classification, computing formula is：
Problem retrieves module 3, the term vector one d × n's of composition for as shown in Figure 4 training a text and word2vec
Matrix R, wherein d are the eigenvalue numbers that text is represented using bag of words, but can be removed stop words.N is word2vec
The number of vocabulary word, what each row were represented is feature description of some word in d dimensions.Convert the text to term vector square
Battle array, then two Feature Words can be understood as two points in n-dimensional space, and between their semantic similarity and two points
Distance dependent, can be calculated with Euclidean distance.For two texts, can be regarded as two distributions, be now to
Compare the similarity degree of two texts, that is, to compare the distance between two distributions, but distance is only used for two points
Calculating, the thought of EMD distances can be applied to here the distribution of two texts.When the distribution of two texts EMD distances compared with
Hour, similarity is just high, then just can calculate the similarity for comparing text by EMD distances.
Calculated by the distance of word and word, corresponding calculating is also carried out to text and the distance of text, it is assumed that two texts
Respectively P and Q, its implication is likely to be similar, but does not but include any one identical Feature Words, then the two texts
This is present in space different regions and is distributed, it would be desirable to which finding the word of the semantic similarity in P and Q carries out turning
Change.Assume any word j in any word i and Q in P, its Euclidean distance is dij.Our target finds every a pair of words
Minimum range, because given so its Feature Words of text will be selected, by comparing P and Q in the distance of each pair word just can find
The most short word of distance, and we are converted the most short word of distance, its conversion is defined as apart from c (i, j)：
C (i, j)=| | xi-xj||2 (10)
But the Feature Words number of usual two texts is not equal, it is impossible to which paired carrying out converts, so by one
Similar to EMD apart from inner weight concept, it is assumed that word occurs in that in the text ciIt is secondary, the quality d of this wordiRepresent:
Hypothesis can obtain a word movement matrix T, wherein Tij>=0, expression is transferred to required for word j from word i
Freight volume.In order to be able to word i wholes are converted into into j, carried out by setting following condition on the basis of the hypothesis more than
Limit：
(1) each word should be transferred out of its all of quality in d, be formulated as：
(2) each word in d ' should also receive its all of quality, be formulated as：
(3) it is exactly to make the total gross traffic of transmission minimum to calculate the target of text, therefore this problem becomes a line
Property planning problem, represented with following equation：
WMD=minT≥0Σi,jTijc(i,j) (14)
The minimum scheme of linear programming can be converted the word of semantic similitude, because their distance is closer to and such as
When the quantity difference of two text words of fruit conversion causes quality different, it is approximate that unnecessary quality will be transformed into other meanings
Word on.
The present invention necessary screening is carried out to text using WCD distances and RWCD distances, it is possible to reduce finally calculating WMD away from
From workload, to reach the purpose for reducing the calculating time, the calculating to WMD simplifies.WCD distances are by each document representation
Into the average word vectors of its weighting, WMD >=WCD may certify that by following reckoning.Although the calculating of WCD is very simple
It is single, but it is nor very strict, is a kind of approximate algorithm, it is envisioned that the value of WMD distances is reduced in vector space
.The principle of RWCD distances be reduce WMD distances restrictive condition formula 12 or 13, thus can obtain two schemes L1 and
L2, defines RWCD=max (L1, L2), and do so can obtain stricter boundary.And loosen the result meeting of being possible to of condition
Reduce the value of WMD, that is to say, that WMD >=RWCD, can be proved with mathematical formulae, here it is envisioned that loosening the restriction meeting of being possible to
Cause incomplete word to transmit, often the phrase semantic of some of document word and another document closer to, by
In apart from shorter, loosen condition so that these words can repeatedly be transmitted, and some words in larger distance are not transmitted, most
Total transmission quantity can be caused eventually to diminish.
The present invention optimization realize main process be to calculate text WCD distances to be checked first, and to all WCD away from
From ascending sort is carried out, k WCD text in small distance calculates their WMD values before taking out.Next, we calculate residue
Text RWMD distances, if remaining text RWMD minima has exceeded minima in current k WMD values, then just can delete
Except remaining text, minimum WMD values are exactly what we to be looked in current K.If be not above, repeat value and operate until the above
The occurrence of.Because RWMD is a very strict boundary, RWMD about can make to delete 95% text, also
It is to say that only needing to the text to 5% has carried out WMD calculating, can save the substantial amounts of time.
While WMD algorithms are optimized, the text to needing matching have also been made optimization to the present invention.Problem s that user proposes
Length and target problem q to be matchediLength it is roughly equal, and the answer a of target problemiIt is generally longer, if directly ai
With qiIt is stitched together as a new text, with s WMD values is calculated, the calculating time can be greatly increased.Therefore herein in splicing
Before, first propose a with TF-IDFiIn several key words, then these key words and qiIt is stitched together, WMD is finally calculated again
Value.
Optimization process by more than, we are transformed in tax corpus in question and answer often by calculating user input problem
The WMD values of one problem, by dropping power sequence to WMD values, find out the ID of some minimum problems of WMD values, most corresponding at last
Question and answer returns to user.
Embodiment
Purpose：Under true counseling problem environment, problem is input in present system, the answer of complete dual problem.
The system software programming environment major parameter is as follows：The cores of CPU model Intel Core i7 tetra-；Cpu frequency 2GHz；
Internal memory 8GB；Hard disk 256G；Operating system is OSX EL Capitan 10.11.4；Development environment be Android Studio,
PyCharm、IntelliJ IDEA；Data are mainly with the storage of JSON, bin, txt form；Programming language service software is used
Python writes, and application software is write using java.
Software deployment major parameter is as follows：CPU model Intel i3-2120 double-cores；Cpu frequency 3.30GHz；Internal memory 4GB；
Hard disk 500G；Operating system is 10 professional versions of Windows 32；Deployed environment background service uses Python2.7.3,
The primary clustering of installation has gensim0.12.4, jieba0.38, numpy1.9.1, PuLP1.5.3, scikit-
learn0.14.1、six1.10.0、Theano0.8.2、smart_open1.3.3；Deployed environment foreground uses
Genymotion2.6.0 and VirtualBox5.0.20 carries out the simulation of Android system.
Voice conversion module provides online service using the speech recognition of Xun Fei company limiteies of University of Science and Technology research and development.
Chinese word segmentation module uses stammerer participle (jieba) component of Python.
Text classification module uses long neutral net LSTM model in short-term.Training uses the training set of manual sort, including
Tax file, tax paper, local government's file, tax news etc..
Problem retrieval module uses word displacement WMD model.Training using search dog the whole network news data carry out word to
Amount training, using the word2vec softwares of instrument Google companies, the major parameter that word2vec softwares are used is model selection
Be Skip-gram models, dimension selects that 250 dimensions are excellent, and change algorithms selection is Hierarchical Softmax algorithms.
Problem returns module and arranges return problematic amount 10.
Core tax corpus uses the question and answer storehouse in 12366 paying taxes service hot line tax revenue professional knowledge storehouses.
The Internet Problems that problem base is used be Baidu know, 360 question and answer, search dog ask in three question answering systems with the tax
Similar problem in corpus.When 10 problems are returned, average accuracy is 74.49%.
Claims (9)
1. a kind of intelligent Answer System towards advisory tax business, it is characterised in that include：
One installation Android operation system terminal unit, for gathering user speech problem, and speech data is converted in real time
For the incoming computer of text data；It is additionally operable to for the answer of final matching to return to user and is shown；
One computer, for carrying out understanding retrieval to text question incoming in real time；
The installing terminal equipment has Application Software Program, and the application software includes that voice conversion module 1, problem returns module 4,
For the interface for gathering user voice data, provide the user accurately problem answers, provide the user close friend；
Service software system is installed, the process software system includes that problem Understanding Module 2, problem retrieves mould on the computer
Block 3, for carrying out semantic analysis, to understand problem, for retrieving similarity highest problem；
The speech data that user exports is converted into text data by voice conversion module 1 when system works, and by problem mould is understood
Block 2 carries out semantic understanding, using the problem retrieval retrieval answer of module 3, and is transferred to result by problem return module 4
Terminal use.
2. the intelligent Answer System towards advisory tax business according to claim 1, it is characterised in that：
Described voice conversion module 1, for completing the function that user voice signal is converted into the information of correspondence text, including logical
The collection to user speech is crossed, feature extraction is carried out to voice messaging, model to be identified is formed, and is carried out with reference model
Match somebody with somebody, find similarity highest model and final output recognition result；Its input is voice messaging, is output as text message；
Described problem Understanding Module 2, for completing the process to text message, including to be input into text carry out participle, to text
Originally classified, deleted stop words for including in text etc.；Its input is question text, is output as Feature Words；
Described problem retrieval module 3, for completing to match user input problem with question and answer in tax corpus, including leading to
The comparison of the matching degree crossed between two given problems, two some most problems of problem common trait are retrieved；
Its input is problem characteristic, and output is multiple similarity highest question numbers；
Described problem returns module 4, for completing the displaying to matching problem, presets return problematic amount, will be similar
Degree highest problem answers return to user；Its input is the question number that need to be returned, and is output as corresponding problem and answer.
3. the intelligent Answer System towards advisory tax business according to claim 2, it is characterised in that described problem
Understanding Module 2 includes Chinese word segmentation module 21, text classification module 22, goes stop words module 23；Wherein, Chinese word segmentation module 21
For carrying out participle to being input into text, to determine text in the Feature Words that include；Text classification module 22 is used for basis and trains
Categorical data, classifies to text；Go stop words module, for deleting text in the stop words that includes；Chinese word segmentation module
The Feature Words included in 21 determination texts are transferred to text classification module 22；Text classification module 22 is classified to problem；Will
Feature Words are transferred to stop words module 23 and are processed.
4. the intelligent Answer System towards advisory tax business according to claim 3, it is characterised in that described text
Sort module 22, using long neutral net (LSTM) model in short-term；
Described problem retrieval module 3, using word displacement (WMD) model Similarity Measure is carried out.
5. the intelligent Answer System towards advisory tax business according to one of claim 1-4, it is characterised in that system
Need to carry out the collection and training of data before input automatic running；Problem retrieval module 3 need to use tax question and answer corpus,
It is the question and answer storehouse of the advisory tax system core, is the Data Source for answering customer problem；Problem retrieval module 3 also needs to use
Term vector trains storehouse, mainly completes the training to different terms feature, is that follow-up Similarity Measure is used；Text classification module
22 need to use the training set through manual sort, extract the feature of each classification, appoint for completing the classification to strange problem
Business；Stop words module 23 is gone to gather deactivation vocabulary.
6. the intelligent Answer System towards advisory tax business according to claim 5, it is characterised in that text classification mould
Block 22 uses LSTM networks, the network design to have memory module for completing the preservation to historical data, and memory module is by remembering
Recall cellularity, information can be transmitted freely in each mnemon and not disappeared by gradient is affected；The memory is thin
Born of the same parents are mainly made up of input gate, forgetting door, out gate；The setting of various doors is mainly used for adjusting memory cell itself and outside
Relation between environment, wherein input gate mainly decide whether that the data to receiving are changed, and forget door and mainly determine note
Whether the state for recalling cell itself previous moment is deleted, and what out gate then affected is other neurocytes.
7. the intelligent Answer System towards advisory tax business according to claim 6, it is characterised in that the memory is thin
Born of the same parents are in the mode that each moment is updated：
Assume h for LSTM units output, C for LSTM mnemons value, x for input, W be corresponding weight matrix, σ,
Tanh is activation primitive, and b is vectorial from rate (BIAS) to take advantage of；Renewal process is described with equation below：
(1) value of moment t neurocytes is designated as
(2) value for calculating input gate is designated as it：
it=σ (Wxixt+Wcict-1+bi) (2)
(3) calculate forgetting door and be designated as f in the value of moment tt：
ft=σ (Wxfxt+Whfht-1+bf) (3)
(4) with reference to neurocyte neurocyte is updated in the value of moment t and the value of forgetting door, the nerve after renewal is thin
Born of the same parents' value is designated as Ct：
(5) hidden layer activated by sigmoid functions with reference to the new value of neurocyte is designated as o calculating which partial information outputt：
ot=σ (Wx0xt+Whoct-1+bo) (5)
(6) with a tanh function to filter to final renewal determining to want final output ht：
ht=ot*tanh(Ct) (6)。
8. the intelligent Answer System towards advisory tax business according to claim 7, it is characterised in that traditional
Increase an average pond function and a logistic regression layer on LSTM network foundations；The input of whole neutral net is a sentence
Son, is made up of term vector, is designated as x0, x1, ..., xn, these term vectors become abstract representation after the process of LSTM networks
h0, h1..., hn, again through average pond, a vector representation for obtaining whole sentence is these abstract representations：
The vector representation h of final sentence is calculated final classification through logistic regression layer by softmax functions, and belongs to
The probability of each classification, computing formula is：
9. the intelligent Answer System towards advisory tax business according to claim 1,2,3,4,6,7 or 8, its feature exists
In in problem retrieval module 3, if a text constitutes the matrix R, wherein d of a d × n with the term vector of word2vec training
It is eigenvalue number that text is represented using bag of words, but stop words can be removed, n is word2vec vocabulary words
Number, each row represent feature description of some word in d dimensions；Convert the text to term vector matrix, then two features
Word can be understood as two points in n-dimensional space, and their semantic similarity is calculated using Euclidean distance；For two texts
For, regarded as two distributions, the similarity of two texts is calculated using EMD distances；
Calculated by the distance of word and word, corresponding calculating is also carried out to text and the distance of text, it is assumed that two text difference
For P and Q, its implication is likely to be similar, but does not but include any one identical Feature Words, then the two texts exist
Different regions are present in space to be distributed, and need the word for finding the semantic similarity in P and Q to be changed；In assuming P
Any word j arbitrarily in word i and Q, its Euclidean distance is dij；Target is to find the minimum range of every a pair of words；By than
The most short word of distance can be just found compared with the distance of each pair word in P and Q, and the most short word of distance is converted, its conversion is apart from c
(i, j) is defined as：
C (i, j)=| | xi-xj||2 (10)
But the Feature Words number of usual two texts is not equal, it is impossible to which paired carrying out converts, so similar by one
In EMD apart from inner weight concept, it is assumed that word occurs in that in the text ciIt is secondary, the quality d of this wordiRepresent:
Hypothesis can obtain a word movement matrix T, wherein Tij>=0, expression is transferred to the fortune required for word j from word i
Throughput rate；In order to be able to word i wholes are converted into into j, limited by setting following condition on the basis of the hypothesis more than：
(1) each word should be transferred out of its all of quality in d, be formulated as：
(2) each word in d ' should also receive its all of quality, be formulated as：
(3) target for calculating text is exactly to make the total gross traffic of transmission minimum, and this problem becomes asking for linear programming
Topic, is represented with following equation：
WMD=minT≥0∑i,jTijc(i,j) (14)
The minimum scheme of linear programming can be converted the word of semantic similitude, if the quantity of two text words of conversion
When difference causes quality different, unnecessary quality will be transformed on the approximate word of other meanings；
RWCD=max (L1, L2) is defined, is referred to as the documentation center distance loosened；Here, L1 is public for the restrictive condition of WMD distances
The scheme of formula (12), L2 is the scheme of the restrictive condition formula (13) of WMD distances；
Necessary screening is carried out to text using RWCD distances, detailed process is to calculate text WCD distances to be checked first,
And all WCD values are carried out with ascending sort, take out the less text of front k WCD value and calculate their WMD values；Next, calculating
Remaining text RWMD values, if residue text RWMD minima exceedes minima in current k WMD values, then just delete surplus
Remaining text；At present WMD values minimum in K seek to what is looked for；If remaining text RWMD minima is not above current k WMD
Minima in value, then repeatedly value operation, till the generation of case above；
The WMD values of each problem in question and answer in tax corpus are so transformed into by calculating user input problem, by right
The drop power sequence of WMD values, finds out the ID of some minimum problems of WMD values, and most at last corresponding question and answer returns to user.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610990193.9A CN106649561B (en) | 2016-11-10 | 2016-11-10 | Intelligent question-answering system for tax consultation service |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610990193.9A CN106649561B (en) | 2016-11-10 | 2016-11-10 | Intelligent question-answering system for tax consultation service |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106649561A true CN106649561A (en) | 2017-05-10 |
| CN106649561B CN106649561B (en) | 2020-05-26 |
Family
ID=58806015
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610990193.9A Expired - Fee Related CN106649561B (en) | 2016-11-10 | 2016-11-10 | Intelligent question-answering system for tax consultation service |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106649561B (en) |
Cited By (51)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107239574A (en) * | 2017-06-29 | 2017-10-10 | 北京神州泰岳软件股份有限公司 | A kind of method and device of intelligent Answer System knowledge problem matching |
| CN107368548A (en) * | 2017-06-30 | 2017-11-21 | 华中科技大学鄂州工业技术研究院 | Intelligent government affairs service interaction method and system |
| CN107391614A (en) * | 2017-07-04 | 2017-11-24 | 重庆智慧思特大数据有限公司 | A kind of Chinese question and answer matching process based on WMD |
| CN107679234A (en) * | 2017-10-24 | 2018-02-09 | 上海携程国际旅行社有限公司 | Customer service information providing method, device, electronic equipment, storage medium |
| CN107798116A (en) * | 2017-11-07 | 2018-03-13 | 盐城工学院 | Traffic rules answering method and device |
| CN108133297A (en) * | 2018-01-26 | 2018-06-08 | 广州大学 | Dissolved oxygen prediction method and system based on shot and long term memory network |
| CN108153868A (en) * | 2017-12-25 | 2018-06-12 | 安徽磐众信息科技有限公司 | A kind of voice encyclopaedia for option trade answers machine |
| CN108595696A (en) * | 2018-05-09 | 2018-09-28 | 长沙学院 | A kind of human-computer interaction intelligent answering method and system based on cloud platform |
| CN108763477A (en) * | 2018-05-29 | 2018-11-06 | 厦门快商通信息技术有限公司 | A kind of short text classification method and system |
| CN108777751A (en) * | 2018-06-07 | 2018-11-09 | 上海航动科技有限公司 | A kind of call center system and its voice interactive method, device and equipment |
| CN108804492A (en) * | 2018-03-27 | 2018-11-13 | 优视科技新加坡有限公司 | The method and device recommended for multimedia object |
| CN108846035A (en) * | 2018-05-26 | 2018-11-20 | 许崇兰 | A kind of intelligence advisory tax system |
| CN108922634A (en) * | 2018-06-13 | 2018-11-30 | 平安医疗科技有限公司 | The problem of based on online interrogation, replies processing method, device and computer equipment |
| CN108959396A (en) * | 2018-06-04 | 2018-12-07 | 众安信息技术服务有限公司 | Machine reading model training method and device, answering method and device |
| WO2018224034A1 (en) * | 2017-06-09 | 2018-12-13 | 腾讯科技（深圳）有限公司 | Intelligent question answering method, server, terminal and storage medium |
| CN109165291A (en) * | 2018-06-29 | 2019-01-08 | 厦门快商通信息技术有限公司 | A kind of text matching technique and electronic equipment |
| CN109241240A (en) * | 2018-08-17 | 2019-01-18 | 国家电网有限公司客户服务中心 | Power failure repairing information automatically forwarding method |
| CN109359302A (en) * | 2018-10-26 | 2019-02-19 | 重庆大学 | A kind of optimization method of field term vector and fusion sort method based on it |
| CN109410911A (en) * | 2018-09-13 | 2019-03-01 | 何艳玲 | Artificial intelligence learning method based on speech recognition |
| CN109446314A (en) * | 2018-11-14 | 2019-03-08 | 沈文策 | A kind of customer service question processing method and device |
| CN109493186A (en) * | 2018-11-20 | 2019-03-19 | 北京京东尚科信息技术有限公司 | The method and apparatus for determining pushed information |
| CN109657037A (en) * | 2018-12-21 | 2019-04-19 | 焦点科技股份有限公司 | A kind of knowledge mapping answering method and system based on entity type and semantic similarity |
| CN109684475A (en) * | 2018-11-21 | 2019-04-26 | 斑马网络技术有限公司 | Processing method, device, equipment and the storage medium of complaint |
| CN109726372A (en) * | 2017-10-31 | 2019-05-07 | 百度在线网络技术（北京）有限公司 | Generation method, device and the computer-readable medium of work order based on message registration |
| CN109739985A (en) * | 2018-12-26 | 2019-05-10 | 斑马网络技术有限公司 | Automatic document classification method, equipment and storage medium |
| CN109766421A (en) * | 2018-12-28 | 2019-05-17 | 上海汇付数据服务有限公司 | Intelligent Answer System and method |
| CN109947909A (en) * | 2018-06-19 | 2019-06-28 | 平安科技（深圳）有限公司 | Intelligent customer service answer method, equipment, storage medium and device |
| CN109993507A (en) * | 2019-04-17 | 2019-07-09 | 王亚萍 | A kind of pilot-operated type service system of paying taxes |
| CN109993649A (en) * | 2019-03-13 | 2019-07-09 | 王亚萍 | A kind of pilot-operated type consultancy relating to payment of tax method for pushing |
| CN110019710A (en) * | 2017-11-27 | 2019-07-16 | 厦门快商通信息技术有限公司 | A kind of topic forest formula interactive method and system |
| CN110113422A (en) * | 2019-05-10 | 2019-08-09 | 南京硅基智能科技有限公司 | A kind of intension recognizing method and system of the virtual mobile phone based on cloud |
| CN110334110A (en) * | 2019-05-28 | 2019-10-15 | 平安科技（深圳）有限公司 | Natural language classification method, device, computer equipment and storage medium |
| CN110489533A (en) * | 2019-07-09 | 2019-11-22 | 深圳追一科技有限公司 | Interactive method and relevant device |
| CN110555343A (en) * | 2018-06-01 | 2019-12-10 | 北京师范大学 | method and system for extracting three elements of forest, shrub and grass in typical resource elements |
| CN110727778A (en) * | 2019-10-15 | 2020-01-24 | 大连中河科技有限公司 | Intelligent question-answering system for tax affairs |
| CN110827797A (en) * | 2019-11-06 | 2020-02-21 | 北京沃东天骏信息技术有限公司 | Voice response event classification processing method and device |
| CN110909142A (en) * | 2019-11-20 | 2020-03-24 | 腾讯科技（深圳）有限公司 | Question and sentence processing method and device of question-answer model, electronic equipment and storage medium |
| CN111221950A (en) * | 2019-12-30 | 2020-06-02 | 航天信息股份有限公司 | Method and device for analyzing weak emotion of user |
| CN111275536A (en) * | 2020-01-13 | 2020-06-12 | 山东浪潮商用系统有限公司 | Tax intelligent consultation method and system integrating manual intervention |
| CN111581338A (en) * | 2020-03-25 | 2020-08-25 | 北京市农林科学院 | Agricultural technical service robot man-machine fusion consultation question-answering method and system |
| CN111626684A (en) * | 2019-02-28 | 2020-09-04 | 百度在线网络技术（北京）有限公司 | Intelligent tax processing method, device, terminal and medium |
| CN111626686A (en) * | 2019-02-28 | 2020-09-04 | 百度在线网络技术（北京）有限公司 | Intelligent tax processing method, device, terminal and medium |
| CN111666381A (en) * | 2020-06-17 | 2020-09-15 | 中国电子科技集团公司第二十八研究所 | Task type question-answer interaction system oriented to intelligent control |
| CN111680089A (en) * | 2017-09-19 | 2020-09-18 | 广州市妇女儿童医疗中心 | Text structuring method, device and system and non-volatile storage medium |
| CN112599120A (en) * | 2020-12-11 | 2021-04-02 | 上海中通吉网络技术有限公司 | Semantic determination method and device based on user-defined weighted WMD algorithm |
| CN112861757A (en) * | 2021-02-23 | 2021-05-28 | 天津汇智星源信息技术有限公司 | Intelligent record auditing method based on text semantic understanding and electronic equipment |
| US11222054B2 (en) | 2018-03-12 | 2022-01-11 | International Business Machines Corporation | Low-complexity methods for assessing distances between pairs of documents |
| CN114138949A (en) * | 2021-10-28 | 2022-03-04 | 数税(江苏)信息科技有限公司 | Tax intelligent question-answering customer service processing system |
| CN114297501A (en) * | 2021-12-29 | 2022-04-08 | 中国建设银行股份有限公司 | Text recommendation method, device, equipment and medium |
| CN117041618A (en) * | 2023-10-10 | 2023-11-10 | 北京装库创意科技有限公司 | Intelligent voice customer service method and system for electronic commerce |
| CN117198290A (en) * | 2023-11-06 | 2023-12-08 | 深圳市金鼎胜照明有限公司 | Acoustic control-based multi-mode LED intelligent control method and apparatus |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060235689A1 (en) * | 2005-04-13 | 2006-10-19 | Fuji Xerox Co., Ltd. | Question answering system, data search method, and computer program |
| CN101566998A (en) * | 2009-05-26 | 2009-10-28 | 华中师范大学 | Chinese question-answering system based on neural network |
| CN103198155A (en) * | 2013-04-27 | 2013-07-10 | 俞志晨 | Mobile terminal based smart question answering interaction system and method |
-
2016
- 2016-11-10 CN CN201610990193.9A patent/CN106649561B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060235689A1 (en) * | 2005-04-13 | 2006-10-19 | Fuji Xerox Co., Ltd. | Question answering system, data search method, and computer program |
| CN101566998A (en) * | 2009-05-26 | 2009-10-28 | 华中师范大学 | Chinese question-answering system based on neural network |
| CN103198155A (en) * | 2013-04-27 | 2013-07-10 | 俞志晨 | Mobile terminal based smart question answering interaction system and method |
Cited By (66)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018224034A1 (en) * | 2017-06-09 | 2018-12-13 | 腾讯科技（深圳）有限公司 | Intelligent question answering method, server, terminal and storage medium |
| CN107239574B (en) * | 2017-06-29 | 2018-11-02 | 北京神州泰岳软件股份有限公司 | A kind of intelligent Answer System knowledge-matched method and device of problem |
| CN107239574A (en) * | 2017-06-29 | 2017-10-10 | 北京神州泰岳软件股份有限公司 | A kind of method and device of intelligent Answer System knowledge problem matching |
| CN107368548A (en) * | 2017-06-30 | 2017-11-21 | 华中科技大学鄂州工业技术研究院 | Intelligent government affairs service interaction method and system |
| CN107391614A (en) * | 2017-07-04 | 2017-11-24 | 重庆智慧思特大数据有限公司 | A kind of Chinese question and answer matching process based on WMD |
| CN111680089A (en) * | 2017-09-19 | 2020-09-18 | 广州市妇女儿童医疗中心 | Text structuring method, device and system and non-volatile storage medium |
| CN111680089B (en) * | 2017-09-19 | 2023-03-21 | 广州市妇女儿童医疗中心 | Text structuring method, device and system and non-volatile storage medium |
| CN111680090B (en) * | 2017-09-19 | 2023-03-21 | 广州市妇女儿童医疗中心 | Text structuring method, device and system and non-volatile storage medium |
| CN111680090A (en) * | 2017-09-19 | 2020-09-18 | 广州市妇女儿童医疗中心 | Text structuring method, device and system and non-volatile storage medium |
| CN107679234A (en) * | 2017-10-24 | 2018-02-09 | 上海携程国际旅行社有限公司 | Customer service information providing method, device, electronic equipment, storage medium |
| CN107679234B (en) * | 2017-10-24 | 2020-02-11 | 上海携程国际旅行社有限公司 | Customer service information providing method, customer service information providing device, electronic equipment and storage medium |
| CN109726372A (en) * | 2017-10-31 | 2019-05-07 | 百度在线网络技术（北京）有限公司 | Generation method, device and the computer-readable medium of work order based on message registration |
| CN107798116A (en) * | 2017-11-07 | 2018-03-13 | 盐城工学院 | Traffic rules answering method and device |
| CN110019710A (en) * | 2017-11-27 | 2019-07-16 | 厦门快商通信息技术有限公司 | A kind of topic forest formula interactive method and system |
| CN108153868A (en) * | 2017-12-25 | 2018-06-12 | 安徽磐众信息科技有限公司 | A kind of voice encyclopaedia for option trade answers machine |
| CN108133297A (en) * | 2018-01-26 | 2018-06-08 | 广州大学 | Dissolved oxygen prediction method and system based on shot and long term memory network |
| US11222054B2 (en) | 2018-03-12 | 2022-01-11 | International Business Machines Corporation | Low-complexity methods for assessing distances between pairs of documents |
| CN108804492A (en) * | 2018-03-27 | 2018-11-13 | 优视科技新加坡有限公司 | The method and device recommended for multimedia object |
| CN108595696A (en) * | 2018-05-09 | 2018-09-28 | 长沙学院 | A kind of human-computer interaction intelligent answering method and system based on cloud platform |
| CN108846035A (en) * | 2018-05-26 | 2018-11-20 | 许崇兰 | A kind of intelligence advisory tax system |
| CN108763477A (en) * | 2018-05-29 | 2018-11-06 | 厦门快商通信息技术有限公司 | A kind of short text classification method and system |
| CN110555343B (en) * | 2018-06-01 | 2022-04-05 | 北京师范大学 | Method and system for extracting three elements of forest, shrub and grass in typical resource elements |
| CN110555343A (en) * | 2018-06-01 | 2019-12-10 | 北京师范大学 | method and system for extracting three elements of forest, shrub and grass in typical resource elements |
| CN108959396A (en) * | 2018-06-04 | 2018-12-07 | 众安信息技术服务有限公司 | Machine reading model training method and device, answering method and device |
| CN108777751A (en) * | 2018-06-07 | 2018-11-09 | 上海航动科技有限公司 | A kind of call center system and its voice interactive method, device and equipment |
| CN108922634A (en) * | 2018-06-13 | 2018-11-30 | 平安医疗科技有限公司 | The problem of based on online interrogation, replies processing method, device and computer equipment |
| CN109947909B (en) * | 2018-06-19 | 2024-03-12 | 平安科技（深圳）有限公司 | Intelligent customer service response method, equipment, storage medium and device |
| CN109947909A (en) * | 2018-06-19 | 2019-06-28 | 平安科技（深圳）有限公司 | Intelligent customer service answer method, equipment, storage medium and device |
| CN109165291A (en) * | 2018-06-29 | 2019-01-08 | 厦门快商通信息技术有限公司 | A kind of text matching technique and electronic equipment |
| CN109165291B (en) * | 2018-06-29 | 2021-07-09 | 厦门快商通信息技术有限公司 | Text matching method and electronic equipment |
| CN109241240A (en) * | 2018-08-17 | 2019-01-18 | 国家电网有限公司客户服务中心 | Power failure repairing information automatically forwarding method |
| CN109410911A (en) * | 2018-09-13 | 2019-03-01 | 何艳玲 | Artificial intelligence learning method based on speech recognition |
| CN109359302A (en) * | 2018-10-26 | 2019-02-19 | 重庆大学 | A kind of optimization method of field term vector and fusion sort method based on it |
| CN109446314A (en) * | 2018-11-14 | 2019-03-08 | 沈文策 | A kind of customer service question processing method and device |
| CN109493186A (en) * | 2018-11-20 | 2019-03-19 | 北京京东尚科信息技术有限公司 | The method and apparatus for determining pushed information |
| CN109684475A (en) * | 2018-11-21 | 2019-04-26 | 斑马网络技术有限公司 | Processing method, device, equipment and the storage medium of complaint |
| CN109657037A (en) * | 2018-12-21 | 2019-04-19 | 焦点科技股份有限公司 | A kind of knowledge mapping answering method and system based on entity type and semantic similarity |
| CN109739985A (en) * | 2018-12-26 | 2019-05-10 | 斑马网络技术有限公司 | Automatic document classification method, equipment and storage medium |
| CN109766421A (en) * | 2018-12-28 | 2019-05-17 | 上海汇付数据服务有限公司 | Intelligent Answer System and method |
| CN111626686B (en) * | 2019-02-28 | 2023-08-18 | 百度在线网络技术（北京）有限公司 | Intelligent tax processing method, device, terminal and medium |
| CN111626684A (en) * | 2019-02-28 | 2020-09-04 | 百度在线网络技术（北京）有限公司 | Intelligent tax processing method, device, terminal and medium |
| CN111626686A (en) * | 2019-02-28 | 2020-09-04 | 百度在线网络技术（北京）有限公司 | Intelligent tax processing method, device, terminal and medium |
| CN109993649A (en) * | 2019-03-13 | 2019-07-09 | 王亚萍 | A kind of pilot-operated type consultancy relating to payment of tax method for pushing |
| CN109993507A (en) * | 2019-04-17 | 2019-07-09 | 王亚萍 | A kind of pilot-operated type service system of paying taxes |
| CN110113422A (en) * | 2019-05-10 | 2019-08-09 | 南京硅基智能科技有限公司 | A kind of intension recognizing method and system of the virtual mobile phone based on cloud |
| CN110334110A (en) * | 2019-05-28 | 2019-10-15 | 平安科技（深圳）有限公司 | Natural language classification method, device, computer equipment and storage medium |
| CN110489533A (en) * | 2019-07-09 | 2019-11-22 | 深圳追一科技有限公司 | Interactive method and relevant device |
| CN110727778A (en) * | 2019-10-15 | 2020-01-24 | 大连中河科技有限公司 | Intelligent question-answering system for tax affairs |
| CN110827797B (en) * | 2019-11-06 | 2022-04-12 | 北京沃东天骏信息技术有限公司 | Voice response event classification processing method and device |
| CN110827797A (en) * | 2019-11-06 | 2020-02-21 | 北京沃东天骏信息技术有限公司 | Voice response event classification processing method and device |
| CN110909142B (en) * | 2019-11-20 | 2023-03-31 | 腾讯科技（深圳）有限公司 | Question and sentence processing method and device of question-answer model, electronic equipment and storage medium |
| CN110909142A (en) * | 2019-11-20 | 2020-03-24 | 腾讯科技（深圳）有限公司 | Question and sentence processing method and device of question-answer model, electronic equipment and storage medium |
| CN111221950A (en) * | 2019-12-30 | 2020-06-02 | 航天信息股份有限公司 | Method and device for analyzing weak emotion of user |
| CN111275536A (en) * | 2020-01-13 | 2020-06-12 | 山东浪潮商用系统有限公司 | Tax intelligent consultation method and system integrating manual intervention |
| CN111581338B (en) * | 2020-03-25 | 2022-01-28 | 北京市农林科学院 | Agricultural technical service robot man-machine fusion consultation question-answering method and system |
| CN111581338A (en) * | 2020-03-25 | 2020-08-25 | 北京市农林科学院 | Agricultural technical service robot man-machine fusion consultation question-answering method and system |
| CN111666381B (en) * | 2020-06-17 | 2022-11-18 | 中国电子科技集团公司第二十八研究所 | Task type question-answer interaction system oriented to intelligent control |
| CN111666381A (en) * | 2020-06-17 | 2020-09-15 | 中国电子科技集团公司第二十八研究所 | Task type question-answer interaction system oriented to intelligent control |
| CN112599120A (en) * | 2020-12-11 | 2021-04-02 | 上海中通吉网络技术有限公司 | Semantic determination method and device based on user-defined weighted WMD algorithm |
| CN112861757B (en) * | 2021-02-23 | 2022-11-22 | 天津汇智星源信息技术有限公司 | Intelligent record auditing method based on text semantic understanding and electronic equipment |
| CN112861757A (en) * | 2021-02-23 | 2021-05-28 | 天津汇智星源信息技术有限公司 | Intelligent record auditing method based on text semantic understanding and electronic equipment |
| CN114138949A (en) * | 2021-10-28 | 2022-03-04 | 数税(江苏)信息科技有限公司 | Tax intelligent question-answering customer service processing system |
| CN114297501A (en) * | 2021-12-29 | 2022-04-08 | 中国建设银行股份有限公司 | Text recommendation method, device, equipment and medium |
| CN117041618A (en) * | 2023-10-10 | 2023-11-10 | 北京装库创意科技有限公司 | Intelligent voice customer service method and system for electronic commerce |
| CN117041618B (en) * | 2023-10-10 | 2024-02-06 | 北京装库创意科技有限公司 | Intelligent voice customer service method and system for electronic commerce |
| CN117198290A (en) * | 2023-11-06 | 2023-12-08 | 深圳市金鼎胜照明有限公司 | Acoustic control-based multi-mode LED intelligent control method and apparatus |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106649561B (en) | 2020-05-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106649561A (en) | Intelligent question-answering system for tax consultation service | |
| CN110427623B (en) | Semi-structured document knowledge extraction method and device, electronic equipment and storage medium | |
| CN107992597B (en) | Text structuring method for power grid fault case | |
| CN113239186B (en) | Graph convolution network relation extraction method based on multi-dependency relation representation mechanism | |
| CN109858041B (en) | Named entity recognition method combining semi-supervised learning with user-defined dictionary | |
| CN109753566A (en) | The model training method of cross-cutting sentiment analysis based on convolutional neural networks | |
| CN108363690A (en) | Dialog semantics Intention Anticipation method based on neural network and learning training method | |
| Bokka et al. | Deep Learning for Natural Language Processing: Solve your natural language processing problems with smart deep neural networks | |
| Kandhro et al. | Sentiment analysis of students’ comment using long-short term model | |
| CN113255366B (en) | Aspect-level text emotion analysis method based on heterogeneous graph neural network | |
| US12032907B2 (en) | Transfer learning and prediction consistency for detecting offensive spans of text | |
| Deng et al. | [Retracted] News Text Classification Method Based on the GRU_CNN Model | |
| Yan et al. | Leveraging contextual sentences for text classification by using a neural attention model | |
| CN115687610A (en) | Text intention classification model training method, recognition device, electronic equipment and storage medium | |
| CN109446326A (en) | Biomedical event based on replicanism combines abstracting method | |
| Srinivasarao et al. | SMS sentiment classification using an evolutionary optimization based fuzzy recurrent neural network | |
| Korade et al. | Strengthening Sentence Similarity Identification Through OpenAI Embeddings and Deep Learning. | |
| KR102506778B1 (en) | Method and apparatus for analyzing risk of contract | |
| Lauren et al. | A low-dimensional vector representation for words using an extreme learning machine | |
| Stuart et al. | Intelligent semantic-based system for corpus analysis through hybrid probabilistic neural networks | |
| Valipour et al. | Unsupervised transfer learning via BERT neuron selection | |
| CN117216617A (en) | Text classification model training method, device, computer equipment and storage medium | |
| CN115796635A (en) | Bank digital transformation maturity evaluation system based on big data and machine learning | |
| Jiang et al. | Sentiment classification based on clause polarity and fusion via convolutional neural network | |
| Kreyssig | Deep learning for user simulation in a dialogue system |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20200526 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |