CN109585014A - A kind of result of decision determines method, apparatus, equipment and readable storage medium storing program for executing - Google Patents
A kind of result of decision determines method, apparatus, equipment and readable storage medium storing program for executing Download PDFInfo
- Publication number
- CN109585014A CN109585014A CN201811481536.4A CN201811481536A CN109585014A CN 109585014 A CN109585014 A CN 109585014A CN 201811481536 A CN201811481536 A CN 201811481536A CN 109585014 A CN109585014 A CN 109585014A
- Authority
- CN
- China
- Prior art keywords
- candidate answers
- evidence
- corpus
- target problem
- target
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
Landscapes
- Health & Medical Sciences (AREA)
- Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Medical Informatics (AREA)
- Public Health (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Pathology (AREA)
- Epidemiology (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
A kind of result of decision of the application determines method, device, equipment and readable storage medium storing program for executing, method includes: to get at least two candidate answers corresponding with target problem under target scene, for each candidate answers, evidence corpus relevant to the combination of candidate answers and target problem is retrieved in the corresponding corpus of target scene, it and according to evidence corpus include situation to word in candidate answers and target problem, obtain the matching characteristic of candidate answers and target problem, according to the matching characteristic of each candidate answers and target problem, the matching answer of target problem is determined in each candidate answers.Angle of the application scheme from natural language understanding and reasoning, it is directly based upon the corresponding corpus of target scene and carries out decision, the relationship in expert's building and more new knowledge base between problem and answer is not needed, it does not need to summarize inference rule yet, it saves cost and guarantees that rule conflict is not present in decision process, so that the result of decision determined is relatively reliable.
Description
This application claims on 07 03rd, 2018 submission Patent Office of the People's Republic of China, application No. is 201810717343.8, invention
The priority of the domestic applications of entitled " a kind of the result of decision determine method, apparatus, equipment and readable storage medium storing program for executing ", whole
Content is hereby incorporated by reference in the application.
Technical field
This application involves natural language processing technique fields, determine method, dress more specifically to a kind of result of decision
It sets, equipment and readable storage medium storing program for executing.
Background technique
Although China's medical industry is fast-developing in recent years, but still cannot fully meet people's medical treatment aspect
Great demand, therefore national policy supports the development of primary care energetically at present.And for developing primary care, core is
Promote the treatment level of base doctor.
Clinical Decision Support Systems (Clinical Decision Support System, CDSS) is a kind of for doctor
The computer assistance administrator system of diagnosis and treatment, it stores and uses a large amount of medical knowledges, according to the essential information of patient, state of an illness information
Deng providing various auxiliary and prompt in the diagnosis and treatment process of doctor, to help doctor more rationally efficiently to complete diagnosis and treatment
Work promotes whole medical treatment service level.Existing Clinical Decision Support Systems generally comprises: base module and reasoning mould
Block.
1, base module, is mainly responsible for the storage and calling of knowledge, and the knowledge of knowledge base storage can be structuring and know
Know, such as disease library, check library, drug storage and its between relationship.The building of knowledge base needs artificial participation, and needs not
The disconnected renewal of knowledge.
2, reasoning module contains a series of inference rule in knowledge based libraries summarized by medical expert, these rules
It is the direct reflection of expertise and knowledge.Reasoning module carries out logic judgment according to user data, using inference rule, obtains
Diagnosis is referred to for medical staff.
Inventor the study found that existing Clinical Decision Support Systems there are certain disadvantages, such as the building of knowledge base
And update the artificial investment for needing a large amount of medical experts, higher cost.It is and due to medical knowledge numerous and complicated, these knowledge are total
It bears and, the inference rule for being expressed as can be used for reasoning from logic is not easy to, and a large amount of expert is needed to put into.And due to very much
Medical knowledge and experience are fuzzy, it is not easy to be expressed as rule.Also, when rule reaches certain amount, may exist
Conflict in logic.
Therefore, the prior art needs the new result of decision of one kind and determines scheme, to avoid the defect of the prior art.
Summary of the invention
In view of this, this application provides a kind of result of decision to determine method, apparatus, equipment and readable storage medium storing program for executing, use
In dependence of the reduction to knowledge base, and in the case where not needing to summarize inference rule, the determination of the result of decision is realized.
To achieve the goals above, it is proposed that scheme it is as follows:
A kind of result of decision determines method, comprising:
Obtain at least two candidate answers corresponding with target problem under target scene；
For each candidate answers, in the corresponding corpus of target scene, retrieval and the candidate answers and the mesh
The relevant evidence corpus of the combination of mark problem；
According to the evidence corpus in the candidate answers and the target problem word include situation, obtain the time
Select the matching characteristic of answer Yu the target problem；
According to the matching characteristic of each candidate answers and the target problem, institute is determined in each candidate answers
State the matching answer of target problem.
Preferably, described to be directed to each candidate answers, in the corresponding corpus of target scene, retrieval is answered with the candidate
The relevant evidence corpus of the combination of case and the target problem, comprising:
For each candidate answers, retrieval is generated using the word that the candidate answers and the target problem respectively contain
Formula；
It according to the retrieval type, is retrieved in the corresponding corpus of target scene, retrieval obtains evidence corpus.
Preferably, it is described according to the evidence corpus in the candidate answers and the target problem word include feelings
Condition determines the matching characteristic of the candidate answers Yu the target problem, comprising:
Chain of evidence figure is constructed with reference to the evidence corpus, the node in the chain of evidence figure is by both appearing in the evidence language
In material, and appear in the composition of the word in the candidate answers or the target problem；
According to the chain of evidence figure, the matching characteristic of the candidate answers Yu the target problem is obtained.
It is preferably, described to construct chain of evidence figure with reference to the evidence corpus, comprising:
For evidence corpus described in each, corresponding document is constructed according to figure, the document is according to the node in figure by described
Word composition in candidate answers that evidence corpus includes and the target problem；
By the corresponding document of evidence corpus described in each item according to figure, by comprising same edge on the basis of merged, obtain
To fused chain of evidence figure.
Preferably, described for evidence corpus described in each, corresponding document is constructed according to figure, comprising:
Corresponding n kind document evidence is constructed for evidence corpus described in each according to n (8) the kind building mode of setting
Figure；
It is described by the corresponding document of evidence corpus described in each item according to figure, by comprising same edge on the basis of melted
It closes, obtains fused chain of evidence figure, comprising:
By evidence corpus described in each item according to document constructed by same building mode according to figure, with comprising same edge be
Benchmark is merged according to m (2) the kind amalgamation mode of setting, obtains the fused chain of evidence figure of n*m kind.
It is preferably, described that the matching characteristic of the candidate answers Yu the target problem is obtained according to the chain of evidence figure,
Include:
According to the matching characteristic template of setting, in the chain of evidence figure, obtains the candidate answers and asked with the target
The matching characteristic of topic.
Preferably, relevant to the combination of the candidate answers and the target problem evidence corpus of the retrieval it
Before, this method further include:
The candidate answers and the target problem are pre-processed, which includes: to be segmented, removed
Spcial character and punctuate remove stop words, determine word weight.
Preferably, the matching characteristic according to each candidate answers and the target problem, in each candidate
The matching answer of the target problem is determined in answer, comprising:
The matching characteristic of each candidate answers and the target problem is inputted into preset answer assessment models, is obtained
The matching status of the candidate answers of answer assessment models output；
The answer assessment models are, in advance using the matching characteristic of problem training data and candidate answers training data as sample
This, using each candidate answers training data whether be match answer annotation results be trained to obtain as sample label；
According to the matching status of each candidate answers, the matching answer of the target problem is therefrom determined.
A kind of result of decision determining device, comprising:
Data capture unit, for obtaining at least two candidate answers corresponding with target problem under target scene；
Retrieval unit, for being directed to each candidate answers, in the corresponding corpus of target scene, retrieval and the candidate
Answer and the relevant evidence corpus of the combination of the target problem；
Matching characteristic acquiring unit is used for according to the evidence corpus to word in the candidate answers and the target problem
Include situation, obtain the matching characteristic of the candidate answers Yu the target problem；
Answer determination unit is matched, for the matching characteristic according to each candidate answers and the target problem,
The matching answer of the target problem is determined in each candidate answers.
Preferably, the retrieval unit includes:
Retrieval type generation unit, it is each using the candidate answers and the target problem for being directed to each candidate answers
Self-contained word generates retrieval type；
Retrieval type retrieval unit, for retrieving, retrieving in the corresponding corpus of target scene according to the retrieval type
To evidence corpus.
Preferably, the matching characteristic acquiring unit includes:
Chain of evidence figure construction unit, for constructing chain of evidence figure with reference to the evidence corpus, the section in the chain of evidence figure
Point is by not only appearing in the evidence corpus, but also appears in the word composition in the candidate answers or the target problem；
Chain of evidence figure uses unit, for obtaining the candidate answers and the target problem according to the chain of evidence figure
Matching characteristic.
Preferably, the chain of evidence figure construction unit includes:
Document is according to figure construction unit, for constructing corresponding document according to figure, the list for evidence corpus described in each
Word in candidate answers that node in evidence figure includes by the evidence corpus and the target problem forms；
Document according to figure integrated unit, for by the corresponding document of evidence corpus described in each item according to figure, with comprising phase
With being merged on the basis of side, fused chain of evidence figure is obtained.
Preferably, the matching answer determination unit includes:
Model uses unit, preset for inputting the matching characteristic of each candidate answers and the target problem
Answer assessment models obtain the matching status of the candidate answers of answer assessment models output；
The answer assessment models are, in advance using the matching characteristic of problem training data and candidate answers training data as sample
This, using each candidate answers training data whether be match answer annotation results be trained to obtain as sample label；
As a result determination unit therefrom determines the target problem for the matching status according to each candidate answers
Match answer.
A kind of result of decision determines equipment, including memory and processor；
The memory, for storing program；
The processor realizes that the result of decision disclosed above determines each step of method for executing described program.
A kind of readable storage medium storing program for executing is stored thereon with computer program, real when the computer program is executed by processor
Now the result of decision disclosed above determines each step of method.
It can be seen from the above technical scheme that the result of decision provided by the embodiments of the present application determines method, mesh is got
Mark scene under at least two candidate answers corresponding with target problem, the target scene can be medical diagnosis scene or other
Scape, corresponding target problem and candidate answers can be patient information and medical diagnosis on disease scheme, and the application is answered for each candidate
Case retrieves evidence corpus relevant to the combination of candidate answers and target problem, and root in the corresponding corpus of target scene
It include situation to word in candidate answers and target problem according to evidence corpus, the matching for obtaining candidate answers and target problem is special
Sign, the matching characteristic reflect the support of matching answer of the candidate answers as target problem, therefore can be according to each time
The matching characteristic for selecting answer and target problem determines the matching answer of target problem in each candidate answers.Application scheme from
The angle of natural language understanding and reasoning is directly based upon the corresponding corpus of target scene and carries out decision, do not need expert's building
And the relationship in more new knowledge base between problem and answer, it does not need to summarize inference rule yet, saves cost and guarantee decision mistake
Rule conflict is not present in journey, so that the result of decision determined is relatively reliable.
Detailed description of the invention
In order to illustrate the technical solutions in the embodiments of the present application or in the prior art more clearly, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
The embodiment of application for those of ordinary skill in the art without creative efforts, can also basis
The attached drawing of offer obtains other attached drawings.
Fig. 1 determines method flow diagram for a kind of result of decision disclosed in the embodiment of the present application；
Fig. 2 a- Fig. 2 b illustrates the document of two evidence corpus building according to figure respectively；
Fig. 3 illustrates two documents and obtains chain of evidence figure according to figure fusion；
Fig. 4 illustrates the corresponding maximum spanning tree schematic diagram of chain of evidence figure；
Fig. 5 illustrates the corresponding bigraph (bipartite graph) of chain of evidence figure；
Fig. 6 is a kind of result of decision determining device structural schematic diagram disclosed in the embodiment of the present application；
Fig. 7 determines the hardware block diagram of equipment for a kind of result of decision disclosed in the embodiment of the present application.
Specific embodiment
Below in conjunction with the attached drawing in the embodiment of the present application, technical solutions in the embodiments of the present application carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of embodiments of the present application, instead of all the embodiments.It is based on
Embodiment in the application, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall in the protection scope of this application.
Exist under many scenes and ask-answer demand, that is, after providing a problem, needs to answer to corresponding matching of ging wrong
Case.By taking medical diagnosis scene as an example, Clinical Decision Support Systems be solve ask-answer demand.The patient information of input is
Problem needs to provide corresponding diagnosis scheme, which is answer.Diagnosis scheme can be medical diagnosis on disease, treatment, hand
Art, medication etc..
Still by taking the Clinical Decision Support Systems of medical diagnosis scene as an example, have the defects that numerous:
The building of knowledge base and the artificial investment for updating a large amount of medical experts of needs, higher cost.And due to medical knowledge
Numerous and complicated sums up these knowledge to come, and the inference rule for being expressed as can be used for reasoning from logic is not easy to, and needs a large amount of
Expert investment.And since many medical knowledges and experience are fuzzy, it is not easy to be expressed as rule.Also, when rule reaches
When certain amount, there may be conflict in logic.
In order to solve the defects of prior art, creative proposes a solution to inventor, is avoiding knowing
In the case where knowing library and reasoning module, the angle directly from natural language understanding and reasoning is based on the corresponding corpus of target scene
Library carries out decision support.Next, application scheme is discussed in detail, as shown in Figure 1, method includes:
Step S100, at least two candidate answers corresponding with target problem under target scene are obtained.
Specifically, application scheme can be adapted for several scenes, simply by the presence of ask-answer demand.Target scene can
To be medical diagnosis scene or other scenes.
In this step, at least two candidate answers corresponding with target problem under target scene are obtained.Wherein, target problem
The problem to be solved provided, user can provide the corresponding multiple candidate answers of target problem.Certainly, if user does not give
Candidate answers out then can regard all possible answer of target problem as candidate answers.
What medicine by taking medical diagnosis as an example, target problem is: user, which catches a cold, to have a running nose, and eats? it can be provided by user several
A candidate answers can also regard all drugs as candidate answers.
Step S110, each candidate answers are directed to, in the corresponding corpus of target scene, retrieval and the candidate answers
And the relevant evidence corpus of combination of the target problem.
Specifically, the corresponding corpus storage of target scene is corpus relevant to target scene.With target scene
For medical diagnosis scene, medical book, document, diagnosis report, case etc. can store in corresponding corpus.It can manage
Solution, the corresponding corpus of target scene contain all problems and corresponding answer under target scene.
Based on this, each candidate answers is directed in this step, using it with target problem as combining, in target scene pair
Relevant evidence corpus is combined in retrieval in the corpus answered.The evidence corpus retrieved becomes as candidate answers in support combination
The evidence of the matching answer of target problem.
Different according to the organizational form of corpus, the evidence corpus retrieved can be sentence, paragraph or complete chapter etc..
Step S120, according to the evidence corpus in the candidate answers and the target problem word include situation,
Obtain the matching characteristic of the candidate answers Yu the target problem.
It specifically, due to aforementioned retrieval is evidence corpus relevant with the combination of target problem to candidate answers,
Evidence corpus can be simultaneously comprising the word in candidate answers and target problem, but the information root such as the quantity for including, distance between word
There can be variation according to different evidence corpus.In this step, according to evidence corpus to the packet of word in candidate answers and target problem
Containing situation, the matching characteristic of candidate answers and target problem is obtained.The matching characteristic reflects candidate answers and target problem
Matching degree, i.e. candidate answers become a possibility that matching answer of target problem.
Step S130, it according to the matching characteristic of each candidate answers and the target problem, is answered in each candidate
The matching answer of the target problem is determined in case.
Specifically, aforementioned to be directed to each candidate answers, the matching characteristic of itself and target problem has been determined.Matching is special
Sign reflects the matching degree of candidate answers and target problem, therefore according to matching characteristic in this step, in each candidate answers really
Make the matching answer of target problem.
It is understood that the candidate answers of the matching answer as target problem, the matching characteristic with target problem
The matching degree reflected should be it is higher, such as choose matching degree highest one or topN candidate answers, asked as target
The matching answer of topic.
The result of decision provided by the embodiments of the present application determines method, from the angle of natural language understanding and reasoning, direct base
Decision is carried out in the corresponding corpus of target scene, does not need the pass in expert's building and more new knowledge base between problem and answer
System does not need to summarize inference rule yet, saves cost and guarantees that rule conflict is not present in decision process, so that the decision determined
As a result relatively reliable.
Optionally, between above-mentioned steps S100 and step S110, the embodiment of the present application can also be further increased to time
Answer and target problem is selected to carry out pretreated process.
Pretreated process may include:
1, word segmentation processing is carried out to candidate answers, target problem.
2, candidate answers, additional character and punctuate in target problem are removed.
3, candidate answers, the stop words in target problem are removed.Stop words can be IDF (Inverse Document
Frequency, inverse document frequency) lower word, as " ", " ".
4, its word weight is determined to the word that word segmentation processing obtains.It specifically can be using the IDF of standard as word weight.
In another embodiment of the application, to above-mentioned steps S110, for each candidate answers, in target scene pair
In the corpus answered, the process for retrieving evidence corpus relevant to the combination of the candidate answers and the target problem is situated between
It continues.
In order to realize the retrieval to candidate answers corpus related to the combination of target problem, can be directed in the present embodiment
Each candidate answers generate retrieval type using the word that the candidate answers and target problem respectively contain.Further, according to generation
Retrieval type is retrieved in the corresponding corpus of target scene, and retrieval obtains evidence corpus.
Wherein, it is connected between the word that candidate answers include in the retrieval type with "or", between the word that target problem includes
It is connected with "or", to improve recall rate.
Further, connected between candidate answers and target problem with "and", thus guarantee the obtained evidence corpus of retrieval with
The combination of candidate answers and target problem is related, that is, guarantees the accurate rate of search result.
The process retrieved in corpus in the present embodiment according to retrieval type can be examined using Lucene search engine
Rope.It is, of course, also possible to be retrieved using other search engines.
Optionally, foregoing description carries out pretreated process to candidate answers and target problem.Preprocessing process carries out
Participle, additional character and punctuate removal, stop words removal and word weight determine.Therefore, when generating retrieval type, in retrieval type
Additional character, punctuate and stop words can not included.Further, word can also mark weight in retrieval type.
Illustrate retrieval type generating process followed by a specific example under medical diagnosis scene.
Target problem is to determine that preferred contraceptive is according to patient information.Patient information includes: female, and uterine neck is in
Rotten to the corn state, cervix opening pine, anteversion of uterus.
Candidate answers include A-E, as follows respectively:
A, Topiramate B, carbamazepine C, ethymal D, Levetiracetam E, Clonazepam
For candidate answers A, its retrieval type corresponding with the combination of target problem is determined are as follows:
(uterine neck ^6.34 | | rotten to the corn ^8.86 | | state ^4.50 | | mouth pine ^5.82 | | uterus ^4.77 | | anteposition ^3.54) &&
(Topiramate ^11.23)
Wherein, the numerical value after " ^ " is word weight, and such as word weight of " uterine neck ^6.34 " expression " uterine neck " is 6.34."||"
Indicate the connection relationship of "or".The connection relationship of " && " expression "and".
Upper predicate weight is determined according to the IDF of word.
After generating retrieval type, it can use Lucene search engine and examined in the corresponding corpus of target scene
Rope is such as retrieved in the corresponding corpus of medical diagnosis, can retain the obtained whole results of retrieval as evidence corpus, can also
To take top n result as evidence corpus.
In another embodiment of the application, to above-mentioned steps S120, the candidate is answered according to the evidence corpus
Word includes situation in case and the target problem, obtains the process of the matching characteristic of the candidate answers and the target problem
It is introduced.
In the present embodiment chain of evidence figure can be constructed with circumstantial evidence corpus.Chain of evidence figure can be undirected diagram form.Card
According to the node in chain figure by not only appearing in evidence corpus, but also appear in the word composition in candidate answers or target problem.Node
Between line represent the corresponding word of two nodes co-occurrence in evidence corpus, and between node line weight by two node equivalents
Appearance situation in co-occurrence evidence corpus determines.It may include frequency of occurrence, two node equivalents altogether that this, which situation occurs,
Distance etc. in existing evidence corpus.
After constructing chain of evidence figure, the matching of candidate answers and target problem can be obtained according to the chain of evidence figure
Feature.
It is understood that being directed to each candidate answers, the corresponding evidence corpus building of candidate answers can use
Corresponding chain of evidence figure, and then according to the matching characteristic of chain of evidence figure acquisition candidate answers and target problem.
The matching characteristic reflects the matching degree of candidate answers and target problem, i.e. of the candidate answers as target problem
A possibility that with answer.
It should be noted that the evidence corpus that the combination for each candidate answers and target problem is retrieved,
The item number of evidence corpus can be a plurality of.Based on this, each candidate answers correspond to a plurality of evidence corpus.The embodiment of the present application is detailed
Introduce the process of above-mentioned circumstantial evidence corpus building chain of evidence figure:
1) it is directed to each evidence corpus of candidate answers, constructs corresponding document according to figure.
Wherein, the phrase in candidate answers and target problem that document includes by this evidence corpus according to the node in figure
At.The weight of line is determined by appearance situation of two node equivalents in this evidence corpus between node.
Assuming that candidate answers A is corresponding with x evidence corpus, then corresponding list can be constructed for each evidence corpus
Evidence figure.Therefore, x document can be constructed according to figure for candidate answers A by amounting to.Due to the corresponding x evidence corpus of candidate answers A
It is different comprising situation to what is segmented in candidate answers and target problem, therefore x document may according to the corresponding participle of figure interior joint
Less, and between node the weight of line may also be different.
Referring to Fig. 2 a and Fig. 2 b, the document for the building of two evidence corpus is illustrated respectively according to figure.
Specifically, target problem is to determine that preferred treatment method is according to patient information.Patient information includes:
Female falls down rear left shoulder and lands injury, shoulder swelling, pain, shoulder mobility obstacle.X-ray film display left side humerus shell shin
Cortical bone continuity is interrupted, without obvious displacement.
Candidate answers include A-E, as follows respectively:
A, the fixed B of triangular bandage suspension patch chest, gypsum fix C, open reduction and Kirschner wire fixation D, Set by small splints from outside E, ruler outside
Bone olecranon Bone traction+Boards wall
By taking candidate answers A as an example, evidence corpus relevant to the combination of candidate answers A and target problem is retrieved.With retrieval
For obtained two evidence corpus 1,2, in which:
Evidence corpus 1: ... there is upper limb stretch outward turning or extendes backIt landsInjured history,Shoulder pain、Swelling... take an examination
Worry hasShoulder jointDislocation is possible ... has sense of emptiness, upper limb is flexible at Scapula glenoidIt is fixed
Evidence corpus 2: pureShoulder jointDislocation can be used after resettingTriangular bandage suspensionUpper limb, 90 degree of elbow joint buckling, armpit
Cotton pad is padded at nestIt is fixed3 weeks
The candidate answers A and the word in target problem of co-occurrence are marked with underscore in above-mentioned evidence corpus.
Wherein, the document of the corresponding building of evidence corpus 1 is as shown in Figure 2 a according to figure, the document evidence of the corresponding building of evidence corpus 2
Figure is as shown in Figure 2 b.
The exemplary document of Fig. 2 a and Fig. 2 b merely illustrates node and connection relation between nodes according to figure, for node weights,
It is in side right weight equal parameter graph and unmarked.
Can be seen that the word for co-occurrence in evidence corpus by Fig. 2 a and Fig. 2 b, corresponding document according in figure with
Joint form occurs, and there is line in figure between any two node, indicates the corresponding word of node in same evidence corpus
Middle co-occurrence.
In Fig. 2 a and Fig. 2 b, node " fixation ", " suspention ", " triangle bandage " are the word occurred in candidate answers A, other nodes
For the word occurred in target problem.
Optionally, it is being directed to each evidence corpus, it, can be according to preset more when constructing corresponding document according to figure
Kind building mode is constructed.It is assumed that having preset n kind building mode, then it is directed to each evidence corpus, constructs corresponding n
Kind document is according to figure.
Under different building modes, constructed document is according to the node weights of figure, the calculation of side right weight, evidence weight.
Referring to the following table 1, the several calculation of node weights, side right weight, evidence weight is illustrated:
Table 1
Obviously, a kind of node weights calculation, 2 kinds of side right re-computation modes, 4 kinds of evidence weights are illustrated in above-mentioned table 1
Calculation, node weights, side right weight, evidence weight calculation in any combination, therefore can co-exist in 1*2*4=8 kind structure
Build mode.
Certainly, table 1 merely illustrates several weight calculation mode, other weight calculation sides in addition to this can also be arranged
Formula.
2) by the corresponding document of evidence corpus described in each item according to figure, by comprising same edge on the basis of merged,
Obtain fused chain of evidence figure.
Specifically, above-mentioned 1) to construct corresponding document according to figure, this step for each evidence corpus of candidate answers
It is middle to merge each document according to figure, obtain fused chain of evidence figure.
Specifically, carry out document according to figure merge when, by comprising same edge on the basis of merged.Wherein, identical
While being the line between identical two nodes.
Still by taking corresponding two documents of Fig. 2 a and Fig. 2 b are according to figure as an example, the fusion process of the two is introduced, it is shown in Figure 3,
Fig. 3 illustrates corresponding two documents of Fig. 2 a and Fig. 2 b according to figure fusion process schematic diagram.
Corresponding two documents of Fig. 2 a and Fig. 2 b according to figure fusion when, between " shoulder ", " joint ", " fixation " three nodes
Side on the basis of merged.Side right weight of the side in fused chain of evidence figure as benchmark is according to setting amalgamation mode weight
New to calculate, remaining side right remains unchanged again.
Amalgamation mode may exist m kind in the present embodiment, and previous embodiment illustrates to may exist n kind document according to figure structure
Mode is built, therefore every evidence corpus is corresponding with n kind document according to figure.It, can be by evidence corpus described in each item according to same based on this
Document constructed by a kind of building mode according to figure, by comprising same edge on the basis of, melted according to the m kind amalgamation mode of setting
It closes, obtains the fused chain of evidence figure of n*m kind.
Next two kinds of optional amalgamation modes are introduced:
The first: same edge weighted superposition
Wedge_merge=(Wnode1+Wnode2)(Wedge1Wevd1+Wedge2Wevd2)
Wherein, Wnode1And Wnode2It is document according to same edge two sides node weights, W in Fig. 1 and 2edge1And Wedge2For document evidence
The side right weight of same edge, W in Fig. 1 and 2evd1And Wevd2It is document according to the evidence weight of Fig. 1 and 2, Wedge_mergeIt is document according to Fig. 1
With 2 in same edge side in chain of evidence figure after fusion side right weight.
Second: maximum side right weight
Wedge_merge=(Wnode1+Wnode2)max(Wedge1Wevd1,Wedge2Wevd2)
Wherein the definition of parameters is identical as first way in formula.
With the exemplary n=8 of above-described embodiment, for m=2, then amount to chain of evidence after available 8*2=16 kind fusion
Figure.
Further, after constructing chain of evidence figure, candidate answers and target problem can be obtained according to chain of evidence figure
Matching characteristic.
Specifically, several matching characteristic template can be set in the present embodiment, such as set β kind matching characteristic template, into
And in chain of evidence figure, matching characteristic is obtained according to the matching characteristic template of setting.
It is understood that n*m kind chain of evidence figure can be generated in above-mentioned have been described above, it can be according to β kind in this step
Matching characteristic is obtained with feature templates, and then n*m* β kind matching characteristic can be got, namely gets n*m* β dimension matching spy
Sign.
Next, introducing several matching characteristic templates.
1, it regard the sum of side right weight of the maximum spanning tree of chain of evidence figure as matching characteristic.
Specifically, the maximum spanning tree of chain of evidence figure is firstly generated, and then by the conduct of the sum of the side right of maximum spanning tree weight
Matching characteristic.
Spanning tree is a kind of processing mode of connected graph, and target is that the ring cutting removed in original image still maintains connection.And most
Big spanning tree is maximum one kind of the sum of side right weight in all possible spanning tree.Spanning tree can embody the connectivity of figure,
The namely connectivity of chain of evidence.
By taking the exemplary fused chain of evidence figure of Fig. 3 as an example, the maximum spanning tree of the chain of evidence figure, effect such as Fig. 4 are generated
It is shown.
2, it regard the sum of side right weight of the minimum spanning tree of chain of evidence figure as matching characteristic.
Specifically, the minimum spanning tree of chain of evidence figure is firstly generated, and then by the conduct of the sum of the side right of minimum spanning tree weight
Matching characteristic.Minimum spanning tree is the smallest one kind of the sum of side right weight in the corresponding all possible spanning tree of chain of evidence figure.
3, it regard the sum of side right weight of the maximum spanning tree of the bigraph (bipartite graph) of chain of evidence figure as matching characteristic.
Specifically, the bigraph (bipartite graph) of chain of evidence figure is firstly generated, the maximum spanning tree of bigraph (bipartite graph) is further generated, maximum is raw
The sum of side right weight of Cheng Shu is used as matching characteristic.
It is understood that the node that chain of evidence figure includes can be divided into two parts, a part is to belong to target problem packet
The word contained, another part are the words for belonging to candidate answers and including.According to this two parts, the bigraph (bipartite graph) of chain of evidence figure is generated.In life
When at bigraph (bipartite graph), the side between the word in chain of evidence figure in each section is removed, only retains the side between two parts.Shown with Fig. 3
For the fused chain of evidence figure of example, the bigraph (bipartite graph) of the chain of evidence figure is generated, effect is as shown in Figure 5.
In Fig. 5, " fixation " on right side, " suspention ", " triangle bandage " belong to the word that candidate answers include, and one as bigraph (bipartite graph)
Part, residue participle belong to the word that target problem includes, another part as bigraph (bipartite graph).
4, it regard the sum of side right weight of the minimum spanning tree of the bigraph (bipartite graph) of chain of evidence figure as matching characteristic.
Specifically, the bigraph (bipartite graph) of chain of evidence figure is firstly generated, the minimum spanning tree of bigraph (bipartite graph) is further generated, it will most your pupil
The sum of side right weight of Cheng Shu is used as matching characteristic.
5, using the node number in chain of evidence figure as matching characteristic.
6, chain of evidence figure interior joint number is accounted for into the ratio of all word numbers of target problem as matching characteristic.
Illustrate 6 kinds of matching characteristic templates, it is to be understood that in addition to this other classes can also be set in the present embodiment
The matching characteristic template of type.
With the exemplary n=8 of above-described embodiment, m=2 for β=6, then it is special to amount to available 8*2*6=96 dimension matching
Sign.
After the n*m* β dimension matching characteristic for getting each candidate answers and target problem of above-described embodiment introduction,
It is further described step S130, according to the matching characteristic of each candidate answers and the target problem, in each candidate
The process of the matching answer of the target problem is determined in answer.
The prediction of matching answer can be realized in the present embodiment by neural network model.Specifically, it can instruct in advance
Practice answer assessment models, which is neural network model, such as MLP (Multi-layer Perceptron, multilayer
Perceptron) neural network model of type or the neural network model of other forms.
The application can obtain problem training data and candidate answers training data in advance, for same problem training number
According to, be corresponding with multiple candidate answers training datas, and be labelled with sample label in advance for each candidate answers training data, i.e., it is logical
It crosses sample label and marks whether it is matching answer.Further, of problem training data and candidate answers training data is obtained
With feature, whether each candidate answers training data is the annotation results for matching answer as sample by the matching characteristic that will acquire
For sample label, training answer assessment models.
It can be trained by each candidate answers and the matching characteristic input of the target problem in the present embodiment
In answer assessment models, and obtain the matching status of the candidate answers of answer assessment models output.Wherein, candidate answers
Matching status can be qualitative results or quantitative result, such as the matching status of candidate answers are as follows: candidate answers and target problem
Match, candidate answers and target problem mismatch；Alternatively, the matching status of candidate answers are as follows: the matching rate score of candidate answers.
According to the matching status of each candidate answers, the matching answer of the target problem is therefrom determined.Such as work as matching
When state is qualitative results, matched candidate answers can be chosen, the matching answer as target problem.When matching status is fixed
When measuring result, the highest candidate answers of matching rate can be chosen, the matching answer as target problem.
Result of decision determining device provided by the embodiments of the present application is described below, the result of decision described below is true
Determine device and determines that method can correspond to each other reference with the above-described result of decision.
Referring to Fig. 6, Fig. 6 is a kind of result of decision determining device structural schematic diagram disclosed in the embodiment of the present application.Such as Fig. 6 institute
Show, the apparatus may include:
Data capture unit 11, for obtaining at least two candidate answers corresponding with target problem under target scene；
Retrieval unit 12, for being directed to each candidate answers, in the corresponding corpus of target scene, retrieval and the time
Select answer and the relevant evidence corpus of the combination of the target problem；
Matching characteristic acquiring unit 13 is used for according to the evidence corpus in the candidate answers and the target problem
Word includes situation, obtains the matching characteristic of the candidate answers Yu the target problem；
Answer determination unit 14 is matched, for the matching characteristic according to each candidate answers and the target problem,
The matching answer of the target problem is determined in each candidate answers.
Optionally, the retrieval unit may include:
Retrieval type generation unit, it is each using the candidate answers and the target problem for being directed to each candidate answers
Self-contained word generates retrieval type；
Retrieval type retrieval unit, for retrieving, retrieving in the corresponding corpus of target scene according to the retrieval type
To evidence corpus.
Optionally, the matching characteristic acquiring unit may include:
Chain of evidence figure construction unit, for constructing chain of evidence figure with reference to the evidence corpus, the section in the chain of evidence figure
Point is by not only appearing in evidence corpus, but also appears in the word composition in the candidate answers or the target problem；
Chain of evidence figure uses unit, for obtaining the candidate answers and the target problem according to the chain of evidence figure
Matching characteristic.
Optionally, the chain of evidence figure construction unit may include:
Document is according to figure construction unit, for constructing corresponding document according to figure, the list for evidence corpus described in each
Word in candidate answers that node in evidence figure includes by the evidence corpus and the target problem forms；
Document according to figure integrated unit, for by the corresponding document of evidence corpus described in each item according to figure, with comprising phase
With being merged on the basis of side, fused chain of evidence figure is obtained.
Optionally, the document may include: according to figure construction unit
First document constructs subelement according to figure, for the n kind building mode according to setting, for evidence language described in each
Material, constructs corresponding n kind document according to figure.Based on this,
The document may include: according to figure integrated unit
First document merges subelement according to figure, is used for evidence corpus described in each item according to constructed by same building mode
Document according to figure, by comprising same edge on the basis of, merged according to the m kind amalgamation mode of setting, after obtaining the fusion of n*m kind
Chain of evidence figure.
Optionally, the chain of evidence figure may include: using unit
First chain of evidence figure uses subelement, obtains in the chain of evidence figure for the matching characteristic template according to setting
Take the matching characteristic of the candidate answers Yu the target problem.
Optionally, result of decision determining device disclosed in the embodiment of the present application can also include:
Pretreatment unit, for retrieving evidence corpus relevant to the combination of the candidate answers and the target problem
Before, the candidate answers and the target problem are pre-processed, which includes: to be segmented, remove spy
Different character and punctuate remove stop words, determine word weight.
Optionally, the matching answer determination unit may include:
Model uses unit, preset for inputting the matching characteristic of each candidate answers and the target problem
Answer assessment models obtain the matching status of the candidate answers of answer assessment models output；
The answer assessment models are, in advance using the matching characteristic of problem training data and candidate answers training data as sample
This, using each candidate answers training data whether be match answer annotation results be trained to obtain as sample label；
As a result determination unit therefrom determines the target problem for the matching status according to each candidate answers
Match answer.
Result of decision determining device provided by the embodiments of the present application can be applied to the result of decision and determine equipment, as PC terminal,
Cloud platform, server and server cluster etc..Optionally, Fig. 7 shows the hardware block diagram that the result of decision determines equipment, ginseng
According to Fig. 7, the result of decision determines that the hardware configuration of equipment may include: at least one processor 1, at least one communication interface 2, until
A few memory 3 and at least one communication bus 4；
In the embodiment of the present application, processor 1, communication interface 2, memory 3, communication bus 4 quantity be at least one,
And processor 1, communication interface 2, memory 3 complete mutual communication by communication bus 4；
Processor 1 may be a central processor CPU or specific integrated circuit ASIC (Application
Specific Integrated Circuit), or be arranged to implement the integrated electricity of one or more of the embodiment of the present invention
Road etc.；
Memory 3 may include high speed RAM memory, it is also possible to further include nonvolatile memory (non-volatile
Memory) etc., a for example, at least magnetic disk storage；
Wherein, memory is stored with program, the program that processor can call memory to store, and described program is used for:
Obtain at least two candidate answers corresponding with target problem under target scene；
For each candidate answers, in the corresponding corpus of target scene, retrieval and the candidate answers and the mesh
The relevant evidence corpus of the combination of mark problem；
According to the evidence corpus in the candidate answers and the target problem word include situation, obtain the time
Select the matching characteristic of answer Yu the target problem；
According to the matching characteristic of each candidate answers and the target problem, institute is determined in each candidate answers
State the matching answer of target problem.
Optionally, the refinement function of described program and extension function can refer to above description.
The embodiment of the present application also provides a kind of storage medium, which can be stored with the journey executed suitable for processor
Sequence, described program are used for:
Obtain at least two candidate answers corresponding with target problem under target scene；
For each candidate answers, in the corresponding corpus of target scene, retrieval and the candidate answers and the mesh
The relevant evidence corpus of the combination of mark problem；
According to the evidence corpus in the candidate answers and the target problem word include situation, obtain the time
Select the matching characteristic of answer Yu the target problem；
According to the matching characteristic of each candidate answers and the target problem, institute is determined in each candidate answers
State the matching answer of target problem.
Optionally, the refinement function of described program and extension function can refer to above description.
Finally, it is to be noted that, herein, relational terms such as first and second and the like be used merely to by
One entity or operation are distinguished with another entity or operation, without necessarily requiring or implying these entities or operation
Between there are any actual relationship or orders.Moreover, the terms "include", "comprise" or its any other variant meaning
Covering non-exclusive inclusion, so that the process, method, article or equipment for including a series of elements not only includes that
A little elements, but also including other elements that are not explicitly listed, or further include for this process, method, article or
The intrinsic element of equipment.In the absence of more restrictions, the element limited by sentence "including a ...", is not arranged
Except there is also other identical elements in the process, method, article or apparatus that includes the element.
Each embodiment in this specification is described in a progressive manner, the highlights of each of the examples are with other
The difference of embodiment, the same or similar parts in each embodiment may refer to each other.
The foregoing description of the disclosed embodiments makes professional and technical personnel in the field can be realized or use the application.
Various modifications to these embodiments will be readily apparent to those skilled in the art, as defined herein
General Principle can be realized in other embodiments without departing from the spirit or scope of the application.Therefore, the application
It is not intended to be limited to the embodiments shown herein, and is to fit to and the principles and novel features disclosed herein phase one
The widest scope of cause.
Claims (15)
1. a kind of result of decision determines method characterized by comprising
Obtain at least two candidate answers corresponding with target problem under target scene；
For each candidate answers, in the corresponding corpus of target scene, retrieval is asked with the candidate answers and the target
The relevant evidence corpus of the combination of topic；
According to the evidence corpus in the candidate answers and the target problem word include situation, obtain the candidate and answer
The matching characteristic of case and the target problem；
According to the matching characteristic of each candidate answers and the target problem, the mesh is determined in each candidate answers
The matching answer of mark problem.
2. being corresponded to the method according to claim 1, wherein described be directed to each candidate answers in target scene
Corpus in, retrieve relevant to the combination of the candidate answers and target problem evidence corpus, comprising:
For each candidate answers, retrieval type is generated using the word that the candidate answers and the target problem respectively contain；
It according to the retrieval type, is retrieved in the corresponding corpus of target scene, retrieval obtains evidence corpus.
3. the method according to claim 1, wherein it is described according to the evidence corpus to the candidate answers and
Word includes situation in the target problem, determines the matching characteristic of the candidate answers Yu the target problem, comprising:
Chain of evidence figure is constructed with reference to the evidence corpus, the node in the chain of evidence figure is by both appearing in the evidence corpus
In, and appear in the composition of the word in the candidate answers or the target problem；
According to the chain of evidence figure, the matching characteristic of the candidate answers Yu the target problem is obtained.
4. according to the method described in claim 3, it is characterized in that, described construct chain of evidence figure, packet with reference to the evidence corpus
It includes:
For evidence corpus described in each, corresponding document is constructed according to figure, the document is according to the node in figure by the evidence
Word composition in candidate answers that corpus includes and the target problem；
By the corresponding document of evidence corpus described in each item according to figure, by comprising same edge on the basis of merged, melted
Chain of evidence figure after conjunction.
5. according to the method described in claim 4, building corresponds to it is characterized in that, described for evidence corpus described in each
Document according to figure, comprising:
Corresponding n kind document is constructed according to figure for evidence corpus described in each according to the n kind building mode of setting；
It is described by the corresponding document of evidence corpus described in each item according to figure, by comprising same edge on the basis of merged, obtain
To fused chain of evidence figure, comprising:
By evidence corpus described in each item according to document constructed by same building mode according to figure, using comprising same edge as base
Standard is merged according to the m kind amalgamation mode of setting, obtains the fused chain of evidence figure of n*m kind.
6. according to the method described in claim 3, obtaining the candidate and answering it is characterized in that, described according to the chain of evidence figure
The matching characteristic of case and the target problem, comprising:
According to the matching characteristic template of setting, in the chain of evidence figure, the candidate answers and the target problem are obtained
Matching characteristic.
7. the method according to claim 1, wherein being asked in the retrieval with the candidate answers and the target
Before the relevant evidence corpus of the combination of topic, this method further include:
The candidate answers and the target problem are pre-processed, the preprocessing process include: segmented, remove it is special
Character and punctuate remove stop words, determine word weight.
8. method according to claim 1-7, which is characterized in that described according to each candidate answers and institute
The matching characteristic for stating target problem determines the matching answer of the target problem in each candidate answers, comprising:
The matching characteristic of each candidate answers and the target problem is inputted into preset answer assessment models, obtains answer
The matching status of the candidate answers of assessment models output；
The answer assessment models are, in advance using the matching characteristic of problem training data and candidate answers training data as sample,
Using each candidate answers training data whether be match answer annotation results be trained to obtain as sample label；
According to the matching status of each candidate answers, the matching answer of the target problem is therefrom determined.
9. a kind of result of decision determining device characterized by comprising
Data capture unit, for obtaining at least two candidate answers corresponding with target problem under target scene；
Retrieval unit, for being directed to each candidate answers, in the corresponding corpus of target scene, retrieval and the candidate answers
And the relevant evidence corpus of combination of the target problem；
Matching characteristic acquiring unit, for the packet according to the evidence corpus to word in the candidate answers and the target problem
Containing situation, the matching characteristic of the candidate answers Yu the target problem is obtained；
Answer determination unit is matched, for the matching characteristic according to each candidate answers and the target problem, in each institute
State the matching answer that the target problem is determined in candidate answers.
10. device according to claim 9, which is characterized in that the retrieval unit includes:
Retrieval type generation unit is respectively wrapped for being directed to each candidate answers using the candidate answers and the target problem
The word contained generates retrieval type；
Retrieval type retrieval unit, for being retrieved in the corresponding corpus of target scene according to the retrieval type, retrieval is demonstrate,proved
According to corpus.
11. device according to claim 9, which is characterized in that the matching characteristic acquiring unit includes:
Chain of evidence figure construction unit, for constructing chain of evidence figure with reference to the evidence corpus, the node in the chain of evidence figure by
Not only it had appeared in the evidence corpus, but also has appeared in the word composition in the candidate answers or the target problem；
Chain of evidence figure uses unit, for obtaining of the candidate answers Yu the target problem according to the chain of evidence figure
With feature.
12. device according to claim 11, which is characterized in that the chain of evidence figure construction unit includes:
Document is according to figure construction unit, for constructing corresponding document according to figure, the document evidence for evidence corpus described in each
Word in candidate answers that node in figure includes by the evidence corpus and the target problem forms；
Document according to figure integrated unit, for by the corresponding document of evidence corpus described in each item according to figure, with comprising same edge
On the basis of merged, obtain fused chain of evidence figure.
13. according to the described in any item devices of claim 9-12, which is characterized in that the matching answer determination unit includes:
Model uses unit, for the matching characteristic of each candidate answers and the target problem to be inputted preset answer
Assessment models obtain the matching status of the candidate answers of answer assessment models output；
The answer assessment models are, in advance using the matching characteristic of problem training data and candidate answers training data as sample,
Using each candidate answers training data whether be match answer annotation results be trained to obtain as sample label；
As a result determination unit therefrom determines the matching of the target problem for the matching status according to each candidate answers
Answer.
14. a kind of result of decision determines equipment, which is characterized in that including memory and processor；
The memory, for storing program；
The processor realizes such as result of decision determination side of any of claims 1-8 for executing described program
Each step of method.
15. a kind of readable storage medium storing program for executing, is stored thereon with computer program, which is characterized in that the computer program is processed
When device executes, realize that the result of decision of any of claims 1-8 such as determines each step of method.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2018107173438 | 2018-07-03 | ||
| CN201810717343.8A CN108962384A (en) | 2018-07-03 | 2018-07-03 | A kind of result of decision determines method, apparatus, equipment and readable storage medium storing program for executing |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109585014A true CN109585014A (en) | 2019-04-05 |
| CN109585014B CN109585014B (en) | 2021-08-13 |
Family
ID=64485183
Family Applications (2)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810717343.8A Pending CN108962384A (en) | 2018-07-03 | 2018-07-03 | A kind of result of decision determines method, apparatus, equipment and readable storage medium storing program for executing |
| CN201811481536.4A Active CN109585014B (en) | 2018-07-03 | 2018-12-05 | Decision result determination method, device, equipment and readable storage medium |
Family Applications Before (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810717343.8A Pending CN108962384A (en) | 2018-07-03 | 2018-07-03 | A kind of result of decision determines method, apparatus, equipment and readable storage medium storing program for executing |
Country Status (1)
| Country | Link |
|---|---|
| CN (2) | CN108962384A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111681765A (en) * | 2020-04-29 | 2020-09-18 | 华南师范大学 | Multi-model fusion method of medical question-answering system |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110674292B (en) * | 2019-08-27 | 2023-04-18 | 腾讯科技（深圳）有限公司 | Man-machine interaction method, device, equipment and medium |
| CN115132314B (en) * | 2022-09-01 | 2022-12-20 | 合肥综合性国家科学中心人工智能研究院(安徽省人工智能实验室) | Examination impression generation model training method, examination impression generation model training device and examination impression generation model generation method |
| CN117171398B (en) * | 2023-11-01 | 2024-02-09 | 浙江大学高端装备研究院 | Method, device and equipment for constructing service tree of industrial Internet platform |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103124980A (en) * | 2010-09-24 | 2013-05-29 | 国际商业机器公司 | Providing answers to questions including assembling answers from multiple document segments |
| EP2750085A1 (en) * | 2012-12-31 | 2014-07-02 | Facebook, Inc. | Natural-language rendering of structured search queries |
-
2018
- 2018-07-03 CN CN201810717343.8A patent/CN108962384A/en active Pending
- 2018-12-05 CN CN201811481536.4A patent/CN109585014B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103124980A (en) * | 2010-09-24 | 2013-05-29 | 国际商业机器公司 | Providing answers to questions including assembling answers from multiple document segments |
| EP2750085A1 (en) * | 2012-12-31 | 2014-07-02 | Facebook, Inc. | Natural-language rendering of structured search queries |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111681765A (en) * | 2020-04-29 | 2020-09-18 | 华南师范大学 | Multi-model fusion method of medical question-answering system |
| CN111681765B (en) * | 2020-04-29 | 2023-08-11 | 华南师范大学 | Multi-model fusion method of medical question-answering system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109585014B (en) | 2021-08-13 |
| CN108962384A (en) | 2018-12-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109585014A (en) | A kind of result of decision determines method, apparatus, equipment and readable storage medium storing program for executing | |
| CN108614885B (en) | Knowledge mapping analysis method and device based on medical information | |
| CN109741806B (en) | Auxiliary generation method and device for medical image diagnosis report | |
| Sofaer | The body as material culture: a theoretical osteoarchaeology | |
| CN110675944A (en) | Triage method and device, computer equipment and medium | |
| JP4380142B2 (en) | Search system and search method | |
| CN109599185A (en) | Disease data processing method, device, electronic equipment and computer-readable medium | |
| CN110289068A (en) | Drug recommended method and equipment | |
| CN110838368A (en) | Robot active inquiry method based on traditional Chinese medicine clinical knowledge graph | |
| CN111048167A (en) | Hierarchical case structuring method and system | |
| CN112201350A (en) | Intelligent triage method and device and electronic equipment | |
| JPWO2009025022A1 (en) | Information processing system and program | |
| Monsen et al. | An empiric analysis of Omaha System targets | |
| CN107066816A (en) | Medical treatment guidance method, device and server based on clinical data | |
| CN116469534A (en) | Hospital number calling management system and method thereof | |
| Ogunyemi et al. | Combining geometric and probabilistic reasoning for computer-based penetrating-trauma assessment | |
| CN110610766A (en) | Apparatus and storage medium for deriving probability of disease based on symptom feature weight | |
| Cimino | Distributed cognition and knowledge-based controlled medical terminologies | |
| CN112635050A (en) | Diagnosis recommendation method, electronic equipment and storage device | |
| CN110289065A (en) | A kind of auxiliary generates the control method and device of medical electronic report | |
| Mendes et al. | Extended clinical discourse representation structure for controlled natural language clinical decision support systems | |
| Roger et al. | Medical Informatics Europe 85: Proceedings, Helsinki, Finland August 25–29, 1985 | |
| CN111986799B (en) | Orthopedics knowledge graph construction system taking joint movement function as core | |
| JP2013164872A (en) | Database system | |
| EP3985679A1 (en) | Technique for providing an interactive display of a medical image |
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 |