CN107992482A - Mathematics subjective item answers the stipulations method and system of step - Google Patents
Mathematics subjective item answers the stipulations method and system of step Download PDFInfo
- Publication number
- CN107992482A CN107992482A CN201711437048.9A CN201711437048A CN107992482A CN 107992482 A CN107992482 A CN 107992482A CN 201711437048 A CN201711437048 A CN 201711437048A CN 107992482 A CN107992482 A CN 107992482A
- Authority
- CN
- China
- Prior art keywords
- answer
- read
- stipulations
- module
- representation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 62
- 238000012545 processing Methods 0.000 claims abstract description 45
- 230000011218 segmentation Effects 0.000 claims abstract description 43
- 238000000926 separation method Methods 0.000 claims abstract description 13
- 238000012916 structural analysis Methods 0.000 claims abstract description 12
- 230000007935 neutral effect Effects 0.000 claims description 31
- 238000000605 extraction Methods 0.000 claims description 12
- 238000004422 calculation algorithm Methods 0.000 claims description 10
- 238000012549 training Methods 0.000 claims description 7
- 238000013519 translation Methods 0.000 claims description 5
- 238000009795 derivation Methods 0.000 description 12
- 230000000875 corresponding effect Effects 0.000 description 6
- 238000004458 analytical method Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 238000006243 chemical reaction Methods 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000012417 linear regression Methods 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- PCTMTFRHKVHKIS-BMFZQQSSSA-N (1s,3r,4e,6e,8e,10e,12e,14e,16e,18s,19r,20r,21s,25r,27r,30r,31r,33s,35r,37s,38r)-3-[(2r,3s,4s,5s,6r)-4-amino-3,5-dihydroxy-6-methyloxan-2-yl]oxy-19,25,27,30,31,33,35,37-octahydroxy-18,20,21-trimethyl-23-oxo-22,39-dioxabicyclo[33.3.1]nonatriaconta-4,6,8,10 Chemical compound C1C=C2C[C@@H](OS(O)(=O)=O)CC[C@]2(C)[C@@H]2[C@@H]1[C@@H]1CC[C@H]([C@H](C)CCCC(C)C)[C@@]1(C)CC2.O[C@H]1[C@@H](N)[C@H](O)[C@@H](C)O[C@H]1O[C@H]1/C=C/C=C/C=C/C=C/C=C/C=C/C=C/[C@H](C)[C@@H](O)[C@@H](C)[C@H](C)OC(=O)C[C@H](O)C[C@H](O)CC[C@@H](O)[C@H](O)C[C@H](O)C[C@](O)(C[C@H](O)[C@H]2C(O)=O)O[C@H]2C1 PCTMTFRHKVHKIS-BMFZQQSSSA-N 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000005194 fractionation Methods 0.000 description 1
- 230000008595 infiltration Effects 0.000 description 1
- 238000001764 infiltration Methods 0.000 description 1
- 230000015654 memory Effects 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 210000004218 nerve net Anatomy 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- SBNFWQZLDJGRLK-UHFFFAOYSA-N phenothrin Chemical compound CC1(C)C(C=C(C)C)C1C(=O)OCC1=CC=CC(OC=2C=CC=CC=2)=C1 SBNFWQZLDJGRLK-UHFFFAOYSA-N 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- 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
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/20—Education
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- General Engineering & Computer Science (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Artificial Intelligence (AREA)
- Tourism & Hospitality (AREA)
- Educational Technology (AREA)
- Educational Administration (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Electrically Operated Instructional Devices (AREA)
Abstract
The invention discloses a kind of stipulations method and system of mathematics subjective item answer step, this method includes：Answering information to be read is received, and word segmentation processing is carried out to the answering information to be read；Answer step separation is carried out to the answering information to be read according to word segmentation processing result, obtains answer step；The mathematical expression form of answer step is converted into stipulations representation.Mathematics subjective item can be answered step and be converted into stipulations representation by the present invention, easy to carry out structural analysis to answer step and answer result is read and appraised.
Description
Technical field
The present invention relates to natural language processing, deep learning field, and in particular to a kind of mathematics subjective item answer step
Stipulations method and system.
Background technology
During conventional teaching, teacher carries substantial amounts of workload for a long time as the main body taken an examination and operation is read and appraised.
Read and appraise and substantial amounts of duplicate message contained in work, especially mathematical solution answer read and appraise scene under, the result of answering of student is past
It is more obvious toward corresponding fixed knowledge point, repeatability.At the same time, as Internet technology and product are gradually to education sector
Infiltration, audient's scale of online education constantly expand, and further add the quantity that examination and operation are read and appraised.On the other hand,
When processing reads and appraises work on a large scale, teacher is easily disturbed be subject to subjective factors such as fatigue, personal preferences, so as to influence to comment
Read, the accuracy and objectivity especially to score.Therefore complete using computer or aid in completion to read and appraise, manually read and appraised with reducing
Workload, lifting reads and appraises, and the accuracy and objectivity especially to score is significant to teaching process.
Automatic for mathematical problem answer is read and appraised, and existing method is mainly：Based on carrying out matching acquisition with model answer
The method of appraisal result, this kind of scheme are primarily adapted for use in reading and appraising for objective item and the subjective item of non-opening.It is existing to read and appraise automatically
The topic type that method is limited primarily directed to expression-form, such as calculation question, gap-filling questions, and for open topic type, as proof question,
Answer is solved, effect is difficult to ensure that.This is because the representation of mathematics answer and the form of expression of traditional writing text exist
Significant difference, such as △ ABC, triangle ABC, existing semantic understanding method are not suitable for mathematics answer, cause mathematics answer
Automatically actual demand cannot be met by reading and appraising effect.
The content of the invention
The present invention provides a kind of stipulations method and system of mathematics subjective item answer step, to solve due to mathematics answer
There are significant difference, causing mathematics answer to read and appraise effect automatically cannot expire the form of expression of representation and traditional writing text
The problem of sufficient actual demand.
For this reason, the present invention provides following technical solution：
A kind of stipulations method of mathematics subjective item answer step, including：
Answering information to be read is received, and word segmentation processing is carried out to the answering information to be read；
Answer step separation is carried out to the answering information to be read according to word segmentation processing result, obtains answer step；
The mathematical expression form of answer step is converted into stipulations representation.
Preferably, it is described the mathematical expression form of answer step is converted into stipulations representation to include：
Preset stipulations representation；
The mathematical expression form of answer step is translated as stipulations representation.
Preferably, the stipulations representation is the nestable representation based on multi-component system, wherein, multi-component system bag
Include：Predicate and several argument relations：Argument.
Preferably, it is described the mathematical expression form of answer step is translated as stipulations representation to include：
Mathematical expression form is translated as by stipulations representation using stipulations model trained in advance, the stipulations model is adopted
With neutral net end to end, wherein, the input of neutral net is answer step, exports and is represented for stipulations.
Preferably, it is described that answer step separation is carried out to the answering information to be read according to word segmentation processing result, answered
Step is inscribed, including：
Answer step point is carried out to the answering information to be read based on word segmentation processing result and the substep model built in advance
From obtaining answer step.
Preferably, the substep model is neutral net, including：Vectorization module, multilayer retrieval module and classification
Module, wherein, the input of vectorization module is the word that word segmentation processing obtains, and the output of vectorization module is term vector sequence, more
The input of sequence of layer acquisition module is term vector sequence, and the output of multilayer retrieval module is sequence vector, sort module
It is the judging result for segmenting point as separating step point to input as sequence vector, the output of sort module.
Preferably, before word segmentation processing is carried out to the answering information to be read, the method further includes：
Mathematical entities identification is carried out to the answering information to be read；
It is described that the answering information progress word segmentation processing to be read is included：
Word segmentation processing is carried out to the answering information to be read based on the mathematical entities recognition result.
Preferably, after the mathematical expression form of answer step is converted into stipulations representation, the method is also wrapped
Include：
Answer step to stipulations representation carries out structural analysis, obtains relation between step；
Feature is read and appraised based on relation extraction first between answer step and step；
Feature is read and appraised using described first and what is built in advance reads and appraises model, and obtain the answering information to be read reads and appraises knot
Fruit.
Correspondingly, present invention also offers a kind of algorithm of mathematics subjective item answer step, including：
Answering information receiving module, for receiving answering information to be read；
Word-dividing mode, for carrying out word segmentation processing to the answering information to be read；
Answer step acquisition module, for carrying out answer step point to the answering information to be read according to word segmentation processing result
From obtaining answer step；
Protocol module, for the mathematical expression form of answer step to be converted into stipulations representation.
Preferably, the protocol module includes：
Form setup unit, for presetting stipulations representation；
Translation unit, for the mathematical expression form of answer step to be translated as stipulations representation.
Preferably, the protocol module is specifically used for being translated as mathematical expression form using stipulations model trained in advance
Stipulations representation, the stipulations model use neutral net end to end, wherein, the input of neutral net is answer step,
Export and represented for stipulations.
Preferably, the answer step acquisition module is specifically used for based on word segmentation processing result and the substep mould built in advance
Type carries out answer step separation to the answering information to be read, and obtains answer step, wherein, the substep model is nerve net
Network, including：Vectorization module, multilayer retrieval module and sort module, wherein, the input of vectorization module is word segmentation processing
Obtained word, the output of vectorization module are term vector sequence, and the input of multilayer retrieval module is term vector sequence, multilayer
The output of retrieval module is sequence vector, and the input of sort module is sequence vector, and the output of sort module is participle point
Judging result as separating step point.
Preferably, the system also includes：
Entity recognition module, for carrying out mathematical entities identification to the answering information to be read；
The word-dividing mode is specifically used for dividing the answering information to be read based on the mathematical entities recognition result
Word processing.
Preferably, the system also includes：
Relation acquisition module between step, for carrying out structural analysis to each answer step of stipulations representation, is walked
Relation between rapid；
Characteristic extracting module is read and appraised, for reading and appraising feature based on relation extraction first between answer step and step；
Read and appraise module, for reading and appraising feature using described first and what is built in advance reads and appraises model, obtain it is described wait to read answer
Topic information reads and appraises result.
The stipulations method and system of mathematics subjective item answer step provided in an embodiment of the present invention, are receiving answer letter to be read
Breath, and word segmentation processing is carried out to the answering information to be read；The answering information to be read is answered according to word segmentation processing result
Step separation is inscribed, obtains answer step；The mathematical expression form of answer step is converted into stipulations representation.Since mathematics is answered
The situations of numerous expression of equal value, such as " AB//CD " and " AB is parallel with CD ", the number of the invention by answer step are included in topic information
Learn expression-form and be converted into stipulations representation, effectively lift the accuracy subsequently read and appraised, the precision especially to score.
Further, stipulations representation described in the embodiment of the present invention are the nestable expression shape based on multi-component system
Formula, is so conducive to that subsequently relation between step is stored and shown in the form of a tree.
Further, the stipulations method and system of mathematics subjective item provided in an embodiment of the present invention answer step, can be with
After the answer step of stipulations expression is obtained, structural analysis is carried out to the answering information to be read, can be obtained by the step
Relation between the answering information of answer to be read is taken, since relation can characterize the answer logic of answer person between answer step and step,
It can be so compared according to the answer logic of answer person and the answer logic of Key for Reference, obtain first and read and appraise feature, into
And read and appraise feature according to first and treat and read answering information and read and appraised, avoid the prior art need to carry out model answer it is whole
Reason, extension, and the expression-form of the answer of answer person, answer thinking may be various, cause spreading result can not cover institute
Possible answer, causes to read and appraise the incorrect situation generation of result, effectively lift open topic type reads and appraises the accurate of result
Degree.
Brief description of the drawings
In order to illustrate the technical solutions in the embodiments of the present application or in the prior art more clearly, below will be to institute in embodiment
Attached drawing to be used is needed to be briefly described, it should be apparent that, drawings in the following description are only one described in the present invention
A little embodiments, for those of ordinary skill in the art, can also obtain other attached drawings according to these attached drawings.
Fig. 1 is the first flow chart of the stipulations method of mathematics subjective item answer step provided in an embodiment of the present invention；
Fig. 2 is a kind of stream of the method provided in an embodiment of the present invention that mathematical expression form is converted into stipulations representation
Cheng Tu；
Fig. 3 is second of flow chart of the stipulations method of mathematics subjective item answer step provided in an embodiment of the present invention；
Fig. 4 is the first structural representation of the algorithm of mathematics subjective item answer step provided in an embodiment of the present invention
Figure；
Fig. 5 is second of structural representation of the algorithm of mathematics subjective item answer step provided in an embodiment of the present invention
Figure.
Embodiment
In order to make those skilled in the art more fully understand the scheme of the embodiment of the present invention, below in conjunction with the accompanying drawings and implement
Mode is described in further detail the embodiment of the present invention.
For the representation of mathematics answer and the form of expression of traditional writing text there are significant difference, cause mathematics
The problem of effect cannot meet actual demand, is read and appraised in answer automatically, and the present invention provides a kind of rule of mathematics subjective item answer step
About method, the mathematical expression form for the answer step that will identify that are converted into stipulations representation, are answered in order to subsequently treat to read
Topic information is read and appraised.
As shown in Figure 1, it is a kind of flow of the stipulations method of mathematics subjective item answer step provided in an embodiment of the present invention
Figure.
In the present embodiment, the stipulations method of the mathematics subjective item answer step may comprise steps of：
Step S01, receives answering information to be read, and carries out word segmentation processing to the answering information to be read.
Wherein, answering information to be read can be answer image information and/or answer text message, when in answering information to be read
During including answer image information, answer text message can be obtained by image recognition.For example, answer image can pass through bat
Answer image is obtained according to means such as, scannings, and then image is identified using OCR technique, obtains answer text message.When
So, answer text message can also be the text message that the modes such as computer answer input, and not limit herein.
On word segmentation processing, existing common segmenting method can be used, such as rule-based method, based on sequence
Model prediction method of mark etc., does not limit this this case.
In another embodiment, in order to improve the accuracy of follow-up word segmentation processing, process is segmented in mathematics answer content
In first mathematical entities can be identified, to ensure the accuracy of final word segmentation result and serviceability.On mathematical entities
Identification, specifically can by the symbol for the mark mathematic(al) object being likely to occur in mathematics answer content, as triangle ABC, angle BAC,
Parallelogram ABCD etc., is defined as mathematical entities, because under normal conditions, expression way of the mathematical entities in answer result compared with
It is limited, therefore rule-based matched method detection mathematical entities can be used to be obtained with higher recognition accuracy,
Neutral net etc. can certainly be used to carry out mathematical entities identification, correspondingly, should if having carried out mathematical entities identification
Word segmentation processing is carried out to the answering information to be read based on the mathematical entities recognition result.
, can be with order to further lift the accuracy of word segmentation processing in addition, when answering information is answer image information
Answer image information is segmented first, the processing such as branch, can be with for example, the hand-written answer for mathematics open-ended question
Check whether there is adhesion row in the answer image information, and adhesion row is split, identify in the answer image information
The special mathematic sign such as fraction line, so as to correct branch, such as nearest row carries out being incorporated as one up and down by fraction line
Mathematics answer row etc., can so carry out answer image information accurate branch, in order to subsequently carry out word segmentation processing.
Step S02, carries out answer step separation to the answering information to be read according to word segmentation processing result, obtains answer step
Suddenly.
Specifically, due to the form of answering information to be read, layout the problems such as, the form of content of finally answering is often not
It is controllable, such as comprising the unsegregated problem of multiple mathematical steps, for example, including multiple answer steps, or adjacent lines with a line
For an answer step, therefore the present invention needs each answer step to be separated in answering information to be read, and obtains each answer step
Suddenly.Specifically, it will can each segment and a little be used as candidate's separating step point, then, the method based on model judgement obtains each
Whether candidate's separating step point is point step by step, wherein, above-mentioned model can be neutral net, its training method can be compared with
Technology, such as gathers training data first, which can be to have the answering information for putting markup information step by step, by training number
According to answering information word segmentation result, such as each word input neutral net, the parameter for adjusting neutral net make it that neutral net is defeated
Go out result and constantly level off to correct markup information, if which participle point is point step by step, which participle point is not point step by step,
When the similarity of neutral net output result and standard results is more than given threshold, then it is assumed that model training is completed, Ran Houke
To utilize each separating step point in trained model prediction answering information to be read.
It is in the present embodiment, described that answer step separation is carried out to the answering information to be read according to word segmentation processing result,
Obtaining answer step includes：The answering information to be read is answered based on word segmentation processing result and the substep model built in advance
Step separation is inscribed, obtains answer step.
Wherein, the substep model is neutral net, including：Vectorization module, multilayer retrieval module and classification mould
Block, wherein, the word that the input of vectorization module is obtained for word segmentation processing, the output of vectorization module is term vector sequence, multilayer
The input of retrieval module is term vector sequence, and the output of multilayer retrieval module is sequence vector, sort module it is defeated
It is the judging result for segmenting point as separating step point to enter for sequence vector, the output of sort module.
In a specific embodiment, it will each segment and a little be used as candidate's separating step point, the method adjudicated based on model,
Obtain each candidate's separating step point whether be separating step point judging result.Illustrated by taking neutral net as an example, model
Predominantly vectorization module+multilayer retrieval module+sort module, the word one by one in the answering information after will segmenting,
Continuous input obtains term vector sequence to vectorization module, then by the term vector sequence inputting at most sequence of layer acquisition module, example
Such as long memory network (Long Short-Term Memory, LSTM), Recognition with Recurrent Neural Network (Recurrent Neural in short-term
Networks, RNN) etc., neutral net is directed to candidate's separating step point, a sequence vector is obtained, finally by the sequence vector
Input to sort module, obtain each candidate's separating step point whether be separating step point judging result.Above-mentioned judging result can
To be candidate's separating step point as the score of separating step point or certain candidate's separating step point is separating step point
Conclusion.
Step S03, stipulations representation is converted into by the mathematical expression form of answer step.
For example, carrying out stipulations in the form of single order predicate logic, the specific manifestation form of the single order predicate logic can be
Representation based on multi-component system, can also be represented with AMR abstract semantics, this this case is not limited
As shown in Fig. 2, it is the method provided in an embodiment of the present invention that mathematical expression form is converted into stipulations representation
A kind of flow chart, it is described the mathematical expression form of answer step is converted into stipulations representation to may comprise steps of：
Step S21, stipulations representation is preset.
Wherein, it is contemplated that mathematical linguistics is mainly the relation between mathematical entities are expressed, and single order is used in the present embodiment
Predicate logic is as stipulations representation.The single order predicate logic can pass through the nestable representation based on multi-component system
It is indicated, wherein, multi-component system includes：Predicate and several argument relations：Argument.The multi-component system refers to predicate Predicate
Argument argument (1) ... argument argument (n), wherein, n is the natural number more than or equal to 2.For example, AB is parallel to CD,
Predicate is parallel, and argument is respectively AB, CD；The nestable multi-component system substantially can be understood as to triple
The extension of form, including：
1) number of argument is not limited to two, while defines the relation of argument and predicate with " argument relation ", and form is such as：
Predicate argument relation 1：Argument argument relation 2：Argument ... argument relations n：Argument.
2) structure can be nested, that is to say, that argument can be sub- triple, and form is such as：
Predicate argument relation：(predicate argument relation：Argument).
In addition, the specific manifestation form of single order predicate logic can also be represented with AMR abstract semantics, this this case is not limited
It is fixed.Correspondingly, relation includes between above-mentioned answering information：Relation between the answer step and step of stipulations representation.
Step S22, the mathematical expression form of answer step is translated as stipulations representation.
Conversion from mathematical expression to stipulations representation is substantially a kind of translation process of language to another language,
Therefore the principle of herein by reference machine network, it is stipulations representation that mathematical expression is converted (namely translation).It can specifically adopt
Mathematical expression form is translated as stipulations representation with stipulations model trained in advance, the stipulations model can be arrived using end
The neutral net at end, wherein, the input of neutral net is answer step, exports and is represented for stipulations.
In a specific embodiment, represented using neural network model end to end to carry out mathematic(al) representation to stipulations
The conversion of form, for example, end to end neural network model can with coder-decoder structure, encoder using two-way LSTM as
Example, decoder is with unidirectional LSTM, and the input of model is a step, and the output of model represents for stipulations, with the number of tree form
According to example is carried out, such as it is expressed as：Intersecting (line (AB, CD)), intersection point (O)).The training of model is same as existing common neutral net instruction
Practice process, no longer elaborate herein.
The embodiment of the present invention is in order to lift the accuracy of relation between the answering information of acquisition, by the mathematical expression of answer step
Form is converted into stipulations representation.Due to including the situation of numerous expression of equal value, normalized emphasis in mathematics answering information
The conversion two that design, mathematical expression including stipulations representation are represented to stipulations walks greatly；Further, the stipulations represent shape
Formula can be the representation based on triple, so be conducive to that subsequently relation between step is stored and shown in the form of a tree.
As shown in figure 3, it is second of stream of the stipulations method of mathematics subjective item answer step provided in an embodiment of the present invention
Cheng Tu.In the present embodiment, the stipulations method of mathematics subjective item answer step, can also comprise the following steps：
Step S31, the answer step to stipulations representation carry out structural analysis, obtain relation between step.
Specifically, the scoring of mathematics answer needs to consider the relation between answer step, therefore in order to further lift scoring
Precision or provide answer intermediate logic process it is whether correct or complete, sequence of steps is parsed into tree herein, to retouch
State the relation between step and step.
In the present embodiment, the structural analysis is mainly relationship analysis between step, wherein, relation is used to characterize between step
Answer logic.For example, can be tree by relation decomposing between step, to describe the relation between step and step.Wherein,
Between the step relation can include it is following any one：Derive, arranged side by side and repetition.Wherein, derivation relation can characterize answer person
Answer logic, which can include：Derivation condition, derive conclusion etc., derives condition to be set up so that deriving conclusion
The condition of institute's foundation, for example, it is topic condition, known quantitative and according to topic condition and/or the known centre quantitatively derived
Conclusion.
For example, in following answer content：" because step 1, step 2 ", step 1 are the derivations of step 2
Condition, therefore step 1 and step 2 are " derivation " modified relationships；Similar, in mathematics answer step relationship analysis, further include
" arranged side by side ", " repetition " and etc. between relation.
For example, above-mentioned relation analysis model is convolutional neural networks, including：Input layer, convolutional layer, classification layer and output
Layer, the input of input layer are answer step vector, and the output of convolutional layer is for determining that the distribution of relation between step is special
Sign vector, the input for layer of classifying is vectorial for the statistical nature of the rule-based extraction of distributed nature vector sum, the output of output layer
The judging result of relation between step.It should be noted that the answer step vector can be the term vector sequence of answer step,
It can also be a vector value of answer step, not limit herein.
The present embodiment is based on relation between neutral net, such as regression model or disaggregated model obtaining step, below with convolution
Illustrated exemplified by neutral net CNN, the structure of model mainly includes input layer, convolutional layer, classification layer and output layer.Input layer
What is inputted is the answer step vector of the answer step of pending analysis, which then accesses convolutional layer,
Convolutional layer uses Multi-layer design, it is therefore an objective to extracts the feature of different level of abstractions, the output of final convolutional layer is to be used to step close
The definite distributed nature vector of system, by the distributed nature and the statistical nature of rule-based extraction vector, input together to
Classification layer, the output of final CNN is the judging result of the relation between step, wherein, above-mentioned judging result can directly be represented
Relation object is other to state or belongs to the probability of each classification.
The statistical nature mainly includes architectural feature, introducer feature, step linked character, keyword feature.
The length of architectural feature --- step and the position in answer, such as which step.
Introducer feature --- the information of contained introducer in step, such as because,.
Step linked character --- the relation obtained between step to be analyzed according to rule judges.
Keyword feature --- the information of keyword is included in step to be analyzed, wherein, the information of the keyword is advance
The information of the keyword of setting.
It should be noted that：Relation between the step of for analyzing, can store and show in the form of a tree, based on relation
The method of spanning tree, can use some existing ripe spanning tree algorithms, the algorithm such as based on state transfer, the calculation based on figure
Method etc., does not limit herein.
The method provided in an embodiment of the present invention that structural analysis is carried out to the answering information to be read, can be automatically from waiting to read
Extracting relation between answering information in answering information, relation can characterize the answer logic of answer person between the answer step, so as to
Obtain reading and appraising feature in based on relation between the answering information.
Step S32, feature is read and appraised based on relation extraction first between the answering information.
In the present embodiment, it is described be based on the answering information between relation extraction first read and appraise feature and can include following step
Suddenly：
Step a, predefine first and read and appraise feature, described first read and appraise feature include it is following any one or more：Close
Key steps characteristic, the derivation relationship characteristic of committed step, answer result feature.
Wherein, read and appraise whether feature mainly occurs including committed step, whether the derivation relation of committed step is abundant, answer
As a result it is whether correct.The derivation condition whether the derivation relation of committed step is fully referred mainly between committed step and previous step is
It is no correct.
Committed step can corresponding answer step and/or the key manually marked when relation is derivation relation between step
Step, wherein, deriving the corresponding answer step of relation can be step corresponding to derivation condition and/or derive the corresponding step of conclusion
Suddenly.In the prior art, committed step is usually all that expert carries out model answer according to experience committed step mark come really
Determine committed step, still, there may be a variety of, to be obtained by way of extension answers to the expression way of same problem in reality
Logic and expression way tend not to the answering mode for covering all correctly answer logics, can not be by way of labor standard
Committed step mark is carried out to all correct options, therefore, it is automatic that the prior art can not carry out machine to open subjective item
Go over examination papers, for example, once there is the model answer for omitting logic or expression way, then can cause that it fails to match, but this is missed
Answer be also correct option, cause computer to read and appraise result automatically incorrect.In addition, the committed step that manually marks may be because
The committed step for causing mark for factors such as personal experiences differs, and is not easy to the answer step to various possible expression-forms
Manually marked.
In the present embodiment, committed step is determined using derivation relation and/or the mode manually marked, for example, lacking
, can be by relation between step due to can be by deriving the answer logic of relation table question and answer game topic person during artificial markup information
Corresponding answer step is as committed step during to derive relation.Certainly, if the information manually marked, key can be caused
The identification of step is more comprehensive.
Specifically, above-mentioned read and appraise is characterized in extracting for the answering information of each topic, as whether committed step occurs
The vector of a multidimensional can be shown as, each dimension represents whether each committed step occurs in the topic respectively；Latter two is special
Sign can be handled equally.
Step b, extracts relation between the answering information of Key for Reference in advance.
The extracting mode of relation between the answering information of Key for Reference, may be referred to close between the answering information of answering information to be read
System, this will not be detailed here.
It should be noted that the Key for Reference can be it is following any one or more：Model answer, to model answer into
The full marks answer etc. of correct option, other persons of answering that row extension obtains, does not limit herein.
Step c, by relation between the answering information of relation and the Key for Reference between the answering information of answering information to be read into
Row compares, and obtains first and reads and appraises feature.
Relation can be understood as treating the semantic understanding process for readding answering information between the answering information, can be with by the process
Obtain characterization answer logic answering information between relation, then can utilize the answering information between relation and Key for Reference answer
Relation is compared between information, so can be obtained by this and first reads and appraises feature, for example, whether committed step occurs, key step
Whether rapid derivation relation is abundant, whether answer result is correct.
It should be noted that the extraction of these above-mentioned features can be used but not limited to the feature extraction based on engineer
Method and the feature extracting method based on neutral net.
Step S33, reads and appraises feature using described first and what is built in advance reads and appraises model, obtains the answering information to be read
Read and appraise result.
In the present embodiment, read and appraise feature based on extracted first, by build regression model or disaggregated model come
Realize Rating Model.Regression model can use linear regression model (LRM) and nonlinear regression model (NLRM).It is complex in step relation
Topic type in, the effect of nonlinear regression model (NLRM) is more preferable.
In a specific embodiment, the model of reading and appraising is nonlinear regression model (NLRM), the output bag for reading and appraising model
Include：Read and appraise fraction and/or evaluation.
It should be noted that in order to more accurately be corrected to complex, horizontal higher topic type, it is described to read and appraise
Model further includes convolutional layer, and the input of the convolutional layer is answer step vector, and the output of the convolutional layer reads and appraises spy for second
Sign, including the step of different grain size is interior and step between information, read and appraised by adjusting described in convolution nuclear parameter and the training of the convolution number of plies
The convolutional layer of model.
It is described to read and appraise feature using described first and what is built in advance reads and appraises model, obtain commenting for the answering information to be read
Readding result includes：Feature, answer step vector are read and appraised using described first and what is built in advance read and appraise model, obtain described waiting to read
Answering information reads and appraises result.
Specifically, the feature extracting method based on neutral net is mainly that correlated characteristic information is embedded in by planned network structure
The hidden layer of neutral net, it is more effective for complex, horizontal higher topic type.For example, using reading and appraising with convolutional layer
Model extraction second reads and appraises feature, and convolutional layer can be using term vector sequence as input, by adjusting convolution nuclear parameter and convolutional layer
The interior information between step of the step of counting, extracting different grain size, feature is read and appraised as second.
That is, it is described read and appraise model can be based on described first read and appraise feature obtain reading and appraising result or
Feature is read and appraised by first, and second obtained by the term vector sequence and convolution of answer step reads and appraises feature, it is common to reading and appraising
The recurrence layer of model, obtains reading and appraising result.
The stipulations method of mathematics subjective item answer step provided in an embodiment of the present invention, also further to the answer to be read
Information carries out structural analysis, relation between the answering information of answer to be read can be obtained by the step, due to answer step and step
Relation can characterize the answer logic of answer person between rapid, and the answer of answer logic and Key for Reference that so can be to answer person is patrolled
Collect and be compared, obtain first and read and appraise feature, and then read and appraise feature by first and read and appraised, avoiding the prior art can not be right
Model answer arranged, extend after, realize and cover all possible answer, cause to read and appraise the incorrect situation of result and occur,
The effectively accuracy for reading and appraising result of the open topic type of lifting.
As shown in figure 4, it is the first knot of the algorithm of mathematics subjective item answer step provided in an embodiment of the present invention
Structure schematic diagram.
In the present embodiment, the algorithm of mathematics subjective item answer step can include：
Answering information receiving module 401, for receiving answering information to be read；
Word-dividing mode 402, for carrying out word segmentation processing to the answering information to be read；
Answer step acquisition module 403, for carrying out answer step to the answering information to be read according to word segmentation processing result
Rapid separation, obtains answer step；
Protocol module 404, for the mathematical expression form of answer step to be converted into stipulations representation.
Preferably, the system also includes：
Entity recognition module 405, for carrying out mathematical entities identification to the answering information to be read；
The word-dividing mode 402 is specifically used for carrying out the answering information to be read based on the mathematical entities recognition result
Word segmentation processing.
Wherein, the protocol module 404 can include：
Form setup unit, for presetting stipulations representation；
Translation unit, for the mathematical expression form of answer step to be translated as stipulations representation.
Specifically, the protocol module 404 is specifically used for turning over mathematical expression form using stipulations model trained in advance
Stipulations representation is translated into, the stipulations model uses neutral net end to end, wherein, the input of neutral net is that answer walks
Suddenly, export and represented for stipulations.
In a specific embodiment, the answer step acquisition module 403 be specifically used for based on word segmentation processing result and
The substep model built in advance carries out answer step separation to the answering information to be read, and obtains answer step, wherein, described point
Step model is neutral net, including：Vectorization module, multilayer retrieval module and sort module, wherein, vectorization module
The word obtained for word segmentation processing is inputted, the output of vectorization module is term vector sequence, and the input of multilayer retrieval module is
Term vector sequence, the output of multilayer retrieval module are sequence vector, and the input of sort module is sequence vector, sort module
Output be judging result of the participle point as separating step point.
The algorithm of mathematics subjective item answer step provided by the invention, the mathematical expression form of answer step is converted
, so can be in order to subsequently so that computer be automatically to mathematics subjective item, especially open mathematics for stipulations representation
The answer result of subjective item carries out structural analysis, and then is read and appraised according to analysis result, can effectively lift answer result and read and appraise
Efficiency and accuracy.
As shown in figure 5, it is second of knot of the algorithm of mathematics subjective item answer step provided in an embodiment of the present invention
Structure schematic diagram.
In the present embodiment, the system also includes：
Relation acquisition module 501 between step, for carrying out structural analysis to each answer step of stipulations representation, obtains
Relation between step；
Characteristic extracting module 502 is read and appraised, for reading and appraising feature based on relation extraction first between answer step and step；
Module 503 is read and appraised, for reading and appraising feature using described first and what is built in advance reads and appraises model, obtains described waiting to read
Answering information reads and appraises result.
Relation acquisition module 501 between above-mentioned steps, read and appraise characteristic extracting module 502 and read and appraise 503 side of may be referred to of module
Related content in method, this will not be detailed here.
In the specification that this place provides, numerous specific details are set forth.It is to be appreciated, however, that the implementation of the present invention
Example can be put into practice in the case of these no details.In some instances, known method, structure is not been shown in detail
And technology, so as not to obscure the understanding of this description.
Similarly, it will be appreciated that in order to simplify the present invention and help to understand one or more of each inventive aspect,
Above in the description to the exemplary embodiment of the present invention, each feature of the invention is grouped together into single implementation sometimes
In example, figure or descriptions thereof.However, the method for the disclosure should be construed to reflect following intention：I.e. required guarantor
The application claims of shield features more more than the feature being expressly recited in each claim.
Those skilled in the art, which are appreciated that, to carry out adaptively the module in the device in embodiment
Change and they are arranged in one or more equipment different from the embodiment.Can be the module or list in embodiment
Member or component be combined into a module or unit or component, and can be divided into addition multiple submodule or subelement or
Sub-component.In addition at least some in such feature and/or process or unit exclude each other, it can use any
Combination is disclosed to all features disclosed in this specification (including adjoint claim, summary and attached drawing) and so to appoint
Where all processes or unit of method or equipment are combined.Unless expressly stated otherwise, this specification (including adjoint power
Profit requires, summary and attached drawing) disclosed in each feature can be by providing the alternative features of identical, equivalent or similar purpose come generation
Replace.
In addition, it will be appreciated by those of skill in the art that although embodiment described herein is included in other embodiments
Included some features rather than further feature, but the combination of the feature of different embodiments means the model in the present invention
Within enclosing and form different embodiments.For example, in the following claims, embodiment claimed it is any
One of mode can use in any combination.
It should be noted that above-described embodiment is that the present invention will be described rather than limits the invention, and
Those skilled in the art can design alternative embodiment without departing from the scope of the appended claims.Positioned at element it
Preceding word "a" or "an" does not exclude the presence of multiple such elements.The present invention can be by means of including some different members
The hardware of part and realized by means of properly programmed computer.In if the unit claim of dry systems is listed, this
Several in a little systems can be embodied by same hardware branch.Word first, second and third make
With not indicating that any order.These words can be construed to title.
Claims (14)
- A kind of 1. stipulations method of mathematics subjective item answer step, it is characterised in that including：Answering information to be read is received, and word segmentation processing is carried out to the answering information to be read；Answer step separation is carried out to the answering information to be read according to word segmentation processing result, obtains answer step；The mathematical expression form of answer step is converted into stipulations representation.
- 2. according to the method described in claim 1, it is characterized in that, described be converted into rule by the mathematical expression form of answer step About representation includes：Preset stipulations representation；The mathematical expression form of answer step is translated as stipulations representation.
- 3. according to the method described in claim 2, it is characterized in that, the stipulations representation is based on multi-component system to be nestable Representation, wherein, multi-component system includes：Predicate and several argument relations：Argument.
- 4. according to the method described in claim 2, it is characterized in that, described be translated as advising by the mathematical expression form of answer step About representation includes：Mathematical expression form is translated as by stipulations representation using stipulations model trained in advance, the stipulations model uses end To the neutral net at end, wherein, the input of neutral net is answer step, exports and is represented for stipulations.
- 5. according to the method described in claim 1, it is characterized in that, described believe the answer to be read according to word segmentation processing result Breath carries out answer step separation, obtains answer step, including：Answer step is carried out to the answering information to be read to separate, obtain based on word segmentation processing result with the substep model built in advance To answer step.
- 6. according to the method described in claim 5, it is characterized in that, the substep model is neutral net, including：Vectorization mould Block, multilayer retrieval module and sort module, wherein, vectorization module inputs the word obtained for word segmentation processing, vectorization The output of module is term vector sequence, and the input of multilayer retrieval module is term vector sequence, multilayer retrieval module Export as sequence vector, the input of sort module is sequence vector, and the output of sort module is used as separating step point for participle point Judging result.
- 7. method according to any one of claims 1 to 6, it is characterised in that divide to the answering information to be read Before word processing, the method further includes：Mathematical entities identification is carried out to the answering information to be read；It is described that the answering information progress word segmentation processing to be read is included：Word segmentation processing is carried out to the answering information to be read based on the mathematical entities recognition result.
- 8. method according to any one of claims 1 to 6, it is characterised in that by the mathematical expression form of answer step It is converted into after stipulations representation, the method further includes：Answer step to stipulations representation carries out structural analysis, obtains relation between step；Feature is read and appraised based on relation extraction first between answer step and step；Feature is read and appraised using described first and what is built in advance reads and appraises model, and obtain the answering information to be read reads and appraises result.
- A kind of 9. algorithm of mathematics subjective item answer step, it is characterised in that including：Answering information receiving module, for receiving answering information to be read；Word-dividing mode, for carrying out word segmentation processing to the answering information to be read；Answer step acquisition module, for carrying out answer step separation to the answering information to be read according to word segmentation processing result, Obtain answer step；Protocol module, for the mathematical expression form of answer step to be converted into stipulations representation.
- 10. system according to claim 9, it is characterised in that the protocol module includes：Form setup unit, for presetting stipulations representation；Translation unit, for the mathematical expression form of answer step to be translated as stipulations representation.
- 11. system according to claim 10, it is characterised in that the protocol module is specifically used for using training in advance Mathematical expression form is translated as stipulations representation by stipulations model, and the stipulations model uses neutral net end to end, its In, the input of neutral net is answer step, exports and is represented for stipulations.
- 12. system according to claim 9, it is characterised in that the answer step acquisition module is specifically used for being based on dividing Word handling result carries out answer step to the answering information to be read with the substep model built in advance and separates, and obtains answer step Suddenly, wherein, the substep model is neutral net, including：Vectorization module, multilayer retrieval module and sort module, its In, the input of vectorization module is the word that word segmentation processing obtains, and the output of vectorization module is term vector sequence, and more sequence of layer obtain The input of modulus block is term vector sequence, and the output of multilayer retrieval module is sequence vector, and the input of sort module is sequence Column vector, the output of sort module is judging result of the participle point as separating step point.
- 13. according to claim 9 to 12 any one of them system, it is characterised in that the system also includes：Entity recognition module, for carrying out mathematical entities identification to the answering information to be read；The word-dividing mode is specifically used for carrying out at participle the answering information to be read based on the mathematical entities recognition result Reason.
- 14. according to claim 9 to 12 any one of them system, it is characterised in that the system also includes：Relation acquisition module between step, for carrying out structural analysis to each answer step of stipulations representation, obtains between step Relation；Characteristic extracting module is read and appraised, for reading and appraising feature based on relation extraction first between answer step and step；Module is read and appraised, for reading and appraising feature using described first and what is built in advance reads and appraises model, obtains the answer letter to be read Breath reads and appraises result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711437048.9A CN107992482B (en) | 2017-12-26 | 2017-12-26 | Protocol method and system for solving steps of mathematic subjective questions |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711437048.9A CN107992482B (en) | 2017-12-26 | 2017-12-26 | Protocol method and system for solving steps of mathematic subjective questions |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107992482A true CN107992482A (en) | 2018-05-04 |
| CN107992482B CN107992482B (en) | 2021-12-07 |
Family
ID=62041456
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711437048.9A Active CN107992482B (en) | 2017-12-26 | 2017-12-26 | Protocol method and system for solving steps of mathematic subjective questions |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107992482B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109684980A (en) * | 2018-09-19 | 2019-04-26 | 腾讯科技（深圳）有限公司 | Automatic marking method and device |
| CN112287083A (en) * | 2020-10-29 | 2021-01-29 | 北京乐学帮网络技术有限公司 | Evaluation method and device, computer equipment and storage device |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130295537A1 (en) * | 2011-01-11 | 2013-11-07 | Iscilab Corporation | Mathematics education service system, service method thereof, apparatus for analyzing and generating mathematical problems, and method thereof |
| CN104933158A (en) * | 2015-06-26 | 2015-09-23 | 百度在线网络技术（北京）有限公司 | Training method and device and reasoning method and device for mathematical problem solving model |
| CN106096564A (en) * | 2016-06-17 | 2016-11-09 | 福建网龙计算机网络信息技术有限公司 | A kind of mathematics corrects method automatically |
| CN106846203A (en) * | 2017-03-09 | 2017-06-13 | 常州工学院 | A kind of intelligently reading method and system |
| CN106897384A (en) * | 2017-01-23 | 2017-06-27 | 科大讯飞股份有限公司 | One kind will bring out the theme automatic evaluation method and device |
| CN107153640A (en) * | 2017-05-08 | 2017-09-12 | 成都准星云学科技有限公司 | A kind of segmenting method towards elementary mathematics field |
| CN107169043A (en) * | 2017-04-24 | 2017-09-15 | 成都准星云学科技有限公司 | A kind of knowledge point extraction method and system based on model answer |
| CN107168949A (en) * | 2017-04-24 | 2017-09-15 | 成都准星云学科技有限公司 | Mathematics natural language processing implementation method, system based on combination of entities |
| US20170337180A1 (en) * | 2016-05-23 | 2017-11-23 | Lishan Wang | Recognition method and system of natural language for machine thinking |
| CN107423286A (en) * | 2017-07-05 | 2017-12-01 | 华中师范大学 | The method and system that elementary mathematics algebraically type topic is answered automatically |
| CN107463553A (en) * | 2017-09-12 | 2017-12-12 | 复旦大学 | For the text semantic extraction, expression and modeling method and system of elementary mathematics topic |
-
2017
- 2017-12-26 CN CN201711437048.9A patent/CN107992482B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130295537A1 (en) * | 2011-01-11 | 2013-11-07 | Iscilab Corporation | Mathematics education service system, service method thereof, apparatus for analyzing and generating mathematical problems, and method thereof |
| CN104933158A (en) * | 2015-06-26 | 2015-09-23 | 百度在线网络技术（北京）有限公司 | Training method and device and reasoning method and device for mathematical problem solving model |
| US20170337180A1 (en) * | 2016-05-23 | 2017-11-23 | Lishan Wang | Recognition method and system of natural language for machine thinking |
| CN106096564A (en) * | 2016-06-17 | 2016-11-09 | 福建网龙计算机网络信息技术有限公司 | A kind of mathematics corrects method automatically |
| CN106897384A (en) * | 2017-01-23 | 2017-06-27 | 科大讯飞股份有限公司 | One kind will bring out the theme automatic evaluation method and device |
| CN106846203A (en) * | 2017-03-09 | 2017-06-13 | 常州工学院 | A kind of intelligently reading method and system |
| CN107169043A (en) * | 2017-04-24 | 2017-09-15 | 成都准星云学科技有限公司 | A kind of knowledge point extraction method and system based on model answer |
| CN107168949A (en) * | 2017-04-24 | 2017-09-15 | 成都准星云学科技有限公司 | Mathematics natural language processing implementation method, system based on combination of entities |
| CN107153640A (en) * | 2017-05-08 | 2017-09-12 | 成都准星云学科技有限公司 | A kind of segmenting method towards elementary mathematics field |
| CN107423286A (en) * | 2017-07-05 | 2017-12-01 | 华中师范大学 | The method and system that elementary mathematics algebraically type topic is answered automatically |
| CN107463553A (en) * | 2017-09-12 | 2017-12-12 | 复旦大学 | For the text semantic extraction, expression and modeling method and system of elementary mathematics topic |
Non-Patent Citations (2)
| Title |
|---|
| 王勇: "基于一阶逻辑的知识表示与自动提取", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 王湘云: "一阶谓词逻辑在人工智能知识表示中的应用", 《重庆工学院学报（社会科学版）》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109684980A (en) * | 2018-09-19 | 2019-04-26 | 腾讯科技（深圳）有限公司 | Automatic marking method and device |
| CN109684980B (en) * | 2018-09-19 | 2022-12-13 | 腾讯科技（深圳）有限公司 | Automatic scoring method and device |
| CN112287083A (en) * | 2020-10-29 | 2021-01-29 | 北京乐学帮网络技术有限公司 | Evaluation method and device, computer equipment and storage device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107992482B (en) | 2021-12-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110110585B (en) | Intelligent paper reading implementation method and system based on deep learning and computer program | |
| CN108121702A (en) | Mathematics subjective item reads and appraises method and system | |
| CN107133220B (en) | Geographic science field named entity identification method | |
| CN108052504A (en) | Mathematics subjective item answers the structure analysis method and system of result | |
| CN110750959B (en) | Text information processing method, model training method and related device | |
| CN107025284A (en) | The recognition methods of network comment text emotion tendency and convolutional neural networks model | |
| CN107562918A (en) | A kind of mathematical problem knowledge point discovery and batch label acquisition method | |
| CN108182177A (en) | A kind of mathematics knowledge-ID automation mask method and device | |
| CN110825867B (en) | Similar text recommendation method and device, electronic equipment and storage medium | |
| CN105279495A (en) | Video description method based on deep learning and text summarization | |
| CN110929034A (en) | Commodity comment fine-grained emotion classification method based on improved LSTM | |
| Haug et al. | Neural multi-step reasoning for question answering on semi-structured tables | |
| CN108090099B (en) | Text processing method and device | |
| CN106897384A (en) | One kind will bring out the theme automatic evaluation method and device | |
| CN110689018A (en) | Intelligent marking system and processing method thereof | |
| Agarwal et al. | Autoeval: A nlp approach for automatic test evaluation system | |
| CN111553140A (en) | Data processing method, data processing apparatus, and computer storage medium | |
| CN107992482A (en) | Mathematics subjective item answers the stipulations method and system of step | |
| JP7303243B2 (en) | Exam question prediction system and exam question prediction method | |
| Saha et al. | The corporeality of infotainment on fans feedback towards sports comment employing convolutional long-short term neural network | |
| CN116521872B (en) | Combined recognition method and system for cognition and emotion and electronic equipment | |
| CN110929013A (en) | Image question-answer implementation method based on bottom-up entry and positioning information fusion | |
| CN116662924A (en) | Aspect-level multi-mode emotion analysis method based on dual-channel and attention mechanism | |
| CN115761235A (en) | Zero sample semantic segmentation method, system, equipment and medium based on knowledge distillation | |
| CN112686020B (en) | Composition scoring method and device, electronic equipment and storage medium |
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 |