Based on deep learning online education Students ' Comprehensive portrait tag control systemTechnical field
G06F electricity Digital data processing or G06Q are specially adapted for the number of educational forecasting purpose in classifying the invention belongs to IPCAccording to the portrait label technique in processing system or method, it is related to network and buries the necks such as point, image recognition, text mining, deep learningDomain is based especially on deep learning online education Students ' Comprehensive portrait tag control system.
Background technology
Online education, that is, e-Learning or long-distance education, on-line study generally referred to as refer in existing concept a kind ofNetwork-based learning behavior, it is similar to online training concept.
A kind of common collecting method that a technology is web analytics is buried, it in Website page key position by plantingEnter multistage code, serial behavior of the tracking user on each interface of platform, between event independently of each other.It can be used for establishing and useFamily is drawn a portrait, and personal behavior model is restored.Text Mining Technology refers to the selection of the expression and its characteristic item to text, is that text is dugBasic problem in pick, information retrieval.The computer that it converts structureless urtext to structuring can identify and locateThe information of reason, it is final to realize the excavation effective information from a large amount of texts to which founding mathematical models describe and replace textPurpose.Text semantic analysis is to identify the process of the semantic informations such as text subject, classification and meaning, in natural language processing, letterIt is all commonly used to cease the fields such as filtering, information classification, information retrieval, semantic excavation.Image recognition technology refers to utilizing computerImage is handled, analyzed and is understood, to identify the target of various different modes and to the technology of picture.It is by structureless figureAs being identified as the information that the computer of concrete structure can be identified and be handled.Image recognition visual sensor or camera andComputer simulates human eye and brain, carries out object identification, tracking and measurement, and then do graphics process and make computer understanding trueThe world.Image recognition technology has many application scenarios, such as：The various image scenes such as recognition of face, identification of taking pictures, object identificationIdentification.Deep learning technology has distributed nature expression, Automatic Feature Extraction, end-to-end machine learning and good extensive energyThe advantages such as power, in the successful application that many fields such as speech recognition, image recognition and natural language processing are attracted people's attention.TogetherWhen its powerful inducing ability but also it has very excellent effect on the problems such as label refines.For example, training classificationDiscrimination model needs a large amount of sample data successive ignition training, data that must have the essential characteristic for differentiating object, there is the different back ofs the bodyScape angle is distinguished, and data sample is abundanter, and the accuracy of identification of model is higher.
Patents documents disclose less.Such as：
Chinese patent application 201510944668.6 proposes that a kind of user based on big data draws a portrait method for building up and userDraw a portrait management system, using in time limit effective time user behavior and/or content establish casual user and draw a portrait, andUser behavior and/or content of the casual user portrait out of user's portrait in the middle succession and time limit effective time is set to matchDescriptive label attribute, and when active between user behavior and/or content and user's portrait in the time limit descriptive label categoryProperty when mismatching, then create descriptive label attribute in casual user draws a portrait.The present invention can realize according to user behavior and/Or effective maintenance that content-data draws a portrait to user, especially in user behavior and/or content, there is a situation where ranks to get over formula mutationUnder, it can quickly eliminate the accumulation data that user's Current Content and behavior are taken advantage but do not met in distribution statistics ruleInfluence.
Chinese patent application 201510965619.0 proposes a kind of user's portrait construction method, passes through distributed reptile firstIt crawls internet Various types of data and merges and get through to form mass knowledge library；Then the internet log that obtains and by internet log with knowKnow library and carries out matching generation user base label；In conjunction with telecom operators' distinctive customer relation management (CRM) data and geographyPosition data builds user property label；Weight is carried out to label data and decay factor processing generates user base portrait, andAnd personalized data mining can be done in conjunction with the feature and industry customer's data of industry user, generate the use for meeting sector applicationIt draws a portrait and service is externally provided in family.The advantage of the invention is that accurate the whole network user portrait, fully profit can be provided for clientWith internet data, provided conveniently for application services such as customer analysis, Products Show, precision marketings.
Chinese patent application 201510564860.2 discloses a kind of method of structure user portrait.Wherein, user is builtThe method of portrait includes：It obtains user's internet internet log data and pre-processes, surf the Internet to pretreated internetDaily record data carries out feature extraction, obtains the attributive character of user, is then based on the label of established multidimensional characteristic library trainingClassification, matches in multidimensional characteristic library according to the attributive character of user, the multidimensional attribute label of user is obtained, according to moreDimensional attribute label builds user's portrait.By the above-mentioned means, the present invention can construct holographic various dimensions user portrait, fromAnd it disclosure satisfy that the recommendation of operator/business/company fast accurate advertisement dispensing and the consumer behavior of user group.
Chinese patent application 201611162106.7 provides one kind for online education and synthesis teaching multimedia objectMethod and system.This be used for synthesize teaching multimedia object method include：According to the individual demand information of teaching objectMultiple multimedia resources are selected from multimedia resources database；Teaching multimedia object is synthesized using the multiple multimedia resource.Multiple multimedia resources are selected from multimedia resources database according to the individual demand information of teaching object and synthesize the more matchmakers of teachingBody object, the actual demand that can be directed to teaching object flexibly synthesizes and provides personalized teaching multimedia, to make teachingObject can more easily carry out being directed to inquiry learning, and better experience is brought for user.
Chinese patent application 201510054208.6 provides a kind of online education evaluation system, including：Knowledge mapping listMember, including the incidence relation between one or more knowledge points and one or more of knowledge points；Examination question unit, including oneOr multiple examination questions to be measured, the binding of one or more of each examination question to be measured and the knowledge point, and according to each describedThe weighted value for contacting determining each examination question to be measured of examination question to be measured and the knowledge point；Test and appraisal unit, according to the weightValue selected section or whole examination questions to be measured from the examination question to be measured are used as test and appraisal examination question, and transfer to measured person's test and appraisal；As a resultGeneration unit generates evaluating result report according to the evaluating result of the measured person.After adopting the above technical scheme, can incite somebody to actionRelation map between the knowledge point of test, and show that measured person grasps in which knowledge point in final test and evaluation reportNot enough.
The data in present education field are although various, but isolated island benefit is apparent, lack portraying and assessing to student system.
Invention content
The object of the present invention is to provide based on deep learning online education Students ' Comprehensive draw a portrait tag control system, with solveAbove-mentioned deficiency, can support Scientific Assessment towards student group and individual level ability and show, and then effectively instruct to learnThe optimization and promotion of raw culture scheme and content of courses plan.
The purpose of the present invention will be realized by following technical measures：It is drawn a portrait based on deep learning online education Students ' ComprehensiveTag control system, including sequentially connected data acquisition unit, data pre-processing unit, portrait label refine unit, andResult output and show unit；Wherein：
1) data acquisition unit is used to carry out the acquisition acquisition and storage of data from multiple data sources, specifically includes：
A. data acquisition unit acquires the information such as student's line upper mounting plate behavior path and content of the act simultaneously using a technology is buriedStorage；
B. data acquisition unit utilizes the brain activity state during E.E.G acquisition technique acquisition student online lower studyEtc. information and store；
C. data acquisition unit utilizes the sight focal position during viewpoint tracer technique acquisition student online lower studyEtc. information and store.
2) data that the data pre-processing unit gets acquisition carry out Data Integration, data audit and cleaning, toolBody includes：
A. Data Integration, identity-based card number or student status unifying identifier, will acquire multi-source heterogeneous data and carry out cross aboveTo fusion；
B. data cleansing carries out completeness and efficiency audit for the above gathered data, rejects or adjust nothing thereinValid value and exceptional value；
3) the portrait label refines unit and is used to screen important dimension information from basic data, refines representative portraitLabel, and top-down complete tag system is formed, it specifically includes：
A. deep learning technology is utilized, it includes student's to be refined from the information such as content of the act and action trail on student's lineContent-preference, behavior pattern, learning style, attitude towards study, Online Learning mode tag including learning performance simultaneously store；
B. the sight focal position during student online lower study is carried out with browsing content using image recognition technologyMatching, content and navigation patterns pattern are paid close attention in locking student's study, and in conjunction with the deciphering to E.E.G state, refinement includes learningRaw attention is horizontal, learns state tag under the line including student's viewpoint stability and stores；
C. Chinese word segmentation is carried out for the on-line off-line comment of student, mutual information calculates, in theme using Text Mining TechnologyHold and refine, refine target object, Sentiment orientation and the subject matter of Students ' Evaluation opinion, form students ' subjective attitude label and stores；
D. classification and upper layer induction ＆ summing-up are carried out to all kinds of labels using data mining technology, is formed top-down completeLabel system；
4) result output and show unit for that will draw a portrait label achievement, with appropriate logic and visual means intoRow show with scene application, specifically include：
A. the representative portrait label in data visualization technique selective system is utilized, with line chart, block diagram, scatterplotScheme, pie chart, map, thermodynamic chart, relational graph, crater blasting, the various forms including instrument board etc., in conjunction with table and word, carries out comprehensiveCombination, three-dimensional are presented；
B. it is option label to be opened, support user by screening to label with combine, locking particular student group,The guidance optimization and policy development being oriented.
Especially, the application method for tag control system of being drawn a portrait based on deep learning online education Students ' Comprehensive, including nineA step, step 1 to step 4 corresponding data collecting unit, step 5 to step 7 corresponding data pretreatment unit, step 8 correspond toLabel of drawing a portrait refines unit, and step 9 corresponds to result output and shows unit, wherein：
Step 1：Behavior on line is carried out to bury an acquisition；
The information such as student's line upper mounting plate behavior path and content of the act are acquired using a technology is buried；
Step 2：E.E.G acquisition conscientious to E.E.G state；
The information such as the brain activity state under acquiring student online using E.E.G acquisition technique during study；
Step 3：Viewpoint track acquisition is carried out to viewpoint track；
The information such as the sight focal position under acquiring student online using viewpoint tracer technique during study；
Step 4：Data prediction is carried out to initial data；
Collected multi-source heterogeneous data are carried out lateral fusion by identity-based card number or student status class unifying identifier；TogetherShi Jinhang completeness and efficiencies are audited, and invalid value therein and exceptional value are rejected or adjust；
Step 5：Deep learning processing is carried out based on the Online Learning pattern in basic data fairground；
Using deep learning technology, it includes in student to be refined from the information such as content of the act and action trail on student's lineHold preference, behavior pattern, learning style, attitude towards study, the Online Learning mode tag including learning performance；
Step 6：Image recognition processing is carried out based on learning state under the line in basic data fairground；
Sight focal position during being learnt student down online using image recognition technology and browsing content progressMatch, content and navigation patterns pattern are paid close attention in locking student's study, and in conjunction with the deciphering to E.E.G state, refinement includes studentAttention is horizontal, learns state tag under the line including student's viewpoint stability；
Step 7：Text mining processing is carried out based on the subjective study comment in basic data fairground；
Chinese word segmentation is carried out for the on-line off-line comment of student, mutual information calculates, subject content using Text Mining TechnologyIt refines, refines target object, Sentiment orientation and the subject matter of Students ' Evaluation opinion, form students ' subjective attitude label；
Step 8：Portrait label system construction；
Classification and upper layer induction ＆ summing-up are carried out to all kinds of labels using data mining technology, form top-down complete markLabel system；
Step 9：Portrait system output with show；
In a manner of system, the achievement for label of drawing a portrait is showed and scene with appropriate logic and visual meansUsing.
Advantages of the present invention and effect：Utilize the cutting edge technologies such as a technology of burying, E.E.G acquisition technique, viewpoint tracer technique realityThe acquisition of comprehensive data now on-line off-line to student, it is complete using Text Mining Technology, image recognition technology and deep learning technologyThe efficient of representative portrait label concludes and efficiently refines in pairs, final portrait label system of the foundation with industry universality,Make up the blank of industry field.
Description of the drawings
Fig. 1 is work flow diagram of the present invention.
Specific implementation mode
The invention will be further described with reference to the accompanying drawings and examples.
Embodiment, as shown in Fig. 1, the use for tag control system of being drawn a portrait based on deep learning online education Students ' ComprehensiveMethod, wherein：
Step 1, which buries acquisition and will bury a technology using webpage, to be carried out webpage to be implanted into the processing such as JS SDK codes in advance, fromAnd so that student is recorded for the navigation patterns of each webpage and click behavior, the content of record can be completely anti-The complete view path for answering the online upper mounting plate of student locks its web page contents checked；
The acquisition of step 2 E.E.G will utilize wear-type E.E.G acquisition instrument, by the detection to cerebral cortex voltage change, andThe processing such as difference, digital-to-analogue conversion are carried out, realize the retention note of the situation of change of the brain wave to student in entire learning processRecord, and then realize the exploration to the instant active state of brain and record；
The acquisition of step 3 viewpoint track will utilize the helmets such as eye tracker, by eyeball focus in eyeball navigation processPosition and state, acquisition student in entire learning process for browsing content sight focal position variation etc. information,And then realize exploration and the record that position is paid close attention to student immediately；
Step 4 data prediction will be based on pupilage card number or student status class unifying identifier, will be acquired in above stepThe lateral fusion of multi-source heterogeneous data progress arrived, i.e. the fusion of progress field level, and the fusion of non-recorded grade.After fusion,Further to new data set carry out completeness and efficiency audit, identify invalid value therein, as and table after Null values, wordSymbol type blank etc., and including not meeting convention, such as average daily study duration is more than 24 hours, and does not meet business in field and recognizeKnow, if single document browsing duration is more than 5 hours exceptional values, is rejected or adjusted for this partial data, is such as directed to null valuePadding, and the thresholding operation for outlier；
The processing of step 5 deep learning will utilize deep learning technology, on the pretreated data basis of step 4, fromRefinement includes the content-preference of student, behavior pattern, learning style, in the information such as content of the act and action trail on student's lineAttitude is practised, the Online Learning mode tag including learning performance such as learns liveness, study engagement, self, interactionParticipation, Fast Learning ability etc.；
In aforementioned, step 6 image recognition processing will utilize image recognition technology, to step 3 middle school student in learning processSight focal position matched with corresponding specific browsing content, identification obtain student each moment browsing content andComplete browsing content chain, and based on the division and judgement to page browsing content, lock the emphasis in students'learningAttentinal contents and its distinctive navigation patterns pattern, refinement include student's content-form preference, student's visual color preference,Learning state label including viewpoint stability；
In aforementioned, step 7 text mining processing will carry out one using Text Mining Technology for the on-line off-line comment of studentThe processing of series, including be based on hidden Markov chain and carry out Chinese text participle, carry out mutual information calculating, base for word segmentation resultIt is distributed the subject classification judgement for carrying out content in Di Li Crays, and then refines and obtains the entity pair that Students ' Evaluation discusses content, dismantling obtainsTarget object, the Sentiment orientation and subject matter for object for obtaining Students ' Evaluation opinion, form the subjective state scale label of student；
In aforementioned, step 8 draws a portrait label system construction will be using dimensionality reductions digging technologies pair such as hierarchical clustering, principal component analysisAll kinds of labels are classified, and carry out the induction ＆ summing-up on upper layer, shape on the basis of modeling result based on business cognition in fieldAt a top-down complete tag system；
In aforementioned, step 9 system output and show using data visualization technique, in conjunction with line chart, pie chart, bar shapedThe forms such as various figures and word, table such as figure, geographical thermal map, instrument board, in a manner of system, by label of drawing a portrait atFruit is exported, and on the one hand carrying out macroscopic aspect to Students ' Comprehensive portrait by group's grade analysis report, visualization large-size screen monitors retouchesIt paints and portrays, on the other hand reported by detailed personal grade, student's microscopic feature is understood and showed.
Through the invention, the acquisition of its on-line off-line comprehensive data may be implemented in the student of online education, and formation hasRepresentative complete portrait label, from two levels of both macro and micro realize to student characteristics it is comprehensive it is careful portray, make up rowThe directional guide of the blank in industry field, effectively support for student and the orientation optimization for teaching.
Example the above is only the implementation of the present invention is not intended to limit the scope of the invention, every to utilize this hairEquivalent structure or equivalent flow shift made by bright specification and embodiment content, it is relevant to be applied directly or indirectly in otherTechnical field is included within the scope of the present invention.