Attribute exploration-based test item implicit knowledge attribute association mining and related test item pushing method and system

Abstract

The present application belongs to the technical field of education data mining, and discloses a test question implicit knowledge attribute association mining method and system based on attribute exploration and a related test question pushing method and system, which comprises the following steps: constructing a formal background of a student's error question set; preprocessing a student's question answer record source data, and then filtering out the student's error answer information in combination with a knowledge attribute set contained in the labeled question; exploring the current formal background by using an attribute exploration knowledge attribute mining algorithm to obtain a knowledge attribute implication association set and a non-redundant test question set through any student error answer information; calculating the similarity between test questions by using concept lattice similarity analysis, finding questions containing similar knowledge points to the practice wrong questions, and selecting a number of related test questions meeting a threshold to push. The present application solves the drawbacks caused by the missing attributes in the students' fuzzy judgment in the traditional analysis mode, and can provide targeted test question recommendations for students.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of educational data mining, and particularly relates to a test question implicit knowledge attribute association mining and related test question pushing method and system based on attribute exploration. BACKGROUND

[0002] Attribute exploration is a method for obtaining the implication relationship between attributes based on formal concept analysis theory, and the implication between attributes can represent the knowledge of the inclusion relationship between individual sets. The attribute exploration algorithm mines all the connotations and main bases under the current background by asking a series of questions to domain experts, and further analysis can be performed according to the exploration results. The domain expert can be a database or a person.

[0003] Clustering is a process of gathering a large amount of data into different groups or clusters, so that the objects in the same cluster are extremely similar, and the objects between different clusters are quite different. Clustering analysis can also classify individual objects in the data according to one or more attributes, and perform data analysis according to the characteristics of different classes.

[0004] Formal concept analysis, also known as concept lattice, is a clustering method that can represent the object set satisfying a certain attribute set and the object set having these attributes in the form of "concept". The objects satisfying a certain attribute are classified by the form of "concept", and the hierarchical relationship between concepts and the generalization and instantiation relationship between concepts can be intuitively represented by Hasse graph. Through concept lattice, not only can educational data be clustered and analyzed, but also Hasse graph can intuitively represent the relationship between test questions and knowledge point attributes, and between concepts with different attributes. Therefore, concept lattice can more intuitively display the results than some other data mining methods through two-dimensional tables.

[0005] In the traditional education mode, students often accumulate a large number of practice questions and examination questions in the learning process, and these questions often cannot be fully mined by students to find their weaknesses, and the sensitivity to the association of incorrect knowledge attributes in the test questions is relatively weak, which may lead to a large number of problems being accumulated in the next step of work, and the students may be in a situation of not knowing where to start. SUMMARY

[0006] The present application is directed to the problem that in the traditional education mode, students often accumulate a large number of exercises and test questions in the learning process, and these questions often cannot be fully explored by students to find their weaknesses, and the sensitivity to the association of the wrong knowledge attributes in the test questions is relatively weak, which will lead to a large number of problems accumulated in the next step of work, and thus the students will be in a situation of not knowing where to start. A test question implicit knowledge attribute association mining and related test question pushing method and system based on attribute exploration are proposed. Students will leave traces in the process of answering questions during the test or test process. According to the response record of the students to the test questions, the wrong questions in the student practice process are screened out, and the associated knowledge attributes of the students in each wrong question are obtained by combining the association table of the test questions and the knowledge attributes annotated by the domain experts (the table represents the knowledge attributes examined by each test question), and the form background of the student is formed. In the present application, only objective questions are considered, that is, the answer records of the students only have correct and wrong, and subjective questions under the condition of multiple values are not considered.

[0007] In order to achieve the above-mentioned purpose, the present application adopts the following technical scheme:

[0008] The present application proposes a test question implicit knowledge attribute association mining and related test question pushing method based on attribute exploration, which comprises:

[0009] Step A, constructing a form background K of a student to a wrong question set: preprocessing the student question answering record source data, and then filtering out the student's wrong answer information by combining the annotated knowledge attribute set contained in the question, wherein the wrong answer information is composed of the student's wrong answer question and the knowledge attribute set contained in each question;

[0010] Step B, using an attribute exploration knowledge attribute mining algorithm to explore the current form background, obtaining the knowledge attribute implication association set and the non-redundant test question set obtained by any student's wrong answer information in step A;

[0011] Step C, calculating the similarity between the test questions by using the concept lattice similarity analysis, finding the questions containing similar knowledge points with the wrong questions, and selecting a plurality of related test questions meeting the threshold to push.

[0012] Further, the step A comprises:

[0013] Step A1, preprocessing the existing student question answering record source data; including: cleaning the source data, deleting irrelevant attributes in the source data, obtaining binary answer information data containing only a student object and his / her answer question true or false; and deleting the correctly answered questions in the student answer record, and retaining the student's wrong answer record;

[0014] Step A2, associating the preprocessed data with the knowledge attribute correlation matrix; including: representing the labeled knowledge attribute set by a matrix, representing each row as a question and each column as a knowledge attribute, using 1 to represent the knowledge attribute if the question contains the knowledge attribute, otherwise using 0, to obtain the knowledge attribute correlation matrix; combining the preprocessed data with the knowledge attribute correlation matrix to obtain the error response information matrix of each error question and knowledge attribute fusion, that is, obtaining the question knowledge attribute form background K of the current student for the error question set.

[0015] Further, the step B includes:

[0016] Step B1, constructing a dictionary sequence set Z of the knowledge attribute set, taking the last knowledge attribute set Z in the dictionary sequence set i , and calculating e in the current form background, that is, the non-redundant question set K , where represents the question set with all attributes in the knowledge attribute set Z i , and represents the knowledge attribute set commonly possessed by all question elements in the set .

[0017] Step B2, calculating in the preprocessed data, that is, the form background K, whether it is true; if not, executing step B3, otherwise executing step B4; where D K (Z i ) represents the question set with all attributes in the knowledge attribute set Z i calculated in the form background K, represents the difference set of the two sets;

[0018] Step B3, finding a question in the form background K that does not satisfy the condition and adding it to the non-redundant question set K e , and updating K e as the new non-redundant question set; returning to step B2;

[0019] Step B4, judging whether the objects commonly contained by the non-redundant question set containing the knowledge attribute set Z i in the form background K are equal to the knowledge attribute set Z i , if not, adding the knowledge attribute implication relationship to the knowledge attribute implication correlation set Y; where represents that the student will also answer the question containing the attribute i incorrectly when answering the question containing the knowledge attribute set Z incorrectly;

[0020] Step B5, calculate the next knowledge attribute set Z in the lexicographic set Z i+1 ; determine whether the knowledge attribute set Z in the current lexicographic set is correct i j Y * ∪(Z i ∩{e1,e2,...e j-1}∪(e j )) is correct, if the condition is correct, the next lexicographic set is Z i+1 =Y * ∪(Z i ∩{e1,e2,...e j-1}∪(e j )), enter step B1; otherwise, j = j-1, and continue to calculate in the current step until j = 0, end the program; wherein Y * represents the set of last knowledge attribute implication formula antecedent elements in Y, the set E = {e1, e2,... e j} represents the knowledge attribute set, and j is the number of marked knowledge attributes;

[0021] Step B6, clean the knowledge attribute implication association set, if the knowledge attribute implication relationship formula in is , it is considered that the knowledge attribute implication relationship formula is a redundant condition, and is deleted;

[0022] Step B7, finally obtain the non-redundant test question set K e and the knowledge attribute implication association set Y through the attribute exploration knowledge attribute mining algorithm of steps B1 to B6.

[0023] Further, the step C includes:

[0024] Step C1, construct the test question knowledge attribute formal context L of the test question library that the current student has not practiced according to the step A mode;

[0025] Step C2, construct the concept lattice of the test question library that the current student has not practiced by using the bordat concept lattice construction algorithm, and construct the concept lattice of the test question library by inputting the formal context L obtained in step C1; the concept lattice is represented by concept nodes and partial order relations between the nodes, and each node in the concept lattice is represented by a binary relation (U, S), wherein U represents the current student's test question set that has not been practiced, S represents the knowledge attribute set, and the concept node represents that the test question set U has the knowledge attribute of the S set;

[0026] Step C3, based on the knowledge attribute implication association set obtained in step B, calculate the similarity between the concept node and the knowledge attribute implication relationship formula by using the cosine similarity calculation formula by traversing all concept nodes of the concept lattice; ​

[0027] C4, ranking the related test questions according to the cosine similarity calculation result, setting a threshold, screening out results greater than the threshold, and selecting top-n test questions for pushing, if the number of screened test questions is small, then reducing the threshold.

[0028] Another aspect of the present application provides a test question implicit knowledge attribute association mining and related test question pushing system based on attribute exploration, comprising:

[0029] The wrong question form background construction module is used for constructing the form background K of the student on the wrong question set, pre-processing the student question answer record source data, and then filtering out the student's wrong answer information by combining the knowledge attribute set contained in the labeled question, wherein the wrong answer information is composed of the student's wrong answer question and the knowledge attribute set contained in each question.

[0030] The non-redundant test question set and knowledge attribute implication association set derivation module is used for exploring the current form background by using the attribute exploration knowledge attribute mining algorithm, and obtaining the knowledge attribute implication association set and the non-redundant test question set derived by any student wrong answer information in the wrong question form background construction module.

[0031] The related test question pushing module is used for calculating the similarity between test questions by using concept lattice similarity analysis, finding questions containing similar knowledge points as the wrong questions, and selecting a plurality of related test questions that meet the threshold to push.

[0032] Further, the wrong question form background construction module is specifically used for:

[0033] Step A1, pre-processing the existing student question answer record source data; including: cleaning the source data, deleting irrelevant attributes in the source data, and obtaining binary answer information data containing only a student object and his / her answer question true or false; at the same time, deleting the correctly answered questions in the student answer record, and retaining the student's wrong answer record;

[0034] Step A2, fusing the knowledge attribute association matrix with the pre-processed data; including: representing the labeled knowledge attribute set by a matrix, representing each row as a question and each column as a knowledge attribute, and using 1 to represent the question containing the knowledge attribute and 0 otherwise, to obtain the knowledge attribute association matrix; combining the pre-processed data with the knowledge attribute association matrix to obtain the error answer information matrix of each wrong answer question and knowledge attribute fusion, and obtaining the test question knowledge attribute form background K of the current student on the wrong question set.

[0035] Further, the non-redundant test question set and knowledge attribute implication association set derivation module is specifically used for:

[0036] Step B1, construct the lexicographic set Z of knowledge attribute sets, take the last knowledge attribute set Z in the lexicographic set i , and calculate in the current formal context, i.e. the non-redundant question set K e represents the question set with all attributes in the knowledge attribute set Z i represents the knowledge attribute set commonly owned by all question elements in the set

[0037] Step B2, calculate in the preprocessed data, i.e. the formal context K whether D K (Z i ) is true; if not, execute step B3, otherwise execute step B4; wherein D K (Z i ) represents the question set with all attributes in the knowledge attribute set Z i calculated in the formal context K represents the difference set of the two sets

[0038] Step B3, find a question in the formal context K that does not satisfy the condition, add it to the non-redundant question set K e , and update K e as the new non-redundant question set; return to step B2

[0039] Step B4, determine whether the objects commonly contained in the non-redundant question set containing the knowledge attribute set Z i in the formal context K are equal to the knowledge attribute set Z i , if not, add the knowledge attribute implication relationship formula to the knowledge attribute implication association set Y; wherein represents that the student will also answer incorrectly the question containing the attribute when answering incorrectly the question containing the knowledge attribute set Z i

[0040] Step B5, calculate the next knowledge attribute set Z i+1 in the lexicographic set Z; determine whether the knowledge attribute set Z i j Y * ∪(Z i ∩{e1,e2,...e j-1}∪(e j )) is true, if the condition is true, the next lexicographic set is Z i+1 = Y * ∪(Z i ∩{e1,e2,...e​​j-1}∪(e j If j = j-1, proceed to step B1; otherwise, continue calculating in the current step until j = 0, then terminate the program; where Y * Let E = {e1, e2, ..., e} represent the set of consequents of the relation implied by the last knowledge attribute in Y. j} represents the set of knowledge attributes, and j is the number of knowledge attributes labeled;

[0041] Step B6 involves cleaning the set of associations implied by knowledge attributes. If the knowledge attribute implication relation contains... The set is The implication relation of this knowledge attribute is considered a redundant condition and is deleted.

[0042] Step B7: Through the attribute exploration and knowledge attribute mining algorithm used in steps B1 to B6, the non-redundant question set K is finally obtained. e And the knowledge attributes imply the related set Y.

[0043] Furthermore, the relevant test question push module is specifically used for:

[0044] Step C1: Construct the background L of the question knowledge attributes of the question bank that the current students have not practiced, following the method in Step A;

[0045] Step C2: The Bordat concept lattice construction algorithm is used to construct a concept lattice for the question bank that the current student has not practiced. The concept lattice for the question bank is constructed by inputting the formal background L obtained in step C1. The concept lattice is represented by concept nodes and partial order relations between nodes. Each node in the concept lattice is represented by a binary relation (U, S), where U represents the set of questions that the current student has not practiced, S represents the set of knowledge attributes, and the concept node indicates that the question set U has the knowledge attributes of the set S.

[0046] Step C3: Based on the knowledge attribute implication association set obtained in step B, calculate the similarity between the concept node and the knowledge attribute implication relation by traversing all concept nodes of the concept lattice and using the cosine similarity calculation formula.

[0047] C4: Sort the relevant test questions based on the cosine similarity calculation results, set a threshold, filter out the results greater than the threshold, and select the top-n test questions to push. If the number of selected test questions is small, lower the threshold.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] The method and system for mining implicit knowledge attributes and recommending related questions based on attribute exploration, as described in this invention, can explore a non-redundant set of questions through step B. In particular, it can mine the implicit relationships between knowledge attributes in the set of questions with incorrect knowledge points through the attribute exploration algorithm, and represent the related knowledge attributes in the set of incorrect questions. Then, step C2 constructs a concept lattice of unpracticed questions. This step can clearly represent the question binary relation groups with different attribute combinations through the concept set in the concept lattice. Step C3 uses the cosine similarity calculation method to calculate the similarity between the attributes in the concept binary set and the attributes in the implied meaning, so as to obtain the set of questions associated with the incorrect knowledge attributes. This realizes the recommendation based on implicit association attributes and related questions, which solves the drawbacks of the lack of attributes caused by students' fuzzy judgment in the traditional analysis mode, and can provide students with targeted question recommendations. Attached Figure Description

[0050] Figure 1 This is a basic flowchart of a method for mining the association of implicit knowledge attributes in test questions and pushing related test questions based on attribute exploration, according to an embodiment of the present invention.

[0051] Figure 2 This is a schematic diagram of the architecture of a test question implicit knowledge attribute association mining and related test question push system based on attribute exploration, according to an embodiment of the present invention. Detailed Implementation

[0052] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments:

[0053] like Figure 1 As shown, a method for mining the association of implicit knowledge attributes in test questions and pushing related test questions based on attribute exploration includes:

[0054] Step A, constructing the formal background K of the student's incorrect question set: preprocess the source data of the student's question answer records, and then filter out the student's incorrect response information by combining the knowledge attribute set contained in the labeled questions. The incorrect response information consists of the student's incorrect answer questions and the knowledge attribute set contained in each question.

[0055] Step B involves using an attribute exploration knowledge attribute mining algorithm to explore the current context and obtain the knowledge attribute implication set and non-redundant question set derived from any student's incorrect response information in Step A.

[0056] Step C involves using concept lattice similarity analysis to calculate the similarity between test questions, finding questions containing knowledge points similar to those in practice questions, and selecting several related test questions that most closely match the threshold for submission.

[0057] Further, step A includes:

[0058] Step A1, pre-process the existing student answer record source data. The student answer record source data may contain multiple object information and multiple attributes of the object. Some of the attribute features, such as the gender and student ID of the student, are irrelevant attributes in the present application. Therefore, the data is cleaned to delete the irrelevant attributes in the source data, and only the binary answer information data of a certain student object and the correct or incorrect answer of the question are obtained. At the same time, the correctly answered questions in the student answer record are deleted, and the error answer record of the student is retained.

[0059] Step A2, fuse the knowledge attribute association matrix of the pre-processed data. The knowledge attribute set labeled by the domain expert is represented by a matrix, each row represents a question, and each column represents a specific knowledge attribute. If the question contains a certain knowledge attribute, it is represented by 1, otherwise by 0. For example, a question is 5*5+6, and the corresponding knowledge attributes of multiplication and addition are set to 1. The pre-processed data is combined with the data information of the knowledge attributes contained in the question given by the expert (knowledge attribute association matrix) to obtain the error answer information matrix of each error answer question and knowledge attribute fusion, that is, the question knowledge attribute form background K of the current student for the error question set.

[0060] Further, the step B includes:

[0061] Step B1, the question knowledge attribute set P = (P1, P2, P3,... Pn) labeled by the domain expert in step A is obtained. n , and the initial state knowledge attribute implication association set Non-redundant question set Construct a lexicographic set of knowledge attribute sets Take the last knowledge attribute set Z in the lexicographic set i (initially, ), and calculate e in the current form background, i.e., the non-redundant question set K After the calculation, go to step B2; wherein represents the question set with all attributes in the knowledge attribute set Z i represents the knowledge attribute set commonly possessed by all question element sets.

[0062] Step B2, calculate whether it is true in the pre-processed data, i.e., the form background K, whether the question set with the attribute set Z i is contained in ​The attribute set is in the question set within the formal context K; if the above judgment is true, proceed to step B4; if the above judgment is false, proceed to step B3; where D K (Z i ) represents calculating the possession of Z in formal context K. i The set of test questions for all attributes in the knowledge attribute set. It represents the difference between two sets.

[0063] Step B3, search for non-satisfied conditions in the formal background K. A line of information is added to the non-redundant question set K. e In the process, a data point that does not meet the conditions is inserted into the non-redundant question set, and K is updated. e As a new set of non-redundant questions; return to step B2.

[0064] Step B4: Determine if Z is contained in the formal background K. i Does the set of non-redundant test questions containing the knowledge attribute set contain the same object as Z? i If the sets of knowledge attributes are not equal, then the implication relation of the knowledge attributes will be expressed. Adding it to the knowledge attribute-implied association set Y indicates that the student's incorrect answer contains the attribute set Z. i The test questions will also contain Questions about attributes were answered incorrectly together; among them This indicates that the student's incorrect answer contains a set of knowledge attributes Z. i The test questions will also contain The attribute-related questions were answered incorrectly. Proceed to step B5;

[0065] Step B5, this step is used to calculate the next set of knowledge attributes Z in the lexicographical set Z. i+1 .

[0066] Determine the set of knowledge attributes Z in the current lexicographically ordered set. i < j Y * ∪(Z i ∩{e1, e2, ... e j-1}∪(e j If the condition is true, then the next element in the lexicographical set is Z. i+1 =Y * ∪(Z i ∩{e1, e2, ... e j-1}∪(e j If the condition is met, proceed to step B1; otherwise, j = j-1, i.e., determine e. j The previous element is used, and the calculation continues in the current step until j = 0, that is, the entire set of knowledge attributes has been traversed, and the program terminates; where Y* E represents the set of the last knowledge attribute implication relation formula in Y, set E = {e1, e2,... e j} represents the set of knowledge attributes, j is the number of knowledge attributes marked by the domain expert, Z1< Z2 j Z2 is true if and only if e j ∈ Z2-Z1, and Z1∩{e1, e2,... e j-1} = Z2∩{e1, e2,... e j-1}, specifically, for Z i < Z2 j Y * ∪(Z i ∩{e1, e2,... e j-1}∪(e j )) is true, Z1 represents Z i , and Z2 represents Y * ∪(Z i ∩{e1, e2,... e j-1}∪(e j )).

[0067] Step B6, the knowledge attribute implication association set is cleaned up, if the knowledge attribute implication relation formula is The set is It is considered that the knowledge attribute implication relation formula is a redundant condition, and is deleted, and after the step is completed, step B7 is entered.

[0068] Step B7, the non-redundant test question set K e and the knowledge attribute implication association set Y are finally obtained through the attribute exploration knowledge attribute mining algorithm of steps B1 to B6.

[0069] Further, the step C includes:

[0070] Step C1, the test question knowledge attribute formal context L of the test question library that the current student has not practiced is constructed according to the step A mode. Specifically, in the test question library that the student has not practiced, there is also an association matrix between the test questions and the knowledge attributes marked by the domain expert. The formal context construction step is the same as A, which will not be repeated here.

[0071] Step C2, the test question library concept lattice is constructed, this step uses the bordat concept lattice construction algorithm to construct the concept lattice of the test question library by inputting the formal context L obtained in step C1. The concept lattice is represented by the concept nodes and the partial order relation between the nodes, and each node in the concept lattice is represented by a binary relation (U, S), U represents the test question set, S represents the knowledge attribute set, and the concept node represents that the test question set U has the knowledge attribute set S.

[0072] Step C3, the implication set Y obtained by step B, the implication A→B in the present application means that the student is missing the set of knowledge attributes of B while having the set of knowledge attributes of A. By traversing all the concept nodes of the concept lattice, the similarity of the concept node and the implication A→B is calculated by the cosine similarity calculation formula.

[0073]

[0074] The attribute set P=AUB, int(c) represents the intension set of the concept node c, that is, the attribute set S of the concept node (U, S).

[0075] C4, according to the cosine similarity calculation result, the questions of the related test questions are sorted, a threshold is set, the results greater than the threshold are screened out, and top-n test questions are selected for pushing, if the number of screened test questions is small, the threshold is reduced. As an implementable way, the threshold θ is set to 0.8, the results greater than the threshold are screened out, and top 5 questions are selected for pushing, if the number of screened questions is less than 3, the threshold is reduced by 0.05 in turn, and finally not less than 0.6.

[0076] On the basis of the above embodiment, the present application also proposes a test question implicit knowledge attribute association mining and related test question pushing system based on attribute exploration, comprising:

[0077] The wrong question form background construction module is used for constructing the form background K of the student to the wrong question set: the student question answer record source data is preprocessed, and then combined with the knowledge attribute set contained in the labeled question, the student's wrong answer information is filtered out, the wrong answer information is composed of the student's wrong answer question and the knowledge attribute set contained in each question, and the wrong answer information is encapsulated and processed;

[0078] The non-redundant test question set and knowledge attribute implication association set derivation module is used for exploring the current form background by using the attribute exploration knowledge attribute mining algorithm, obtaining the knowledge attribute implication association set and the non-redundant test question set derived by any student wrong answer information in the wrong question form background construction module;

[0079] The related test question pushing module is used for calculating the similarity between the test questions by using the concept lattice similarity analysis, finding the questions containing similar knowledge points with the wrong questions, and selecting a plurality of related test questions meeting the threshold to push the questions.

[0080] Further, the wrong question form background construction module is specifically used for:

[0081] Step A1: Preprocess the existing student question answer record source data; including: cleaning the source data, deleting irrelevant attributes in the source data, and obtaining binary answer information data containing only a certain student object and the correctness of its answer to the question; at the same time, delete the questions answered correctly in the student answer record, and retain the student's incorrect answer record;

[0082] Step A2 involves fusing the preprocessed data with a knowledge attribute association matrix, including: representing the labeled knowledge attribute set using a matrix, where each row represents a question and each column represents a knowledge attribute; if a question contains the knowledge attribute, it is represented by 1, otherwise by 0, thus obtaining the knowledge attribute association matrix; combining the preprocessed data with the knowledge attribute association matrix to obtain the error response information matrix that fuses each incorrectly answered question with the knowledge attribute, which is the background K of the current student's knowledge attribute form for the set of incorrect questions.

[0083] Step A3, data encapsulation; includes: importing the error response information matrix, i.e., the formal background K, into the concept lattice generation system Conexp, saving the data to be mined in the form of a data frame, and completing the data encapsulation.

[0084] Furthermore, the module for deriving the non-redundant question set and the knowledge attribute-implication association set is specifically used for:

[0085] Step B1: Construct the lexicographical set Z of the knowledge attribute set, and extract the last knowledge attribute set Z from the lexicographical set. i And in the current context, namely the non-redundant question set K e Calculated from in Indicates possession of Z i The set of test questions for all attributes in the knowledge attribute set. express The set of knowledge attributes that are common to all test question elements;

[0086] Step B2, in the preprocessed data, i.e., the formal background K, calculate If the condition is not met, proceed to step B3; otherwise, proceed to step B4; where D... K (Z i ) represents calculating the possession of Z in formal context K. i The set of test questions for all attributes in the knowledge attribute set. Denotes the difference between two sets;

[0087] Step B3: Find a test item in the formal background K that does not meet the conditions and add it to the non-redundant test item set K. e In, and update K e As a new set of non-redundant questions; return to step B2;

[0088] Step B4, determine whether the object contained in the non-redundant question set of the knowledge attribute set Z i in the formal context K is equal to Z i If not, add the knowledge attribute implication relation formula to the knowledge attribute implication relation set Y; wherein indicates that the student will also answer the question containing the knowledge attribute set Z i incorrectly;

[0089] Step B5, calculate the next knowledge attribute set Z i+1 in the lexicographic set Z; determine whether the knowledge attribute set Z i j Y * ∪(Z i ∩{e1,e2,...e j-1}∪(e j )) is true, if the condition is true, the next lexicographic set is Z i+1 =Y * ∪(Z i ∩{e1,e2,...e j-1}∪(e j )), enter step B1; otherwise, j=j-1, and continue to calculate in the current step until j=0, terminate the program; wherein Y * represents the set of consequent elements of the last knowledge attribute implication relation formula in Y, the set E={e1,e2,...e j} represents the knowledge attribute set, and j is the number of marked knowledge attributes;

[0090] Step B6, clean the knowledge attribute implication relation set, if the antecedent set of the knowledge attribute implication relation formula is then the knowledge attribute implication relation formula is considered to be a redundant condition and is deleted;

[0091] Step B7, finally obtain the non-redundant question set K e and the knowledge attribute implication relation set Y.

[0092] Further, the related question pushing module is specifically configured to:

[0093] Step C1, construct the question knowledge attribute formal context L of the question bank that the current student has not practiced according to the method of step A;

[0094] ​​​Step C2, the Bordat concept lattice construction algorithm is used to construct the concept lattice of the test questions which have not been practiced by the current student, and the formal context L obtained in step C1 is input to construct the concept lattice of the test question library; the concept lattice is represented by concept nodes and the partial order relationship between the nodes, and each node in the concept lattice is represented by a binary relationship (U, S), wherein U represents the test question set which has not been practiced by the current student, and S represents the knowledge attribute set; the concept node represents that the test question set U has the knowledge attribute set S;

[0095] Step C3, based on the knowledge attribute implication association set obtained in step B, the similarity of the concept node and the knowledge attribute implication relationship formula is calculated through the cosine similarity calculation formula by traversing all the concept nodes of the concept lattice;

[0096] C4, according to the cosine similarity calculation result, the questions of the related test questions are sorted, a threshold is set, the results greater than the threshold are screened out, and top-n test question titles are selected for pushing, and if the number of the screened test question titles is small, the threshold is reduced.

[0097] In summary, the test question implicit knowledge attribute association mining and related test question pushing method and system based on attribute exploration can explore the non-redundant set of the test question set through step B, especially can mine the implication relationship between the knowledge attributes in the error knowledge point exercise set of the student through the attribute exploration algorithm, and express the associated knowledge attributes in the error question set. Then, the unpracticed question concept lattice is constructed through step C2, and the binary relationship group of the questions owned by different attribute combinations can be clearly expressed in the concept set in the concept lattice through this step, the attribute similarity in the concept binary set and the implication is calculated through the cosine similarity calculation method in step C3, so as to obtain the test question set associated with the error knowledge attribute, realize the recommendation based on the implicit association attribute and the associated question, solve the disadvantages caused by the missing attribute in the fuzzy judgment of the student in the traditional analysis mode, and can provide targeted test question recommendation for the student.

[0098] The above only shows the preferred embodiments of the present application, and it should be noted that for those skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present application.

Claims

1. An attribute exploration-based test question implicit knowledge attribute association mining and related test question pushing method, characterized in that, The method comprises the following steps: Step A, constructing a formal background K of a student on a set of wrong questions: preprocessing the source data of the student's question answering records, and then filtering out the student's wrong answer information by combining the knowledge attribute set contained in the labeled questions, wherein the wrong answer information is composed of the student's wrong answer questions and the knowledge attribute set contained in each question; Step B, exploring the current formal background using an attribute exploration knowledge attribute mining algorithm to obtain the knowledge attribute implication association set and the non-redundant question set derived from any student's wrong answer information in step A; Step C, calculating the similarity between questions using concept lattice similarity analysis to find questions containing similar knowledge points to the wrong questions, and selecting a number of related questions that meet the threshold value of similarity to push; The step B comprises: Step B1: Construct the lexicographical set Z of the knowledge attribute set, and extract the last knowledge attribute set from the lexicographical set. And in the current context, i.e., a non-redundant set of test questions. Calculated from ;in Indicates ownership The set of test questions for all attributes in the knowledge attribute set. express The set of knowledge attributes that are common to all test question elements; Step B2, in the pre-processed data, i.e. in the formal context K, compute if not, then perform step B3, else perform step B4; wherein denotes the set of all items in the formal context K, denotes the set of all items in the formal context K, denotes the difference of two sets; Step B3, find an item in form context K that does not satisfy the condition and add to the non-redundant item set in the database and update as the new non-redundant item set; go to step B2; Step B4, judge whether the objects contained in the non-redundant question set of the knowledge attribute set in the formal context K are equal to the objects contained in the knowledge attribute set If not, add the knowledge attribute implication relation formula to the knowledge attribute implication relation set Y; wherein indicates that the student will also answer the question containing the knowledge attribute incorrectly when answering the question containing the knowledge attribute set incorrectly; Step B5, calculate the next knowledge attribute set in the lexicographic set Z ; determine whether the knowledge attribute set in the current lexicographic set is correct, if the condition is correct, the next lexicographic set is , enter step B1; otherwise j = j-1, and continue to calculate in the current step until j = 0, terminate the program; wherein denotes the set of last knowledge attribute implication relation formula in Y, set E = denotes the knowledge attribute set, and j is the number of marked knowledge attributes; Step B6, clean the set of knowledge attribute implication association, if the knowledge attribute implication relation formula is Ø, it is considered that the knowledge attribute implication relation formula is redundant condition, and is deleted; Step B7, the non-redundant test question set is finally obtained by the knowledge attribute mining algorithm of the attribute exploration of steps B1 to B6 and the knowledge attribute implication association set Y. 2.The attribute exploration-based test item implicit knowledge attribute association mining and related test item pushing method according to claim 1, characterized in that, The step A comprises: Step A1, preprocessing the existing student question answering record source data; including: cleaning the source data, deleting irrelevant attributes in the source data, and obtaining binary answer information data containing only a student object and his / her answer question true or false; at the same time, deleting the correctly answered questions in the student's answer record and retaining the student's wrong answer record; Step A2, fusing the knowledge attribute association matrix of the preprocessed data; including: representing the labeled knowledge attribute set by a matrix, representing each row as a question and each column as a knowledge attribute, and using 1 to represent the question containing the knowledge attribute and 0 otherwise, to obtain the knowledge attribute association matrix; combining the preprocessed data with the knowledge attribute association matrix to obtain the error answer information matrix of each wrong answer question and knowledge attribute fusion, i.e., obtaining the question knowledge attribute formal background K of the current student on the set of wrong questions. 3.The attribute exploration-based test item implicit knowledge attribute association mining and related test item pushing method according to claim 1, characterized in that, The step C comprises: Step C1, constructing the question knowledge attribute formal background L of the current student's unpracticed question library according to the method of step A; Step C2, constructing the concept lattice of the current student's unpracticed question library using the bordat concept lattice construction algorithm, and constructing the concept lattice of the question library by inputting the formal background L obtained in step C1; the concept lattice is represented by concept nodes and partial order relations between nodes, and each node in the concept lattice is represented by a binary relation (U, S), wherein U represents the current student's unpracticed question set, and S represents the knowledge attribute set; the concept node represents that the question set U has the knowledge attribute of the S set; Step C3, based on the knowledge attribute implication association set obtained in step B, calculating the similarity of the concept node and the knowledge attribute implication relation formula by traversing all concept nodes of the concept lattice and using the cosine similarity calculation formula; C4, sorting the related question questions according to the cosine similarity calculation result, setting a threshold value, screening out the results greater than the threshold value, and selecting top-n question questions for pushing; if the number of screened question questions is small, the threshold value is reduced.

4. An attribute exploration-based test question implicit knowledge attribute association mining and related test question pushing system, characterized in that, The method comprises the following steps: The wrong question form background construction module is configured to construct a form background K of a student on a wrong question set, pre-process a student question answer record source data, and then filter out a student error answer information by combining a knowledge attribute set contained in a labeled question, the error answer information being composed of a student error answer question and a knowledge attribute set contained in each question; The non-redundant question set and knowledge attribute implication association set derivation module is configured to derive a knowledge attribute implication association set and a non-redundant question set from any student error answer information derived by the wrong question form background construction module by using an attribute exploration knowledge attribute mining algorithm to explore the current form background; The related question pushing module is configured to calculate the similarity between questions by using a concept lattice similarity analysis, find questions containing similar knowledge points to the practice wrong questions, and select a number of related questions that meet a threshold to push. The non-redundant question set and knowledge attribute implication association set derivation module is configured to: Step B1: Construct the lexicographical set Z of the knowledge attribute set, and extract the last knowledge attribute set from the lexicographical set. And in the current context, i.e., a non-redundant set of test questions. Calculated from ;in Indicates ownership The set of test questions for all attributes in the knowledge attribute set. express The set of knowledge attributes that are common to all test question elements; Step B2, in the pre-processed data, i.e. in the formal context K, compute if not, then perform step B3, else perform step B4; wherein denotes the computation of the set of all items in the formal context K that have a set of test questions for all attributes in the set of knowledge attributes, denotes the difference set of two sets; Step B3, find an item in form context K that does not satisfy the condition and add to the non-redundant item set in the database and update as the new non-redundant item set; Return to step B2. Step B4, judge whether the objects contained in common by the non-redundant question set of the knowledge attribute set in the formal context K are equal to the objects contained in common by the non-redundant question set of the knowledge attribute set in the formal context K, if not equal, then add the knowledge attribute implication relation formula of the knowledge attribute set to the knowledge attribute implication relation set Y; wherein indicates that the student will also answer the question containing the knowledge attribute incorrectly when answering the question containing the knowledge attribute set incorrectly; incorrectly. Step B5, calculate the next knowledge attribute set in the lexicographic set Z ; determine whether the knowledge attribute set in the current lexicographic set is correct, if the condition is correct, the next lexicographic set is , enter step B1; otherwise j = j-1, and continue to calculate in the current step until j = 0, end the procedure; wherein represents the set of last knowledge attribute implication relation formula in Y, set E = represents the knowledge attribute set, and j is the number of marked knowledge attributes; Step B6, clean the set of knowledge attribute implication association, if the knowledge attribute implication relation formula is Ø, it is considered that the knowledge attribute implication relation formula is redundant condition, and is deleted; Step B7, the non-redundant test question set is finally obtained by the knowledge attribute mining algorithm of the attribute exploration of steps B1 to B6 and the knowledge attribute implication association set Y. 5.The attribute exploration-based test item implicit knowledge attribute association mining and related test item pushing system according to claim 4, characterized in that, The wrong question form background construction module is configured to: In step A1, pre-process the existing student question answer record source data, including: cleaning the source data, deleting irrelevant attributes in the source data, and obtaining binary answer information data containing only a student object and the correctness of the answered questions; and deleting the correctly answered questions in the student answer record and retaining the error answer record of the student. In step A2, fuse the knowledge attribute association matrix with the pre-processed data, including: representing the labeled knowledge attribute set by a matrix, representing each row as a question and each column as a knowledge attribute, and using 1 to represent the question containing the knowledge attribute and 0 otherwise, to obtain a knowledge attribute association matrix; combining the pre-processed data with the knowledge attribute association matrix to obtain an error answer information matrix of each error answer question and knowledge attribute fusion, i.e., obtaining a question knowledge attribute form background K of the current student on the wrong question set. 6.The attribute exploration-based test item implicit knowledge attribute association mining and related test item pushing system according to claim 4, characterized in that, The related question pushing module is configured to: In step C1, construct a question knowledge attribute form background L of a question library that has not been practiced by the current student according to the way in the wrong question form background construction module; In step C2, construct a concept lattice of the question library that has not been practiced by the current student by using a bordat concept lattice construction algorithm, and construct the concept lattice of the question library by inputting the form background L derived in step C1; the concept lattice is represented by concept nodes and partial order relations between the nodes, and each node in the concept lattice is represented by a binary relation (U, S), U representing a current student question set that has not been practiced, and S representing a knowledge attribute set; the concept node represents that the question set U has the knowledge attribute S; In step C3, based on the knowledge attribute implication association set derived by the non-redundant question set and knowledge attribute implication association set derivation module, calculate the similarity of the concept node and the knowledge attribute implication relation formula by traversing all concept nodes of the concept lattice by using a cosine similarity calculation formula. C4, according to the cosine similarity calculation result, the question of the related test question is sorted, the threshold value is set, the result greater than the threshold value is screened out, and the top-n test question is selected and pushed, if the number of screened test questions is small, the threshold value is reduced.

Images