JP7179886B2 - Question item automatic generation method and system - Google Patents

Description

One embodiment of the present invention relates to a question item automatic generation method and system.

The following discussion merely provides background information related to the present embodiments and may not constitute prior art.

Recently, the educational environment is changing rapidly through various changes in the surrounding environment due to the use of the Internet and computers. In particular, the development of diverse educational media has enabled learners to choose and use a wider range of learning methods. Among the broader learning methods, Internet-based educational services overcome time and space constraints and enable low-cost education.

Recently, with the rapid development of e-learning-related technology, even with limited human/material resources, it is possible to provide customized educational services that were not possible with off-line education. For example, by providing level-based learning that is subdivided according to the individuality and ability of the learner, we will break away from the uniform educational method of the past and provide educational content according to the individual ability of the learner. be able to.

However, general custom-made educational services take the form of one-sided cramming education. In other words, a general custom-made education service is a method in which the instructor conducts a lecture according to the table of contents or the level of the student, and the student confirms the learning result through the evaluation process after going through the learning process.

In particular, in the case of mathematics education services, it is common for all learners to be presented with the same type of mathematics problem. A general mathematics education service provides a type of problem in which only numbers in a formula used to explain the problem are randomly changed to other numbers for one mathematics problem. The problem type change method described above changes only the numbers in the explanation of the question, so there is a problem that the degree of change of the question is not deep.

In this embodiment, after classifying the question item type for the input question item, similar question items are generated and provided within the same type as the input question item type based on the problem solving library and the natural language processing model. It is an object of the present invention to provide a question item automatic generation method and system for generating and providing modified question items based on the educational process and learning goals within the question item type similar to the item type.

According to one aspect of this embodiment, a model corresponding to an input question item is selected from among a plurality of models already saved in the knowledge space DB as a matching question item model, and based on the matching question item model, a problem item model selection unit that generates problem item type classification information; and a similar problem item information that has the same problem item structure as the problem item type classification information but has changed constituent elements in the problem item; a similar question item generating unit that divides the question item information into preset slot units and then completes and outputs the similar question item information in a form that fills the inside of the slot; and the question item type classification information. and the similarity values mapped between a plurality of pre-stored models, respectively, select a matching similarity type question item model based on the similarity value, and select the same question item model as the matching similarity type question item model Transformed question item information is generated by changing the constituent elements in the question item although it has a structure, and after dividing the transformed question item information into preset slot units, the inside of the slot is filled. and a modified question item generating unit for completing and outputting the modified question item information.

According to another aspect of this embodiment, a process of selecting a model corresponding to an input question item from among a plurality of models already stored in a knowledge space DB as a matching question item model; a step of generating problem item type classification information based on the above; a step of generating similar question item information having the same problem item structure as the problem item type classification information but with changed constituent elements in the problem item; A process of dividing the similar question item information by a preset slot unit and then completing and outputting the similar question item information in a form that fills the inside of the slot; confirming the similarity values mapped between the plurality of models, selecting a matching similar type question item model based on the similarity values; and determining the same question item structure as the matching similar type question item model. However, the process of generating modified question item information in which the constituent elements in the question item are changed, and dividing the modified question item information into units of preset slots (Slots) and filling the slots and completing and outputting the transformed question item information in step 3.

As explained above, according to this embodiment, after classifying the question item type for the input question item, similar question items are generated based on the problem solving library and the natural language processing model within the same type as the input question item type. In addition, it has the effect of generating and providing modified question items based on the educational process and learning goals within the question item type similar to the input question item type.

1 is a block configuration diagram schematically showing a mathematical problem item automatic generation system according to the present embodiment; FIG. 1 is a block configuration diagram schematically showing a question item automatic generation device according to an embodiment; FIG. FIG. 4 is a diagram showing a method of converting input question items according to the present embodiment; FIG. 4 is a flowchart for explaining the operation process of a question item model selection unit according to the embodiment; 4 is a flowchart for explaining the operation process of a similar question item generation unit according to the embodiment; 4 is a flow chart for explaining the operation process of the modified question item generation unit according to the embodiment; 4 is a flow chart for explaining the operation process of the modified question item generation unit according to the embodiment; FIG. 4 is an exemplary diagram showing similar question item information or modified question item information according to the present embodiment;

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

FIG. 1 is a block configuration diagram schematically showing a question item automatic generation system according to this embodiment.

The question item automatic generation system according to this embodiment includes a new question item input unit 110 , an automatic classification unit 120 , a knowledge space DB 130 , a question item automatic generation device 140 and a question confirmation unit 150 . The components included in the question item automatic generation system are not necessarily limited to these.

A new question item input unit 110 receives a new question item due to a change in the educational process or a modification of the learning goal. The new question item includes metadata including categories (grade, semester, major unit, minor unit, detailed type level) to which at least one of the educational process and learning goal is mapped.

The automatic classification unit 120 receives new question items from the new question item input unit 110 . The automatic classifier 120 extracts metadata contained in the new question item. The automatic classification unit 120 automatically classifies new question items based on the information included in the metadata and stores them in the knowledge space DB 130 .

The automatic classification unit 120 models the input new question item and stores it in the knowledge space DB 130 . For example, the automatic classification unit 120 models the problem item specification information for the first grade of junior high school, 3 units, unit information, and model number in the knowledge space DB 130 , converts it into a database, and stores it in the knowledge space DB 130 .

The knowledge space DB 130 means a general data structure implemented in a storage space (hard disk or memory) of a computer system using a database management program (DBMS). The knowledge space DB 130 represents a data storage form capable of searching (extracting), deleting, editing, and adding data. Information stored in the knowledge space DB 130 has fields or elements to achieve their own functions.

The question item automatic generation device 140 selects a model corresponding to the input question item from among a plurality of models already stored in the knowledge space DB 130 as a matching question item model, and based on the matching question item model, creates a question item type. Generate taxonomy information.

The question item automatic generation device 140 generates similar question item information having the same question item structure as the question item type classification information, but with changed constituent elements in the question item. The automatic question item generation device 140 divides the similar question item information into preset slots, and then completes and outputs the similar question item information in a form that fills the slots.

The problem item automatic generation device 140 confirms the problem item type classification information and the similarity values mapped between a plurality of pre-stored models, respectively, and determines the matching similarity type question items that are determined to be suitable among the similarity values. Select model. The question item automatic generation device 140 generates modified question item information having the same question item structure as that of the similar type question item model, but with changed constituent elements in the question item. The automatic question item generation device 140 divides the transformed question item information into preset slots, and then completes and outputs the transformed question item information in a form that fills the slots.

When the similar question item information or the modified question item information output from the automatic question item generation device 140 is confirmed, the question confirming unit 150 inputs the similar question item information or the modified question item information to the automatic classification unit 120 . The question confirmation unit 150 inputs the similar question item information or the modified question item information to the automatic classification unit 120 so that the similar question item information or the modified question item information is learned.

FIG. 2 is a block configuration diagram schematically showing a question item automatic generation device according to the present embodiment.

The question item automatic generation device 140 according to this embodiment includes a question item model selection unit 210 , a similar question item generation unit 220 , and a modified question item generation unit 230 . The components included in the question item automatic generation device 140 are not necessarily limited to these.

Each component included in the automatic question item generation device 140 is connected to a communication path that connects software modules or hardware modules inside the device and can operate organically between them. Such components communicate using one or more communication buses or signal lines.

Each component of the automatic question item generation device 140 shown in FIG. 2 represents a unit that processes at least one function or operation, and is implemented as a software module, a hardware module, or a combination of software and hardware. can be

The question item model selection unit 210 receives an original file configured in a preset format (eg, HWP file format).

The question item model selection unit 210 encodes a preset format (eg, HWP file format) in HML, which is a markup format, or MathML (Mathematical Markup Language) format, which is a markup format for formula expression. Here, MathML is one of the applications of XML, and means a markup language for representing mathematical expressions and understanding their structure and content. MathML is part of HTML5 aimed at integrating mathematical formulas into World Wide Web pages and other documents.

The question item model selection unit 210 confirms metadata included in the input question item. The question item model selection unit 210 maps categories such as grade, semester, major unit, minor unit, detailed type level, etc. to a plurality of models stored in the knowledge space DB 130 based on the mathematics educational process in the metadata.

The question item model selection unit 210 tags the part of speech based on the natural language of the question item input to the question item model corresponding to the narrowed category among the plurality of models stored in the knowledge space DB 130. Then, the formula information is separately constructed, and the patterns are compared and analyzed using an N-gram-based text similarity algorithm.

The question item model selector 210 uses a Korean part-of-speech tagging model based on a bidirectional LSTM (Long Short-Term Memory Network)-CRF (Conditional Random Field) model, which is a deep learning technology, when tagging parts of speech. can be used, but is not necessarily limited to this.

When measuring text similarity, the question item model selector 210 constructs four types of N-gram tokens and compares sentences included in the input question item with patterns of sentences included in a plurality of stored models. do. Here, N-gram means N adjacent characters.

The question item model selection unit 210 compares the sentences included in the input question item with two different sentences included in a plurality of stored models using N-grams, and then compares the type and frequency of words that appear. , the similarity can be checked. For example, when the question item model selection unit 210 inputs sentences such as "I ate delicious spaghetti at Gangnam today" and "Today's spaghetti at Gangnam was delicious." Separate them one by one.

The question item model selection unit 210 selects the input text by preset character units [“today”, “always”, “kang”, “gangnam”, “nam ni”, “de”, “hey”, “shii”. ”. . . ], [ ``Gangnam'', ``Nam ni'', ``De'', ``O'', ``Tabe'', ``Tabe'', ``Tabeta'', ``O'', ``Today'' . . . ].

The question item model selection unit 210 compares the sentences divided by the preset character unit item by item, and increments the count by +1 each time the same item exists. The question item model selection unit 210 divides the total count by the number of items in the segmented sentence array to obtain the degree of similarity.

The question item model selection unit 210 selects a matching question item model in consideration of the input question item model, the learning goal, and the problem item type, and derives the problem item type and question item information based on the matching question item model.

The question item model selector 210 is based on Bidirectional LSTM (Long Short-Term Memory Network)-CRF (Conditional Random Field), which is a deep learning model structure applying a continuous labeling technique for slot parsing of question items. utilizes the slot parsing model of

The question item model selection unit 210 transfers corresponding extraction information (question item type classification information) to the similar question item generation unit 220 or the modified question item generation unit 230 according to the set question item generation method when the system is called. The question item model selection unit 210 generates question item category classification information using a question item category classifier for each input question item using natural language processing technology.

The similar question item generation unit 220 uses a problem solving library (problem solving engine - addition, subtraction, multiplication, random number generation, small number verification, etc.) that can perform core calculations in the curriculum and an automatic generation algorithm for each question item type to generate a modified formula. to generate

The similar question item generation unit 220 uses a question item expression change model based on an attention sequence-to-sequence structure to express the natural language expression of the question item, although the meaning is similar to the existing one. Construct a twin question item structure that is transformed from .

The similar question item generating unit 220 performs slot filling for the similar question item and fills a transformable formula slot in the twin question item structure whose expression is transformed to create a twin question item by filling a transformable formula slot. Generate. Here, slot filling means that when a parameter is "required" but the value is not satisfied, the value is filled using variable conditions, relational expressions between variables, etc. to satisfy the corresponding "parameter value". Means the process of filling.

The similar question item generating unit 220 generates twin question items according to the set number of question items to be generated when the system is called.

After receiving the problem item type classification information from the problem item model selection unit 210, the similar problem item generation unit 220 has the same basic problem item structure as the problem item type classification information, but the sentence (subject , verbs, parts of speech) or formulas (numbers), or change (replace) the order of multiple sentences to generate similar question items. As shown in FIG. 8, the similar question item generation unit 220 generates a new type of similar question item different from the input question item.

After generating a new similar question item, the similar question item generation unit 220 uses a previously saved problem solving library (for example, a math subject library) to check whether the similar question item can actually be solved. Use it to solve a similar question item and then change the example corresponding to the similar question item.

The similar question item generating unit 220 solves similar question items using a pre-stored problem solving library (eg, math subject library), calculates the correct answer, and determines whether the correct answer is among the examples of the corresponding question item. Confirm. As a result of the confirmation, if the correct answer is not found in the example of the corresponding question item, the similar question item creation unit 220 inserts or creates the correct answer in the example of the corresponding question item.

The similar question item generator 220 divides the similar question items into preset slots for natural language processing. The similar question item generation unit 220 divides the similar question items into preset slots in order to determine from where to where the new similar question item is to be determined as one semantic. The similar question item generating unit 220 considers (checks whether nouns and verbs contradict each other) parts in which sentences (subjects, verbs, parts of speech) or formulas (numbers) in question items are changed, and divides slots. do.

The similar question item generator 220 checks whether nouns and verbs in the new similar question item contradict each other, and performs slot parsing if the nouns and verbs do not contradict each other.

The similar question item generation unit 220 checks whether the order of the changed or replaced sentences in the new similar question item is inconsistent, and if the order of the changed or replaced sentences is not inconsistent as a result of the confirmation. , to perform slot parsing.

The similar question item generation unit 220 checks whether nouns and verbs in the new similar question item contradict each other, and if the nouns and verbs contradict each other as a result of the checking, slot filling is performed so that the nouns and verbs do not contradict each other. to fill nouns or verbs to generate other constructs.

The similar question item generation unit 220 checks whether the order of the changed or replaced sentences in the new similar question item is inconsistent. , performs slot filling to add conjunctions in multiple sentences, or add (change) particles in sentences.

The similar question item generator 220 performs slot parsing if the correct answer of the new similar question item is in the example. If there is no correct answer for the new similar question item in the example, the similar question item generator 220 inserts the example into the example based on the correct answer by performing slot filling.

The modified question item generation unit 230 receives question item type classification information from the question item model selection unit 210 . The modified question item generation unit 230 selects matching similar category question item models within a certain range based on the question item category classification information and the similarity values between the pre-stored models, and selects matching similar category question item models as matching similar category question item models. Although the basic question item structure is the same, you can change the contents of sentences (subject, verb, part of speech) or mathematical expressions (numbers) within the question item, or change sentences (subject, verb, parts of speech) or mathematical expressions (numbers). Change (replace) the order to generate a modified question item. As shown in FIG. 8, the modified question item generator 230 generates a modified question item in a new form different from the input question items and the similar question items.

The modified question item generation unit 230 derives a similar type question item list using a mapping dictionary between question item types for the question item model (question item type classification information) input from the question item model selection unit 210 .

The modified question item generation unit 230 analyzes the learning goals and question item types included in the corresponding question items by using the meta information of the question items and the problem solving library for the similar type question item list, and generates a suitable similar type. Select a problem item model.

The modified question item generator 230 generates modified formulas through an automatic generation algorithm for each question item type based on the analysis result of the slot parsing model for the selected matching similar type question item model.

The modified question item generation unit 230 performs slot filling based on the question item expression change model, fills the generated modified formulas in the deformable formula slots in the similar type structure whose expression is changed, and generates similar type question items. do. The modified question item generating unit 230 generates similar type question items according to the set number of question items to be generated when the system is called.

The modified question item generation unit 230 receives question item type classification information from the question item model selection unit 210 . As a result of checking the problem item type classification information, the modified question item generation unit 230 checks the similarity between the question type and the saved model if it is confirmed that the problem type is a question type for confirming the relationship of size. . The modified question item generation unit 230 maps the stored models and similarities to generate a similarity type list.

For example, the modified question item generating unit 230 uses a mapping dictionary between question item types to check that question item types 1 and 3 are mapped and that the corresponding similarity is 0.1. The modified question item generation unit 230 uses a mapping dictionary between question item types to check that question item types 1 and 5 are mapped and that the corresponding similarity is 0.3. The modified question item generation unit 230 checks the question item type and the mapped value, checks a plurality of similarity values, and determines the most suitable value (e.g., a value having a similarity of 0.1). sort out.

The process of finding the solution of the problem using the problem-solving library can be automatically calculated. A problem-solving library means a library for calculating when numbers change. A problem-solving library is a library that computes the solution of a modified problem when a formula is entered.

The modified question item generation unit 230 generates matching similar type information based on the similar type list. The modified question item generator 230 generates modified question items based on the matching similarity type information.

The modified question item generating unit 230 changes the content of sentences (subject, verb, part of speech) or formulas (numbers) in the question item similar to the question item type classification information, or changes the contents of sentences (subject, verb, part of speech) or formulas. Change (replace) the order of (numbers) to generate a modified question item.

After generating the modified question item, the modified question item generating unit 230 solves similar problem items using a pre-stored problem solving library in order to check whether the modified question item can actually be solved. Then change the example.

The modified question item generating unit 230 calculates the correct answer by solving the modified question item using the stored problem solution library, and checks whether the correct answer is among the examples of the corresponding question item. As a result of the confirmation, if the correct answer is not found in the examples of the corresponding question item, the modified question item generation unit 230 inserts or generates the correct answer into the example of the corresponding question item.

The modified question item generating unit 230 divides the modified question items into preset slot units for natural language processing. The modified question item generating unit 230 divides the modified question item into preset slots in order to determine from where to where the modified question item is to be judged as one semantic. The modified question item generation unit 230 considers (checks whether nouns and verbs contradict each other) parts in which sentences (subjects, verbs, parts of speech) or formulas (numbers) are changed in the question item, and classifies slots. do.

The modified question item generator 230 checks whether nouns and verbs in the modified question item contradict each other, and if the nouns and verbs do not contradict each other as a result of checking, slot parsing is performed. The modified question item generation unit 230 checks whether the order of the changed or replaced sentences in the modified question item is inconsistent. Perform slot parsing.

The modified question item generating unit 230 checks whether the noun and the verb in the modified question item are contradictory, and if the noun and the verb are contradictory as a result of the confirmation, slot filling is performed so that the noun and the verb are not contradictory. , to meet nouns or verbs to generate other constructs. The modified question item generation unit 230 checks whether the order of the modified or replaced sentences in the modified question item is inconsistent. Perform slot filling to add conjunctions within multiple sentences or add (change) particles within sentences.

The modified question item generator 230 performs slot parsing if the correct answer of the modified question item is in the example. If there is no correct answer for the modified question item in the example, the modified question item generator 230 inserts the example into the example based on the correct answer by performing slot filling.

FIG. 3 is a drawing showing a method of converting input question items according to the present embodiment.

The question item model selection unit 210 receives input question items (original question items). The question item model selection unit 210 selects a model capable of generating a similar question item type or a modified question item type by mapping an input question item (original question item) from a plurality of models stored in the knowledge space DB 130 .

The input question item is a question item (eg, a math question item) within a general question paper, typically having the HWP file format, as illustrated in FIG.

The question item model selection section 210 generates a converting file by converting the input question items (original question items) having the HWP file format into the HML or MathML file format. The question item model selector 210 compares the converting file (converted to HML or MathML file format) with multiple models (eg, models numbered 1 through 1000) stored in Latex MathML.

The question item model selection unit 210 selects a converting file (converted to HML or MathML file format) and the most suitable model ( For example, the 16th model from among the 1st to 1000th models) is selected to generate problem item type classification information.

FIG. 4 is a flow chart for explaining the operation process of the question item model selection unit according to the present embodiment.

The question item model selection unit 210 converts the input question item (eg, HWP file format) into a preset file format (HML or MathML) to generate converted data (S410). The question item model selection unit 210 extracts metadata contained in the input question item from the converting data (S420).

The question item model selection unit 210 selects a category (grade, semester, major unit, minor unit, detailed model) to which at least one of the educational process and learning goal corresponding to the metadata is mapped among the plurality of pre-stored models. type level) is selected (S430).

The question item model selection unit 210 tags the part of speech for each word recognized by natural language processing of the input question item for the question item model in the category (S440). In step S440, the question item model selection unit 210 uses a part-of-speech tagging model using bidirectional LSTM (Long Short-Term Memory Network)-CRF (Conditional Random Field) when tagging part-of-speech for each word.

The question item model selection unit 210 applies a language model (for example, N-gram Language Model)-based sentence similarity algorithm to each part of speech for each word, compares patterns with a plurality of pre-stored models, and selects matching question items. Select the model (S450). In step S450, when the question item model selection unit 210 compares patterns of a plurality of models stored for each part of speech of each word, adjacent N characters are identified as tokens (for example, four types of N-gram tokens). After constructing, the patterns are compared to calculate the degree of similarity between sentences, and the matching question item model is selected based on the degree of similarity.

The question item model selection unit 210 compares the text included in the input question item with two different texts included in a plurality of stored models using a language model (e.g., N-gram), and then, Check the types of words that appear, the part of speech, and the number of frequencies, and calculate the degree of similarity based on the type of word, the part of speech, and the number of frequencies.

The question item model selection unit 210 classifies the sentences included in the input question item and two different sentences included in a plurality of stored models by a preset character unit (slot unit). The question item model selection unit 210 compares two sentences that are different from each other in preset character units (slot units) by one item (type of word, part of speech), and counts each time the same item exists. ) is incremented (by +1) to calculate the total count. The question item model selection unit 210 calculates the similarity by dividing the total count by the number of items in the segmented sentence array.

The question item model selection unit 210 selects a model having a high degree of similarity with the input question item from among a plurality of stored models as a combined question item model, and classifies the question items included in the matching question item model as a question item type. Generate as classification information.

In FIG. 4, steps S410 to S450 are described as being executed sequentially, but the present invention is not necessarily limited to this. In other words, FIG. 4 is limited to a chronological order, as the steps described in FIG. 4 may be modified to be performed, or one or more steps may be performed in parallel. not something.

As described above, the operation process of the question item model selection unit according to the present embodiment illustrated in FIG. 4 can be implemented as a program and recorded in a computer-readable recording medium. A program for implementing the operation process of the question item model selection unit according to the present embodiment is recorded, and the computer-readable recording medium includes all kinds of recording devices that store data that can be read by a computer system.

FIG. 5 is a flow chart for explaining the operation process of the similar question item generation unit according to the present embodiment.

The similar question item generating unit 220 generates similar question item information having the same question item structure as the question item type classification information, but with changed constituent elements in the question item (S510). In step S510, the similar question item generating unit 220 changes the contents of sentences (subjects, verbs, parts of speech) or formulas (numbers) within the question item, although the basic question item structure is the same as the question item type classification information. Alternatively, the order of sentences (subject, verb, part of speech) or mathematical expressions (numbers) is changed (replaced) to generate similar question item information.

After generating the similar-question item information, the similar-question item generation unit 220 generates similar-question item information using a pre-stored problem-solving library to check whether the similar-question item information can actually be solved. A solution to the information is obtained (S520).

The similar question item generation unit 220 changes the example information included in the similar question item information based on the solution (S530). In step S530, the similar question item generation unit 220 solves the similar question item information using a stored problem solving library (library for calculating the solution of the input formula), and when the solution is calculated, the solution is calculated. After setting the correct answer among the examples, it is checked whether the correct answer exists in the examples of the similar question item information.

As a result of the confirmation, if the correct answer does not exist in the example of the similar question item information, the similar question item generation unit 220 inserts the correct answer into the example. In other words, the similar question item generating unit 220 performs slot parsing when the correct answer exists in the example of similar question item information, and performs slot filling when the correct answer does not exist in the example of similar question item information. and insert one as the correct answer in the example.

The similar question item generation unit 220 performs natural language processing on the similar question item information and divides it into preset slots in order to recognize it as semantic (S540). In step S540, the similar question item generating unit 220 divides slots by considering the contents of sentences (subject, verb, part of speech) or formulas (numbers) changed in the similar question item.

The similar question item generation unit 220 performs slot parsing or slot filling depending on whether there is contradiction in the changed components in the similar question item information for each slot (S550). In step S550, the similar question item generator 220 checks whether sentences (subjects, verbs, parts of speech) or formulas (numbers) changed in units of slots in the similar question item information are inconsistent.

As a result of checking, if the sentences (subject, verb, part of speech) or formulas (numbers) are not inconsistent, the similar question item generation unit 220 performs two-way LSTM-CRF-based slot parsing using a continuous labeling technique. do.

As a result of confirmation, if sentences (subject, verb, part of speech) or formulas (numbers) are inconsistent, the similar question item generation unit 220 changes the expression of the question item based on the attention sequence-to-sequence structure. Using the model, the natural language expression of the question item is similar to the existing meaning, but the twin question item structure is constructed so that the sentence (subject, verb, part of speech) or mathematical expression (number) does not contradict. do. The similar question item generation unit 220 performs slot filling to assign twin question item structures to slots that are not filled due to contradiction among slots to which essential parameter values are assigned. to complete the similar question item information (twin question item information).

The similar question item generation unit 220 performs slot parsing or slot filling depending on whether there is a contradiction in the order of the changed components in the similar question item information for each slot (S560). In step S560, the similar question item generation unit 220 checks whether the order of sentences (subjects, verbs, parts of speech) or formulas (numbers) changed (replaced) in the similar question item information is inconsistent.

As a result of checking, if there is no contradiction in the order of the changed (replaced) sentences (subject, verb, part of speech) or formula (number), the similar question item generation unit 220 generates two-way LSTM-CRF based continuous labeling technique. Perform slot parsing.

As a result of the confirmation, if the order of the changed (replaced) sentences (subject, verb, part of speech) or mathematical expressions (numbers) is inconsistent, the similar question item generation unit 220 generates an attention sequence-to-sequence. ) Construct a twin question item structure by adding a conjunction in the similar question item information or adding (changing) a particle in the sentence using the question item expression change model of the structure base.

The similar question item generation unit 220 performs slot filling to assign twin question item structures to slots that are not filled due to contradiction among slots to which essential parameter values are assigned. to complete the similar question item information (twin question item information).

Although steps S510 to S560 are described as sequentially executed in FIG. 5, the present invention is not necessarily limited to this. In other words, FIG. 5 is limited to chronological order, as the steps described in FIG. not something.

As described above, the operation process of the similar question item generator according to the present embodiment illustrated in FIG. 5 can be implemented as a program and recorded in a computer-readable recording medium. A program for implementing the operation process of the similar question item generator according to the present embodiment is recorded, and the computer-readable recording medium includes all kinds of recording devices in which data readable by a computer system is stored.

6 and 7 are flow charts for explaining the operation process of the modified question item generation unit according to this embodiment.

The modified question item generation unit 230 checks the question item type classification information and the similarity values mapped between the plurality of stored models (S610).

The modified question item generating unit 230 generates a similar type list based on the similarity values mapped between the question item types included in the question item type classification information and the pre-stored models (S620). In step S620, the modified question item generator 230 extracts similarity values within a preset critical range to generate a similarity type list.

The modified question item generation unit 230 extracts metadata contained in the input question item from the converting data, and maps at least one of the educational process and learning goal included in the metadata in the similar type list. A model having a category (grade, semester, major unit, minor unit, detailed category level) is selected as a matching similar category question item model (S630).

The modified question item generation unit 230 has the same basic question item structure as the matching similar type question item model, but changes the contents of sentences (subjects, verbs, parts of speech) or formulas (numbers) in the question item, The order of sentences (subject, verb, part of speech) or formula (numbers) is changed (replaced) to generate modified question item information (S640).

After generating the transformed question item information, the transformed question item generation unit 230 generates transformed question items using a pre-stored problem solution library in order to check whether the transformed question item information can actually be solved. A solution to the information is obtained (S650).

The modified question item generator 230 changes the example information included in the modified question item information based on the solution (S660). In step S660, the transformed question item generation unit 230 solves the transformed question item information using the stored problem solving library (library for calculating the solution of the input mathematical expression) and calculates the solution. After setting the correct answer in the example, it is checked whether the correct answer exists in the example of the modified question item information.

As a result of checking, if the correct answer does not exist in the example, the correct answer is inserted into the example. In other words, the modified question item generating unit 230 performs slot parsing when the correct answer exists in the example of similar question item information, and performs slot filling when the correct answer does not exist in the example of similar question item information. and insert one as the correct answer in the example.

The modified question item generation unit 230 performs natural language processing on the modified question item information and divides it into preset slots in order to recognize it as semantic (S670). In step S670, the modified question item generation unit 230 divides the slots by considering the content of sentences (subjects, verbs, parts of speech) or formulas (numbers) changed in the modified question item.

The modified question item generator 230 performs slot parsing or slot filling depending on whether there is contradiction in the changed components in the similar question item information for each slot (S680). In step S680, the modified question item generation unit 230 checks whether sentences (subjects, verbs, parts of speech) or formulas (numbers) changed in units of slots in the modified question item information are inconsistent.

As a result of checking, if sentences (subjects, verbs, parts of speech) or formulas (numbers) are not inconsistent, the modified question item generator 230 performs two-way LSTM-CRF-based slot parsing using a continuous labeling technique. do.

As a result of checking, if sentences (subject, verb, part of speech) or formulas (numbers) are inconsistent, the modified question item generator 230 changes the question item expression based on the Attention Sequence-to-Sequence structure. Using the model, the natural language expression of the question item is similar to the existing meaning, but the twin question item structure is constructed so that the sentence (subject, verb, part of speech) or mathematical expression (number) does not contradict. do. The modified question item generation unit 230 performs slot filling to assign twin question item structures to slots that are not filled due to contradiction among slots to which essential parameter values are assigned. to complete the transformed question item information (twin question item information).

The modified question item generating unit 230 performs slot parsing or slot filling depending on whether there is a contradiction in the order of the changed components in the modified question item information for each slot (S690). In step S690, the modified question item generation section 230 checks whether the order of the sentences (subject, verb, part of speech) or formulas (numbers) changed (replaced) in the modified question item information is inconsistent.

As a result of checking, if there is no contradiction in the order of the changed (replaced) sentences (subject, verb, part of speech) or mathematical expressions (numbers), the modified question item generation unit 230 generates two-way LSTM-CRF based continuous labeling technique. Perform slot parsing.

As a result of the confirmation, if the order of the changed (replaced) sentences (subject, verb, part of speech) or mathematical expressions (numbers) is inconsistent, the modified question item generation unit 230 generates an attention sequence-to-sequence. ) Construct a twin question item structure by adding a conjunction in the modified question item information or adding (changing) a particle in the sentence using the question item expression change model of the structure base.

The modified question item generation unit 230 performs slot filling to assign twin question item structures to slots that are not filled due to contradiction among slots to which essential parameter values are assigned. to complete the transformed question item information (twin question item information).

Although steps S610 to S690 are described as sequentially executed in FIG. 6, the present invention is not necessarily limited to this. In other words, FIG. 6 is limited to chronological order, as the steps described in FIG. not something.

As described above, the operation process of the modified question item generator according to the present embodiment illustrated in FIG. 6 can be implemented as a program and recorded in a computer-readable recording medium. A program for implementing the operation process of the modified question item generator according to the present embodiment is recorded, and the computer-readable recording medium includes all kinds of recording devices that store data that can be read by a computer system.

The above description is merely illustrative of the technical idea of this embodiment, and those who have ordinary knowledge in the technical field to which this embodiment belongs will not deviate from the essential characteristics of this embodiment. A wide variety of modifications and variations would be possible. Therefore, the present embodiment is not intended to limit the technical idea of the present embodiment but to explain it, and the scope of the technical idea of the present embodiment is not limited by such an embodiment. . The protection scope of this embodiment should be construed according to the following claims, and all technical ideas within the equivalent scope should be construed as included in the scope of rights of this embodiment. .

110: New question item input unit 120: Automatic classification unit 130: Knowledge space DB
140: Question item automatic generation device 150: Question confirmation unit 210: Question item model selection unit 220: Similar question item generation unit 230: Modified question item generation unit

Claims (7)

A problem in which a model corresponding to an input question item is selected from among a plurality of models already stored in a knowledge space DB as a matching question item model, and problem item type classification information is generated based on the matching question item model. an item model selection unit;
Generating similar question item information having the same question item structure as the question item type classification information but changing the constituent elements in the question item, and classifying the similar question item information by preset slot units a similar question item generation unit that completes and outputs the similar question item information in a form that fills the inside of the slot after the completion of the similar question item information;
confirming similarity values mapped between the problem item type classification information and a plurality of pre-stored models, selecting a matching similarity type question item model based on the similarity values, and selecting the matching similarity type question item Although it has the same question item structure as the model, modified question item information is generated by changing the components in the question item, and after dividing the modified question item information into preset slot units, the slot a modified question item generating unit that completes and outputs the modified question item information in a form that fills the inside of the
including
The question item model selection unit classifies the text included in the input question item and two different texts included in the plurality of stored models by a preset character unit, and Comparing the two sentences different from each other one by one according to the type of word or the part of speech, calculating the total count by increasing the count each time the same item exists, and calculating the total count as a segmented sentence array After calculating the value obtained by dividing by the number of items as the similarity value, selecting the matching question item model based on the similarity value,
The similar question item generating unit checks whether the nouns and verbs in the similar question item information are contradictory or not, and whether or not the order of the changed or replaced plural sentences is contradictory. If not, slot parsing is performed, and if a contradiction occurs as a result of confirmation, a twin question item structure is generated, and slot filling is performed on the twin question item structure to divide it into a noun and a noun. fill nouns or verbs to generate other constructions, add conjunctions within multiple sentences, add particles within sentences, and
The modified question item generating unit checks whether the nouns and verbs in the modified question item information contradict each other, and whether the order of the changed or replaced plural sentences contradicts each other. If there is no contradiction, slot parsing is performed, and if a contradiction occurs as a result of confirmation, a twin question item structure is generated and slot filling is performed on the twin question item structure so that nouns and verbs are not contradictory. A question item automatic generation device characterized by satisfying verbs to generate other constructions, adding conjunctions in a plurality of sentences, and adding particles in sentences. The question item model selection unit
converting the input question items into a preset file format to generate converting data; performing natural language processing on the converting data to classify question item types for each input question item; 2. A question item automatic generation apparatus according to claim 1, wherein a model corresponding to said question item type is selected as said matching question item model from among a plurality of models. The similar question item generation unit
While maintaining the same basic question item structure as the problem item type classification information, the similar question item information is generated by changing the contents of sentences or mathematical formulas in the question item or changing the order of sentences or mathematical formulas. The question item automatic generation device according to claim 1, characterized by: The similar question item generation unit
4. The automatic generation of question items according to claim 3, wherein said slots are filled based on whether or not said similar question item information has been solved and based on the result of natural language processing for said similar question item information confirmed in units of said slots. Device. The modified question item generating unit
generating a similar type list based on the similarity mapped between the problem item type included in the problem item type classification information and the previously saved model, and including the input problem item in the similar type list; 5. The question item automatic generation apparatus according to claim 4, wherein a model having a category mapped to metadata is selected as the matching similar type question item model. The modified question item generation unit
While maintaining the same basic question item structure as the matching similar type question item model, the modified question item information is generated by changing the contents of the sentences or formulas in the question item or by changing the order of the sentences or formulas. 6. The question item automatic generation device according to claim 5, characterized in that: When the question item model selection unit selects a model corresponding to an input question item from among a plurality of models already stored in the knowledge space DB as a matching question item model , the text included in the input question item and The two different sentences included in the plurality of pre-stored models are classified by a preset character unit, and the two different sentences are compared one by one with word types or parts of speech in the character unit. Each time the same item exists, the count is incremented to calculate the total count, and the value obtained by dividing the total count by the number of items in the segmented sentence sequence is calculated as the similarity value. selecting the matching question item model based on
a step of generating problem item type classification information based on the matching problem item model in the problem item model selection unit ;
a process of generating similar question item information in which the similar question item generation unit has the same question item structure as the question item type classification information but has changed constituent elements in the question item;
When the similar question item information is divided into preset slots by the similar question item generating unit and the similar question item information is completed and output in a form that fills the slot , the similar question item information is output. Check whether the nouns and verbs in the question item information are inconsistent or not, and if the order of the changed or replaced multiple sentences is inconsistent, and if there is no inconsistency as a result of the confirmation, perform slot parsing; If a contradiction occurs as a result of the confirmation, a twin question item structure is generated, slot filling is performed on the twin question item structure, and another construction is generated by filling the noun or verb so that the noun and verb are not contradictory. the process of adding conjunctions within multiple sentences and adding particles within sentences ;
A process of confirming the similarity values mapped between the question item type classification information and the plurality of pre-stored models in the modified question item generation unit , and selecting a matching similar type question item model based on the similarity values. When,
a step of generating transformed question item information in which the transformed question item generation unit has the same question item structure as the matching similar type question item model but has changed components within the question item;
When the transformed question item information is divided by the preset slot unit in the transformed question item generating unit and then the transformed question item information is completed and output in a form that fills the inside of the slot , the transformed question item information is output. Check whether the nouns and verbs in the question item information are inconsistent or not, and if the order of the changed or replaced multiple sentences is inconsistent, and if there is no inconsistency as a result of the confirmation, perform slot parsing; If a contradiction occurs as a result of the confirmation, a twin question item structure is generated, slot filling is performed on the twin question item structure, and another construction is generated by filling the noun or verb so that the noun and verb are not contradictory. the process of adding conjunctions within multiple sentences and adding particles within sentences ;
A question item automatic generation method, comprising:

Images