JP7082333B2 - Question automatic generation program and question automatic generation device - Google Patents

Description

The present invention relates to a technique for automatically generating a question regarding the content of a read sentence.

Traditionally, in educational settings, the task of creating test questions based on textbooks, reference books, specialized books, etc. is often left to each faculty member, and the task of creating test questions is a burden on faculty members. There was a case. In addition, when questions are created for each individual, there is a problem that sharing between faculty and staff, accumulation of data across years, and handing over are not performed well.

From the above situation, a system that automatically generates questions has been desired. For example, Patent Document 1 is an example of a problem that is automatically created. This Patent Document 1 discloses an automatic problem generation device that automatically generates a problem sentence from a content described in a predetermined format existing on the Internet.

Japanese Unexamined Patent Publication No. 2012-018300

The patent document 1 has a function of automatically generating a problem statement from the contents described in a predetermined format. However, there is a problem that the documents that can generate the problem sentence are limited because the problem sentence cannot be generated from the sentence that deviates from the fixed format.

The present invention has been made in view of the above problems, and an object of the present invention is to provide an automatic question generation program and an automatic question generation device capable of automatically generating questions from various sentences.

The question automatic generation program according to the present invention is a question automatic generation program for realizing a process of automatically generating a question regarding the content of a read sentence on a computer, and the text data of the sentence to be question-generated on the computer. The text data acquisition function to acquire the text data, the logical structure extraction function to extract the technical terms contained in the acquired text data, and the logical structure between a plurality of technical terms, and the question about the technical terms using the logical structure. It is characterized by realizing a question generation function that generates.

Further, the logical structure extraction function includes a function of classifying the relationship of a logical structure between a plurality of technical terms by a logical structure classification including definition, confrontation, inclusion, ancillary properties, etc., and the question generation function is described above. It is characterized by including a function of generating a question based on an algorithm according to a logical structure classification.

Further, the logical structure extraction function is characterized by including a function of automatically generating a reverse logical structure of the logical structure based on the extracted logical structure.

Further, the text data acquisition function has an image area classification function for classifying images for each area according to a predetermined classification such as a table of contents, a text, a chart, and an index for image data of a sentence to be question-generated, and a function for classifying each area after classification. It is characterized by including an image text conversion function for converting text data by optical character recognition and acquiring text data with tag information of a predetermined classification.

Further, the extraction of technical terms in the logical structure extraction function is performed using a technical term database.

Further, the technical term database is characterized in that at least a part of the text data related to the index portion of the book including the text data of the text to be asked is used.

Further, the logical structure extraction function is characterized in that the logical structure between technical terms included in the text data is extracted based on a predetermined keyword corresponding to the logical structure.

Further, the logical structure extraction function is characterized in that the acquired text data is decomposed for each sentence and the logical structure is extracted for each sentence.

Further, the logical structure extraction function is characterized by including a function of extracting a logical structure from a text data portion related to an acquired chart.

Further, the computer is characterized by realizing a logical structure table output function for outputting the relationship of the logical structure between technical terms expressed by the logical structure extracted by the logical structure extraction function in a tabular format.

The question automatic generation device according to the present invention is a question automatic generation device that automatically generates a question regarding the content of a read sentence, and has a text data acquisition unit for acquiring text data of the sentence to be asked and an acquired text. The logic is provided with a logical structure extraction unit that extracts technical terms contained in data and extracts a logical structure between a plurality of technical terms, and a question generation unit that generates questions related to technical terms using the logical structure. The structure extraction unit is configured to classify the relationship of the logical structure between a plurality of technical terms by the logical structure classification including definition, confrontation, inclusion, incidental property, etc., and the question generation unit is the logical structure classification. It is characterized in that it is configured to generate a question based on an algorithm corresponding to the above .

According to the question automatic generation device according to the present invention, text data of a sentence to be question-generated is acquired, technical terms included in the acquired text data are extracted, and a logical structure between a plurality of technical terms is extracted. Since questions related to technical terms are generated using a logical structure, questions can be automatically generated from text data such as textbooks, technical books, and reference books, which makes it possible to reduce the burden on faculty and staff. In addition, if the generated questions are stored and created in a database, it will be possible to use them for sharing among faculty and staff, accumulation of data across years, and handing over. Furthermore, since it is possible to output a logical structure table that shows the relationship of the logical structure used to generate the question, the logical structure table can be used as an auxiliary teaching material for understanding textbooks, etc., or as a material for re-learning after the test. It becomes possible to do.

It is a block diagram which showed the structure of the question automatic generation apparatus 10 which concerns on this invention. It is a flowchart which showed the flow of the question automatic generation processing in the question automatic generation apparatus 10. It is an explanatory diagram showing an example of a textbook including the contents for which a question is to be generated. It is explanatory drawing which showed the content which converted the page of the textbook shown in FIG. 3 into text data. It is explanatory drawing which showed the extraction result when the logical structure was extracted for each of the extraction sentences shown in FIG. It is explanatory drawing which showed an example of the generation of the logical structure in the reverse direction. It is explanatory drawing which showed an example of the question which is automatically generated based on a logical structure. It is explanatory drawing which showed an example of the logical structure table which showed the relationship of a logical structure. It is an explanatory diagram showing an example of a chart included in a textbook or the like.

Hereinafter, an example of the question automatic generation device according to the embodiment will be described with reference to the drawings. FIG. 1 is a block diagram showing a configuration of a question automatic generation device 10 according to the present invention. The question automatic generation device 10 may be a device designed as a dedicated machine, but it is assumed that it can be realized by a general computer. In this case, the question automatic generation device 10 includes a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), a memory, and a hard disk drive, which a general computer would normally have. It is assumed that it is equipped with storage such as (not shown). Further, it goes without saying that various processes are executed by the program in order to make these general computers function as the question automatic generation device 10 of this example. Further, it is not necessary for the question automatic generation device 10 itself to have all the components of the question automatic generation device 10 described below, and a partial configuration can be provided in another device, for example, a connection via a communication network is possible. The server device may be provided with a part of the configuration, and the question automatic generation device 10 may use the configuration provided in the server device while communicating.

As shown in FIG. 1, the question automatic generation device 10 includes at least a text data acquisition unit 11, a logical structure extraction unit 12, a question generation unit 13, a logical structure table output unit 14, and a storage unit 15. There is. Further, in this example, the text data acquisition unit 11 includes an image area classification unit 111 and an image text conversion unit 112.

The text data acquisition unit 11 has a function of acquiring text data of a sentence for which a question is generated. It is a configuration for acquiring text data of sentences including the contents for which you want to generate questions such as textbooks, reference books, and specialized books. It also has a function to convert books, specialized books, etc. into image data by an image scanner, and to execute a process of converting the contents of sentences included in the image data into text in the following image area classification unit 111 and image text conversion unit 112. .. The acquired text data may remain in a state in which a plurality of sentences are continuously described, but in this example, in order to execute the process one sentence at a time in the logical structure extraction unit 12 described later. In the text data acquisition unit 11, the acquired text data is decomposed into sentences and stored. At that time, the cells may be stored separately for each sentence in the tabular data.

The image area classification unit 111 has a function of classifying the image data of the text to be question-generated into each area according to a predetermined classification such as a table of contents, a text, a chart, and an index. Since some books such as textbooks have different contents such as table of contents, text, charts, and indexes, classify them. As the classification method, an algorithm (which may be discrimination by machine learning) that automatically discriminates from the description contents may be adopted, or the user may specify the classification for each area.

The image text conversion unit 112 converts the image data for which classification is specified for each area in the image area classification unit 111 into text data for each area after classification by optical character recognition (OCR), and tags for a predetermined classification. It has a function to acquire text data with information. Areas classified into the text are acquired as text data of the text, areas classified into charts are acquired as text data of charts, and so on, and are acquired as text data with tag information of each classification. Further, the text data about the area classified in the index may be used as a technical term database described later.

The logical structure extraction unit 12 has a function of extracting technical terms included in the acquired text data and extracting a logical structure between a plurality of technical terms. The reason for extracting the technical term is that it is necessary to determine whether or not the contained word is a technical term in order to extract the logical structure for the part related to the technical term. For the extraction of technical terms, for example, a technical term database may be used. The technical term database may be a technical term database related to sentences created and stored in advance. Further, the text data read about the area classified in the index may be used as it is as a technical term database as it is a text that is likely to contain technical terms.

The logical structure extraction process is a process of extracting as a logical structure representing a relationship between words including technical terms, and specifically, a specialty included in text data based on a predetermined keyword corresponding to the logical structure. Extract the logical structure between terms. For example, for the sentence "A is classified into B and C", if A, B, and C are technical terms, the logic "A → (including) → B, C" is used from this sentence. The structure and the logical structure of "B-> (opposite)-> C" can be extracted. This memorizes in advance the logical structure of inclusion and the regularity that there is a high possibility that the logical structure of opposition can be extracted when the keyword "... is classified into ... and ..." is included. Then, when the keyword "classified" is found, the process of extracting the logical structure of inclusion and the logical structure of the opposite is executed. In addition, there are keywords for extracting the logical structure of inclusion such as "~ has ~" and "~ contains ~", and "~ does not exist ~" and "~ exists". The logical structure between technical terms is extracted based on various keywords such as the keyword for extracting the logical structure that defines the relationship between existence and non-existence such as "...". The extraction of the logical structure is not limited to the case where all the words for which the logical relationship is to be extracted are technical terms, and the logical structure may be extracted even when only one of them is a technical term. As for the definition of words, not only the logical structure between words but also the entire definition part may be extracted as a logical structure. For example, when the keyword "... is ..." is used as a condition for extracting the logical structure of the definition, the entire explanation part of the definition may be extracted instead of the word. In most cases, it is highly possible that some logical structure can be extracted from a sentence in which two or more technical terms appear, although there are exceptions. Therefore, it is desirable to set a possible expression as a keyword for expressing the relationship between two or more technical terms as a keyword for extracting a logical structure as much as possible.

Further, for each of the logical structures extracted by the logical structure extraction unit 12, a logical structure in the opposite direction is generated. The logical structure in the reverse direction is a logical structure that expresses the relationship when the logical structure can be expressed in the reverse direction. Specifically, for example, from the logical structure of "A-> (included)-> B, C", the logical structure of "B-> (included)-> A" and "C-> (included)-> A" in the opposite direction. It means to extract. Extracting the logical structure in the opposite direction has the effect of increasing the choices of the logical structure used for question generation.

The question generation unit 13 has a function of generating questions related to technical terms using a logical structure. More specifically, the content of the question to be generated is determined according to the type of logical structure. For example, from the logical structure of "B-> (including)-> D, E", it is possible to generate the question "B is a concept including D. Is it 〇 or ×?". If there are three logical structures, "B-> (including)-> D, E", "C-> (including)-> F, G", and "B-> (opposite)-> C", these three logics. Using all the structures, it is possible to generate the question "Select what is included in B. Correct choices: D, E, Incorrect choices: F, G". It is also possible to consider a method of generating a question by picking up a plurality of logical structures related to a definition, using the element of the content of the definition to be asked as a question as a correct answer option, and using the content of another definition as an incorrect answer option. These are just examples, and it is possible to automatically generate various questions by utilizing the relationships of various logical structures. As for the number of questions to be generated, the number of questions desired by the user may be specified, or all the questions that can be generated from the logical structure may be generated without duplication.

The logical structure table output unit 14 has a function of outputting the relationship of the logical structure between the technical terms expressed by the logical structure extracted by the logical structure extraction unit 12 in a tabular form. Based on the logical structure, the question generation unit 13 not only generates questions and asks questions in the test, but also outputs a table showing the relationship of the logical structure between technical terms, for example, re-learning after the test. It is possible to use the logical structure table as a reference material.

The storage unit 15 has a function of storing various data required for processing of each unit and data obtained as a result of processing in each unit. For example, the text data acquired by the text data acquisition unit 11, the logical structure extracted by the logical structure extraction unit 12, the questions generated by the question generation unit 13, the table generated by the logical structure table output unit 14, the technical term database, and the like can be found. It may be stored in the storage unit 16.

Next, the flow of the question automatic generation process in the question automatic generation device 10 will be described. FIG. 2 is a flowchart showing the flow of the question automatic generation process in the question automatic generation device 10. In FIG. 2, first, the question automatic generation device 10 acquires text data (S101). The text data may be acquired by any method, and for example, the text included in the image data may be converted into text data by optical character recognition processing (OCR processing). Then, the question automatic generation device 10 classifies the text data acquired according to a predetermined classification such as a headline, a text, a chart, an index, etc. (S102). When converting to text by OCR processing, classification may be performed at the stage of image data, and text may be converted after classification. Then, with respect to the acquired and classified text data, a process is performed in which a plurality of parts are concatenated to form a sentence, particularly the text is decomposed into sentences one by one (S103). This is to execute the extraction process of the logical structure in sentence units.

Next, the question automatic generation device 10 executes the extraction process of the logical structure sentence by sentence (S104). To extract the logical structure, try to extract the logical structure not only from the text but also from figures and tables if possible. Further, the question automatic generation device 10 generates a logical structure in the reverse direction with respect to the extracted logical structure (S105). The question automatic generation device 10 automatically generates a predetermined number of questions based on both the forward and reverse logical structures (S106). Then, the question automatic generation device 10 outputs the generated question (S107). The output may be displayed on a display screen for confirmation, or may be printed out on paper or the like. Further, as a matter of course, the process of storing in the storage unit 16 is also executed. If necessary, a logical structure table showing the relationship between the logical structures is also output (S108), and the process is terminated.

Next, the operation of the automatic question generation device 10 will be described with reference to a more specific example. FIG. 3 is an explanatory diagram showing an example of a textbook containing the content for which a question is to be generated. As shown in FIG. 3, the textbook includes a heading part, a text part, a chart part, and the like. Further, it is assumed that FIG. 3 shows a textbook converted into image data by an image scanner. In that case, a process of classifying the contents included in the image of FIG. 3 into each area is executed. The classification process may be classified by an algorithm that automatically discriminates from the description contents, or may be a method in which the user specifies the classification for each area. After classification, it is converted into text data for each area by OCR processing.

FIG. 4 is an explanatory diagram showing the contents of the textbook pages shown in FIG. 3 converted into text data. As shown in FIG. 4, the heading portion and the extracted sentence extracted as the text are distinguished and converted into text. In addition, when it is converted into text, what was formed by concatenating multiple parts into sentences is decomposed into sentences and managed. The data at this time may be stored sentence by sentence in different cells of the tabular data as shown in FIG.

FIG. 5 is an explanatory diagram showing the extraction results when the logical structure is extracted for each of the extracted sentences shown in FIG. As shown in FIG. 5, four logical structures including the logical structure of "stochastic influence-> (opposite)-> deterministic influence" are extracted from the first sentence. From the second sentence, the logical structure of "stochastic effect-> (including)-> carcinogenesis, genetic effect" is extracted. From the third sentence, the logical structure of "deterministic effect-> (including)-> cataract, skin inflammation" is extracted.

FIG. 6 is an explanatory diagram showing an example of generation of a logical structure in the reverse direction. When the logical structure of "probabilistic effect-> (including)-> (effect)-> carcinogenesis, genetic effect" is extracted as the forward logical structure, the reverse logical structure is "carcinogenesis-> (included)". Two opposite logical structures are generated: → (classification of effects) → probabilistic effects” and “genetic effects → (included) → (classification of effects) → probabilistic effects”. Such a logical structure in the opposite direction is also used for question generation.

FIG. 7 is an explanatory diagram showing an example of a question automatically generated based on a logical structure. For example, the processing logic for automatically generating a question that selects the correct answer from multiple choices is (1) selecting the logical structure of the confrontation, and (2) automatically collecting the elements included in each of the logical structures of the confrontation. The question is generated in the flow of (3) creating a question by using the elements included in the confrontation for which the correct answer is to be answered as the correct answer option and the elements contained in the rest of the confrontation as the incorrect answer option. By generating a question from the logical structure extracted in FIG. 5 according to this processing logic, for example, the question of Example 1 or Example 2 can be generated. It should be noted that Examples 1 and 2 are questions in which the relationship between the correct answer option and the incorrect answer option is reversed, and the logical structures used for creating the example 1 and the example 2 are exactly the same.

FIG. 8 is an explanatory diagram showing an example of a logical structure table showing the relationship of logical structures. When the above-mentioned processing logic for automatic question generation is executed in the explanation of FIG. 7, the relationship of the essay structure used for the question generation is extracted. Questions are generated using the relationships of the extracted logical structures, but if the relationships of these logical structures are output as they are as a logical structure table as shown in Fig. 8, it will help the learner's understanding. It will be possible to use it as a supplementary teaching material. Since this logical structure table can be created at the time of question generation, it may be distributed and used as a review material after the test.

FIG. 9 is an explanatory diagram showing an example of a chart included in a textbook or the like. FIG. 9 shows that the classification of radiation effects includes stochastic and deterministic effects, that carcinogenic and genetic effects are included as specific examples of stochastic effects, and that cataracts are specific examples of deterministic effects. It can be read that skin inflammation is included. When the chart shown in FIG. 9 is shown, for example, in the chart area of FIG. 3 and the text does not include this content, the logical structure is extracted from the chart of FIG. 9 and shown in FIG. If the logical structure table as shown can be created, it is possible to extract the logical structure from the contents contained only in the chart and generate a question. In the example of FIG. 9, it is determined that it is most natural to interpret that the logical structure of inclusion exists between the left cell and the right cell of the chart, and the logical structure of inclusion is extracted. There is. It is also possible to extract a logical structure in which the items described in parallel on the vertical axis are in a contradictory relationship. In this way, the possibility that the logical structure will be extracted by estimation increases as compared with the case of sentences, but the logical structure can also be extracted from figures and tables.

10 Automatic question generation device 11 Text data acquisition unit 111 Image area classification unit 112 Image text conversion unit 12 Logical structure extraction unit 13 Question generation unit 14 Logical structure table output unit 15 Storage unit

Claims (10)

It is a question automatic generation program to realize the process of automatically generating questions about the contents of the read sentences on the computer.
To the computer
A text data acquisition function that acquires the text data of the text for which the question is generated, and
In extracting the technical terms contained in the acquired text data and extracting the logical structure between multiple technical terms, the definition, confrontation, inclusion, incidental properties, etc. of the relationship of the logical structure between multiple technical terms are defined. The logical structure extraction function that classifies by the included logical structure classification ,
An automatic question generation program that realizes a question generation function that generates questions based on an algorithm corresponding to the logical structure classification when generating questions related to technical terms using the logical structure. The question automatic generation program according to claim 1 , wherein the logical structure extraction function includes a function of automatically generating a reverse logical structure of the logical structure based on the extracted logical structure. The text data acquisition function is an image area classification function that classifies images for each area according to a predetermined classification such as a table of contents, a text, a chart, and an index for image data of a sentence to be question-generated, and an optical image for each area after classification. The question automatic generation program according to any one of claim 1 or 2 , which includes an image text conversion function for converting text data by target character recognition and acquiring text data with tag information of a predetermined classification. The automatic question generation program according to any one of claims 1 to 3 , wherein the technical term extraction in the logical structure extraction function is performed using the technical term database. The automatic question generation program according to claim 4 , wherein the technical term database uses at least a part of the text data related to the index portion of the book including the text data of the text to be asked. The question according to any one of claims 1 to 5 , wherein the logical structure extraction function extracts a logical structure between technical terms included in text data based on a predetermined keyword corresponding to the logical structure. Automatic generation program. The question automatic generation program according to any one of claims 1 to 6 , wherein the logical structure extraction function decomposes the acquired text data into sentences and extracts the logical structure for each sentence. The automatic question generation program according to any one of claims 1 to 7 , wherein the logical structure extraction function includes a function of extracting a logical structure from a text data portion related to an acquired chart. To the computer
The method according to any one of claims 1 to 8 , which realizes a logical structure table output function that outputs the relationship of the logical structure between the technical terms expressed by the logical structure extracted by the logical structure extraction function in a tabular format. Question automatic generation program. It is a question automatic generation device that automatically generates questions about the contents of the read text.
The text data acquisition unit that acquires the text data of the text for which the question is generated, and
A logical structure extraction unit that extracts technical terms contained in the acquired text data and extracts the logical structure between multiple technical terms.
It is equipped with a question generation unit that generates questions related to technical terms using the above logical structure.
The logical structure extraction unit is configured to classify the relationship of logical structures between a plurality of technical terms by a logical structure classification including definitions, confrontations, inclusions, incidental properties, and the like.
The question generation unit is configured to generate questions based on an algorithm corresponding to the logical structure classification.
An automatic question generator that features this .

Images