Test question generation method and device
By determining the knowledge points and entity information of the natural language processing technology of the test questions, generating the second test questions of the same type but different descriptions, the problem of inability to fully test students' understanding and prevent cheating in the existing technology is solved, and a more fair and fairer situation is achieved exam.
Patent Information
- Application Number
- CN202110185141.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-02-10
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2041-02-10
AI Technical Summary
The existing technology cannot fully test students' understanding and mastery of knowledge points, and cannot effectively prevent students from cheating in the exam.
By performing natural language processing on the question information of the test questions, the knowledge points tested are determined, and the question stem information is processed to extract physical information. Based on this information, a second question with the same type as the original question but a different description is generated to ensure that the knowledge points are not Change and increase the difficulty of plagiarism.
Effectively test students' understanding and mastery of knowledge points, prevent cheating, and ensure the fairness and impartiality of test scores.
Smart Images

Figure CN112800182B_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of information processing technology, and in particular to a test question generating method and device. Background Art
[0002] In the prior art, in order to prevent students from cheating in exams, different test papers are generally formed by randomly shuffling the test question numbers (or the order of candidate answers). Although the existing anti-cheating methods can play a certain role, the description of the question stem generally remains unchanged, so that students can still plagiarize based on the question and candidate answers. In addition, in the prior art, the description of the question stem is single, which cannot fully test students' understanding and mastery of knowledge points. Summary of the invention
[0003] The disclosed embodiments provide a test question generating method and device, which can solve the problems in the prior art that students' understanding and mastery of knowledge points cannot be fully tested and students' cheating cannot be effectively prevented.
[0004] According to one of the solutions of the present disclosure, a method for generating test questions is provided, comprising:
[0005] Performing first natural language processing on the question information of the first test question to determine the knowledge points tested in the first test question;
[0006] Performing a second natural language processing on the stem information of the first test question to obtain entity information of the first test question; wherein the entity information is a core word contained in the first test question;
[0007] At least one second test question is generated based on the knowledge point and the entity information, wherein the second test question is of the same question type as the first test question.
[0008] In some embodiments, performing first natural language processing on the question information of the first test question to determine the knowledge points tested in the first test question includes:
[0009] Determine keywords of the subject information;
[0010] Compare the keyword with knowledge points of a preset category to determine the semantic similarity between the keyword and the knowledge points of the preset category;
[0011] When the semantic similarity between the keyword and the knowledge point of the preset category is greater than a preset semantic threshold, the knowledge point of the preset category with the greatest semantic similarity is determined as the knowledge point tested in the first test question.
[0012] In some embodiments, determining keywords of the topic information includes:
[0013] Performing word segmentation processing on the title information of the first test question to determine candidate keywords;
[0014] Determine synonyms of the candidate keywords, and use the candidate keywords and the synonyms as keywords of the topic information.
[0015] In some embodiments, determining keywords of the topic information includes:
[0016] Determining the question type of the first test question;
[0017] Based on the question type of the first test question, determining the information category included in the question information of the first test question;
[0018] The keywords of the subject information are determined according to the information category.
[0019] In some embodiments, when the first test question is a multiple-choice question, the question information of the first test question includes question stem information and candidate answer information, and determining keywords of the question information according to the information category includes:
[0020] respectively determining keywords in the stem information and candidate answer information of the first test question, and determining keywords in the stem information that match the keywords in the candidate answer information as keywords in the first test question; or,
[0021] When the first test question is a composition question or an analysis question, the question information of the first test question includes question stem information and question information, and the determining of keywords of the question information according to the information category includes:
[0022] Keywords in the stem information and question information of the first test question are determined respectively, and keywords in the stem information that match the keywords in the question information are determined as keywords of the first test question.
[0023] In some embodiments, generating at least one second test question based on the knowledge point and the entity information includes:
[0024] Determine the associated words corresponding to the core words based on the text features of the core words and the syntactic features of the question stem information;
[0025] Generate at least one second test question stem information that conforms to the grammatical structure based on the knowledge points, core words and associated words.
[0026] In some embodiments, determining the associated words corresponding to the core words based on the text features of the core words and the syntactic features of the question stem information includes:
[0027] Determine the semantics expressed by the core word in the question stem information based on the text features of the core word and the syntactic features of the question stem information;
[0028] Based on the semantics expressed by the core word in the question stem information, the core word is semantically expanded to obtain a first associated word; and / or
[0029] Based on the text features of the core word and the syntactic features of the question stem information, a context-related word corresponding to the core word in the question stem information is determined, and the context-related word is semantically expanded to obtain a second related word.
[0030] In some embodiments, the text features of the core word include at least one of the category of the core word, the part of speech, the position of the core word in the question stem information, and the dependency relationship with other core words.
[0031] In some embodiments, the method further comprises:
[0032] Determining the difficulty coefficient of the first test question;
[0033] At least one second test question is generated based on the knowledge point, the entity information and the difficulty coefficient.
[0034] According to one of the solutions of the present disclosure, there is also provided a test question generating device, comprising:
[0035] A determination module, configured to perform a first natural language processing on the title information of the first test question to determine the knowledge points tested in the first test question;
[0036] an acquisition module, configured to perform a second natural language processing on the stem information of the first test question to acquire entity information of the first test question; wherein the entity information is a core word contained in the first test question;
[0037] A generation module is configured to generate at least one second test question based on the knowledge point and the entity information, wherein the second test question is of the same question type as the first test question.
[0038] According to one of the solutions of the present disclosure, there is also provided an electronic device, including a processor and a memory, wherein the memory is used to store computer-executable instructions, and the processor implements the above-mentioned test question generating method when executing the computer-executable instructions.
[0039] According to one of the solutions of the present disclosure, a computer-readable storage medium is provided, on which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, the above-mentioned test question generating method is implemented.
[0040] The test question generation method and device provided by various embodiments of the present disclosure use natural language processing technology to determine the knowledge points tested by the first test question from the title information of the first test question to be generated, and extract the entity information in the first test question from the stem information of the first test question, and then use the above knowledge points and entity information to regenerate at least one second test question, which can generate different stem descriptions while ensuring that the tested knowledge points remain unchanged, and fully test the students' understanding and mastery of the knowledge points; at the same time, different test questions use different stem descriptions, which can effectively create barriers for plagiarism between students, more effectively prevent the occurrence of cheating in the exam, and thus maximize the fairness and justice of the exam results. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Figure 1 A flow chart showing a method for generating test questions according to an embodiment of the present disclosure;
[0042] Figure 2 An example diagram showing the first test question when generating test questions in an embodiment of the present disclosure;
[0043] Figure 3 An example diagram showing a second test question generated in an embodiment of the present disclosure;
[0044] Figure 4 Another flow chart showing the test question generating method according to an embodiment of the present disclosure;
[0045] Figure 5 Another flow chart showing the test question generating method according to the embodiment of the present disclosure;
[0046] Figure 6 Another flow chart showing the test question generating method according to the embodiment of the present disclosure;
[0047] Figure 7 A schematic diagram showing the structure of a test question generating device according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION
[0048] Various aspects and features of the present disclosure are described herein with reference to the drawings.
[0049] It should be understood that various modifications may be made to the embodiments of the present application. Therefore, the above description should not be considered as limiting, but only as an example of an embodiment. Other modifications within the scope and spirit of the present disclosure will occur to those skilled in the art.
[0050] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the present disclosure and, together with the general description of the present disclosure given above and the detailed description of the embodiments given below, serve to explain the principles of the present disclosure.
[0051] These and other characteristics of the present disclosure will become apparent from the following description of a preferred form of embodiment given as a non-limiting example, with reference to the attached drawings.
[0052] It should also be understood that although the disclosure has been described with reference to certain specific examples, those skilled in the art will be able to readily implement many other equivalent forms of the disclosure.
[0053] The above and other aspects, features and advantages of the present disclosure will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings.
[0054] Specific embodiments of the present disclosure are described hereinafter with reference to the accompanying drawings; however, it should be understood that the embodiments claimed are merely examples of the present disclosure, which may be implemented in a variety of ways. Well-known and / or repeated functions and structures are not described in detail to avoid obscuring the present disclosure with unnecessary or redundant details. Therefore, the specific structural and functional details claimed herein are not intended to be limiting, but merely serve as a basis and representative basis for the claims to teach those skilled in the art to use the present disclosure in a variety of ways with substantially any suitable detailed structure.
[0055] Figure 1 FIG. 1 is a flow chart showing a method for generating test questions according to an embodiment of the present disclosure. Figure 1 As shown, the present disclosure provides a test question generating method, comprising:
[0056] S1: Performing a first natural language processing on the question information of the first test question to determine the knowledge points tested in the first test question.
[0057] The first test question is the first test question, and the question information is all the information contained in the first test question. The first test questions of different question types have different question information. Figure 2 As shown, the first test question is a multiple-choice question, and the question information includes question stem information and candidate answer information.
[0058] The first natural language processing is text analysis based on text recognition, text classification, etc., to accurately determine the knowledge points tested by the first test question. Figure 2 As shown, "relational database management system" is the knowledge point determined to be tested in the first test question. This knowledge point is clearly indicated in the question stem information of the first test question and can be directly determined through text recognition.
[0059] In other embodiments, the knowledge point is not clearly indicated in the title information of the first test question and needs to be obtained through reasoning and summarization. Therefore, it needs to be determined through a combination of methods such as text recognition and text classification.
[0060] S2: Perform a second natural language processing on the stem information of the first test question to obtain entity information of the first test question; wherein the entity information is a core word contained in the first test question.
[0061] Specifically, core words include keywords, logical relationship words, demonstrative words, directional words, and other words that play a core role in the sentence structure of the question stem information. Figure 2 As shown, "relational database system", "management" and "relationship" in the question information can all be core words.
[0062] Since the entity information is the specific content contained in the first test question, the second natural language processing is mainly a text analysis based on text recognition to identify and extract core words from the first test question.
[0063] S3: Generate at least one second test question based on the knowledge point and the entity information, wherein the second test question is of the same type as the first test question.
[0064] After obtaining the knowledge points and entity information, the two can be combined to further perform text analysis and processing to generate at least one second test question of the same type as the first test question. Figure 2 The knowledge points and entity information determined by the first test question shown in the figure are generated as follows Figure 3 The second test question shown, and the first test question and the generated second test question are both multiple-choice questions.
[0065] Among them, generating at least one second test question refers to generating the stem information of the second test question. For example, when the first test question is a fill-in-the-blank question or a true-or-false question, the question information only includes the stem information, and the generated second test question is the stem information of the second test question. When the first test question is a multiple-choice question, a question-and-answer question, an analysis question, etc., the question information includes not only the stem information but also the candidate answers or question information. Since the entity information is obtained from the stem information of the first test question, the stem information of the second test question is generated based on the knowledge points and the entity information.
[0066] The test question generation method provided by the embodiment of the present disclosure utilizes natural language processing technology to determine the knowledge points tested by the first test question from the title information of the first test question to be generated, and extracts the entity information in the first test question from the stem information of the first test question, and then regenerates at least one second test question using the above knowledge points and entity information. It can generate different stem descriptions while ensuring that the tested knowledge points remain unchanged, and fully test the students' understanding and mastery of the knowledge points. At the same time, different test questions use different stem descriptions, which can effectively create barriers for plagiarism between students and more effectively prevent the occurrence of cheating in the exam, thereby ensuring the fairness and justice of the exam results to the greatest extent.
[0067] In some embodiments, Figure 4 As shown, in step S1, the first natural language processing is performed on the question information of the first test question to determine the knowledge points tested in the first test question, including:
[0068] S11: Determine keywords of the topic information;
[0069] S12: Compare the keyword with knowledge points of a preset category to determine the semantic similarity between the keyword and the knowledge points of the preset category;
[0070] S13: When the semantic similarity between the keyword and the knowledge point of the preset category is greater than a preset semantic threshold, the knowledge point of the preset category with the greatest semantic similarity is determined as the knowledge point tested in the first test question.
[0071] Since the knowledge point tested by the first test question is the key point of the entire test question, it is necessary to first determine the keywords in the title information of the first test question. The determination of keywords can be performed using text recognition methods. For example, the title information can be first segmented or segmented, and then the keywords in the title information can be determined from different segmentations or segments. The keyword can be one or more. For example Figure 2 In the test, “relational database management system” is the keyword in the question information of the first question.
[0072] After the keyword is determined, the keyword can be compared with knowledge points of preset categories in a preset knowledge point library to calculate the semantic similarity between the two, wherein a certain number of accurate knowledge points of known categories are stored in the knowledge point library.
[0073] The knowledge points of the preset category can be the knowledge points with the same semantic environment as the keywords. For example, the subject, chapter of the textbook, etc. to which "relational database management system" belongs can be determined, and the subject, chapter of the textbook, etc. to which "relational database management system" belongs can be used as the semantic environment of the keyword. Then, the knowledge points in the above subjects, chapters, etc. are extracted from the preset knowledge point library as the knowledge points of the preset category to be compared with the keywords. When the semantic similarity between the keyword and the knowledge point of the preset category is greater than the preset semantic threshold, the knowledge point of the preset category that meets the preset threshold can be determined as the candidate knowledge point of the first test question. After the candidate knowledge points are determined, the knowledge point of the preset category with the greatest semantic similarity is determined as the knowledge point tested in the final first test question.
[0074] When there are multiple knowledge points of the preset category that meet the above preset threshold, some of the knowledge points are non-key knowledge points, so these non-key knowledge points can be eliminated to ensure that the final knowledge points screened out are the knowledge points tested in the first test question. For example, when the knowledge point is "management of relational database management systems", the candidate knowledge points determined based on the keyword "relational database management system" may be "management of relational database management systems", "classification of relational database management systems" and other subdivided knowledge points. At this time, "management of relational database management systems" with the largest similarity threshold is determined as the knowledge point tested in the first test question, so that more accurate knowledge points can be obtained.
[0075] The specific method for calculating semantic similarity can be a statistical method based on word co-occurrence, mainly by counting the word frequency in the sentence, such as the TF-IDF algorithm, etc.; it can also be a corpus training feature extraction method based on a neural network, etc. The specific calculation method is not specifically limited in this disclosure.
[0076] In some embodiments, in step S11, determining keywords of the topic information includes:
[0077] S111: performing word segmentation processing on the title information of the first test question to determine candidate keywords;
[0078] S112: Determine synonyms of the candidate keywords, and use the candidate keywords and the synonyms as keywords of the topic information.
[0079] The word segmentation process includes vocabulary splitting, punctuation filtering, etc. A plurality of word segmentation fragments are obtained through the word segmentation process, and the word segmentation fragments can be a single character, a single word, or a compound word composed of multiple words.
[0080] After the word segmentation process, the probability of each word segmentation fragment appearing in the preset keyword library is determined according to the word frequency statistical method, etc. If the probability of the word segmentation fragment appearing in the preset keyword library meets the preset threshold, the word segmentation fragment is selected as a candidate keyword; if the probability of the word segmentation fragment appearing in the preset keyword library does not meet the preset threshold, it is determined that the word segmentation fragment is not a candidate keyword. In a specific implementation, the candidate keyword can also be determined by determining whether the above-mentioned word segmentation fragment exists in the preset keyword library.
[0081] After the candidate keywords are determined, synonyms with the candidate keywords are extracted from the preset keyword library, and the candidate keywords and synonyms are used as keywords for the question information of the first test question, so that more comprehensive and accurate keywords can be obtained. Figure 2In the text, RDBMS is the abbreviation of relational database management system. After determining that the keyword is "relational database management system", it is necessary to select "RDBMS" or its full English name, which is synonymous with "relational database management system", from the preset keyword library or the existing dictionary. If only the keywords in the question information are used, the comparison between "RDBMS" and the preset knowledge points may be missed, resulting in reduced accuracy in determining the knowledge points.
[0082] In some other embodiments, in step S11, determining the keywords of the topic information includes:
[0083] S113: Determine the type of the first test question;
[0084] S114: Determine, based on the question type of the first test question, the information category included in the question information of the first test question;
[0085] S115: Determine keywords of the topic information according to the information category.
[0086] Since the information categories contained in the title information of different question types are different, for example, when the first test question is a multiple-choice question, the title information includes the stem information and the candidate answer information. If the keywords are obtained only through the above steps S111 and S112, then some interference information that appears more frequently in the candidate answers may also be keywords. For example, the same keyword appears in the interference options of the single-choice question. In addition, for the first test questions of different question types, although the keywords of the title information can be determined from the stem information, determining the keywords only from the stem information may miss important keywords that appear in the correct options, reducing the accuracy of keyword extraction of the title information. Therefore, in order to ensure the accuracy of keyword determination of the title information, in steps S113-115, the information category contained in the title information of the first test question is determined according to the question type of the first test question, and then the keywords of the title information are determined.
[0087] In a specific embodiment, when the first test question is a multiple-choice question, the question information of the first test question includes question stem information and candidate answer information, and determining keywords of the question information according to the information category includes:
[0088] Keywords in the question stem information and candidate answer information of the first test question are determined respectively, and keywords in the question stem information that match the keywords in the candidate answer information are determined as keywords of the first test question.
[0089] That is, when the first test question is a multiple-choice question, the information category of the question information includes stem information and candidate answer information. The stem information and the candidate answer information are respectively segmented to obtain the keywords in the stem information and the candidate answer information, and then the determined first keyword in the stem information and the second keyword in the candidate answer information are matched. If the first keyword and the second keyword both contain the same keyword, it indicates that the keyword is both a keyword in the stem information and a keyword in the candidate answer. In this way, the influence of the keywords in the interference options on the keyword determination of the question information can be eliminated, thereby ensuring the accuracy of the keyword determination of the question information.
[0090] In another specific embodiment, when the first test question is a composition question or an analysis question, the question information of the first test question includes question stem information and question information, and determining keywords of the question information according to the information category includes:
[0091] Keywords in the stem information and question information of the first test question are determined respectively, and keywords in the stem information that match the keywords in the question information are determined as keywords of the first test question.
[0092] The determination of keywords for the question information of essay questions or analysis questions is similar to the determination of keywords for the question information of the multiple-choice questions mentioned above, and will not be repeated here.
[0093] The above steps S111-S112 and steps S113-S115 may be performed independently or in combination. For example, in step S115, the above steps S111-S112 may be used to determine keywords of the topic information.
[0094] In some embodiments, Figure 5 As shown, in step S3, at least one second test question is generated based on the knowledge point and the entity information, including:
[0095] S31: determining a related word corresponding to the core word based on the text feature of the core word and the syntactic feature of the question stem information;
[0096] S32: Generate question stem information of at least one second test question that conforms to the grammatical structure based on the knowledge points, core words and associated words.
[0097] Among them, the text features of the core words include at least one of the category, part of speech, position in the question stem information, and dependency relationship with other core words. The category of the core words can be proper nouns such as names of people, institutions, places, actions, and meaningful time, etc.; the part of speech of the core words can be nouns, verbs, adjectives, adverbs, etc.; the position of the core words in the question stem information can be the beginning of the word, the end of the word, etc., and the word order of multiple core words can be determined based on the position of the core words; the dependency relationship with other core words includes the relationship between the front and back, the subordinate relationship, the modification relationship, or whether it must be used together with other core words, etc. The syntactic features of the question stem information include the sentence structure and the logical semantic relationship between each participle. The sentence structure can include the subject-predicate structure, the subject-predicate-object structure, etc. of the entire question stem information or some phrases in the question stem information.
[0098] For example Figure 2 In the sentence “the relationship managed by the relational database management system is”, after word segmentation, it can be divided into “relational database management system”, “management”, “of”, “relationship”, and “is”. Among them, the core words “relational database management system”, “management”, and “relationship” are the subject, predicate, and object respectively, and “management” and “relationship” need the help of the auxiliary word “of” to form a phrase.
[0099] By analyzing the text features of the core words and the syntactic features of the question stem information, we can get the associated words corresponding to the core words, so as to get more words related to the core words and re-compose the sentences. After getting the associated words corresponding to the core words, we can screen and combine the core words and associated words, and get a second test question with more quantity and more accurate semantic expression when the knowledge points are the same.
[0100] It should be noted that the above-mentioned grammatical structure refers to a complete sentence expression that conforms to conventional general grammar.
[0101] In some embodiments, in step S31, based on the text features of the core word and the syntactic features of the question stem information, determining the associated words corresponding to the core word includes:
[0102] Step S311: determining the semantics expressed by the core word in the question stem information based on the text features of the core word and the syntactic features of the question stem information;
[0103] Step S312: Based on the semantics expressed by the core word in the question stem information, the core word is semantically expanded to obtain a first associated word; and / or
[0104] Step S313: Based on the text features of the core word and the syntactic features of the question stem information, determine the context-related words corresponding to the core word in the question stem information, and perform semantic expansion on the context-related words to obtain second related words.
[0105] Among them, step S311 and step S312 are the steps of expanding the extended semantic words of the core words. Step S311 is used to accurately determine the semantics expressed by the core words in the question stem information to prevent semantic misunderstanding. In particular, for the case where the same core word expresses different semantics in different semantic environments, semantic recognition can be accurately performed. For example, the relationship in "relational database management system" is used as a modifier to modify "database management system" to determine the type of "database management system"; and the relationship in "management relationship" is a noun, therefore, it is necessary to accurately identify the semantics expressed by each core word in order to obtain a more accurate first associated word. After accurately determining the semantics expressed by the core word, the core word is expanded to obtain more core words for generating the second test question, so as to generate more second test questions.
[0106] For example Figure 2 In the question stem information "The relationship managed by a relational database management system is", the core words "relational database management system", "management", and "relationship" are noun, verb, and noun respectively, and the syntactic feature of the question stem information is a standard subject-predicate-object structure. From the above, we can know that "relational database management system" is a proper noun that can be used alone. Therefore, we can directly expand its synonyms to get its English abbreviation "RDBMS", or perform semantic reasoning on "relational database management system" to get that "relational database management system" manages relational data, thus obtaining the following Figure 3 For another example, the "management relationship" in the above question stem information is a subject-object relationship. At the same time, based on the position of "management" and "relationship" or the dependency relationship between the two, it is determined that the data type should be filled in after "is". Therefore, based on the two core words "management" and "relationship" and the syntactic features of the question stem information, the following can be obtained: Figure 3The extended semantic word "storage" is shown. Then, based on common grammatical expressions or idiomatic expressions, the selected knowledge points, core words and the first associated words are recombined to obtain the second test question. For example, "storage" is a verb, and the common expressions following it are generally "stored in" or "stored as", and "storage is" is usually not used. "Stored in" is generally followed by a noun indicating a place, and a data type can be added after "stored as". Therefore, "stored as" is determined to be used for the second test question; the "management" used in the first test question can continue to be used, and the core word "relational database management system" can be used multiple times. When used again, it can be abbreviated as "system". Since in this implementation, the knowledge point and one of the core words are the same, the two can be merged, and finally the stem information of the first test question "The relationship managed by the relational database management system is" is regenerated as "In RDBMS, the relationship managed by the system is stored as".
[0107] In some embodiments, when the topic information also includes information of other information categories, step S31 further includes:
[0108] Based on the text features of the core word, the syntactic features of the question stem information, and other information in the question information except the question stem information, an associated word corresponding to the core word is determined.
[0109] Specifically, based on Figure 2 From the candidate answers for the multiple-choice questions shown, we can determine that the file type should be filled in after the question information "The relationship managed by the relational database management system is"; therefore, after determining the extended semantic word "storage", we can directly determine that "store as" is used in the second question, which can improve the efficiency of question generation.
[0110] Before step S311, the method further includes:
[0111] S3101: Determine whether the semantics expressed by the core word is single;
[0112] S3102: If yes, then the core word is not semantically expanded; if no, the core word expressing non-single semantics is extracted and step S311 is executed.
[0113] That is, when the core word in the question stem information expresses a single semantic meaning, the core word can be determined as an independent word. No matter what semantic environment it is in, the semantic meaning it expresses is unique and there is no need to expand it, which can improve the efficiency of text analysis and processing.
[0114] After obtaining the extended semantic words of the core words, the above-mentioned knowledge points, core words and the extended semantic words may be input into a test question generation model including a preset grammatical structure to generate a second test question.
[0115] It should be noted that, in the above-mentioned test question generation process, the core words and their extended semantic words can be replaced with each other, and can be used multiple times in the process of arranging and combining sentences according to actual needs. The specific usage is not specifically limited in this embodiment.
[0116] In a specific implementation, different core words and their extended semantic words are used as much as possible in different test questions (including the first test question and the generated second test question), so as to enrich the expression of the test questions and effectively prevent cheating.
[0117] Step S313 is a step for determining the contextual associated words of the core word and expanding the contextual associated words. If the second test question is generated based only on knowledge points, core words and extended semantic words of the core words, since core words are usually important words such as nouns and verbs, the generated second test question will be stiff or not smooth enough. Therefore, in order to ensure that the generated second test question is more natural and fluent, the contextual associated words of the core word in the question stem information are obtained through step S313. The contextual associated words can be core words or non-core words, for example, non-core words such as pronouns and conjunctions.
[0118] After determining the context-related words, classify them, divide the core words into one category, and divide the non-core words into another category. The processing of the core words in the context-related words can be expanded as described in steps S311 and S312, and no further description is given here. For non-core words, expansion can be performed to obtain extended semantic words of the non-core words, and then the context-related words and the extended semantic words of the context-related words are used as second associated words, combined with knowledge points, entity information and / or first associated words, to obtain a second test question with accurate semantic expression, and more natural and fluent; at the same time, more second test questions can be generated by obtaining context-related words and their extended semantic words, and the semantic expression is richer.
[0119] In some embodiments, Figure 6 As shown, the method also includes:
[0120] S4: Determine the difficulty coefficient of the first test question;
[0121] S33: Generate at least one second test question based on the knowledge point, the entity information and the difficulty coefficient.
[0122] Students have different understanding abilities for different question stem information. When the question stem information is expressed fluently, students can easily understand the semantics expressed by the question stem information and can answer quickly. However, when the question stem information is difficult to understand, students need to analyze it to understand the semantics expressed by the question stem information, which not only affects the answering time but also easily leads to misunderstanding of the question stem. In this way, it will be difficult to accurately test students' understanding and mastery of knowledge points, and it will also affect the fairness of the exam. Therefore, by determining the difficulty coefficient of the first test question, and then generating a second test question with the same or similar difficulty coefficient as the first test question based on the knowledge point and entity information, it can not only accurately test students' understanding and mastery of knowledge points, but also ensure the fairness of the exam.
[0123] In a specific implementation, a sufficient number of first test questions, second test questions, and answers to the first test questions and the second test questions can be used for pre-training to obtain a difficulty coefficient model. After obtaining the difficulty coefficient of the first test question through the difficulty coefficient model, the knowledge point, entity information, and difficulty coefficient are input into a preset text generation model containing difficulty coefficients to obtain at least one second test question with the same difficulty coefficient as the first test question.
[0124] The above-mentioned test question generation method mainly includes the steps for generating the stem information of the second test question. When the question information of the first test question contains information of other information categories besides the stem information, information of other information categories of the second test question can also be generated at the same time. For example, for multiple-choice questions, candidate answers can be generated while generating the stem information. The specific method for generating the candidate answers is similar to the above-mentioned method for generating the stem information, which will not be repeated here.
[0125] Figure 7 FIG. 2 shows a schematic diagram of the structure of the test question generating device according to an embodiment of the present disclosure. Figure 7 As shown, the embodiment of the present disclosure provides a test question generating device, comprising:
[0126] A determination module 10 is configured to perform a first natural language processing on the question information of the first test question to determine the knowledge points tested in the first test question;
[0127] The acquisition module 20 is configured to perform a second natural language processing on the stem information of the first test question to acquire entity information of the first test question; wherein the entity information is a core word contained in the first test question;
[0128] The generating module 30 is configured to generate at least one second test question based on the knowledge point and the entity information, wherein the second test question is of the same type as the first test question.
[0129] The test question generating device provided by the embodiment of the present disclosure uses natural language processing technology to determine the knowledge points tested by the first test question from the title information of the first test question to be generated, and extracts the entity information in the first test question from the stem information of the first test question, and then uses the above knowledge points and entity information to regenerate at least one second test question. It can generate different stem descriptions while ensuring that the tested knowledge points remain unchanged, so as to fully test the students' understanding and mastery of the knowledge points; at the same time, different test questions use different stem descriptions, which can effectively create barriers for plagiarism between students and more effectively prevent the occurrence of cheating in the examination, thereby ensuring the fairness and justice of the examination results to the greatest extent.
[0130] The test question generating device provided in the embodiment of the present disclosure corresponds to the test question generating method in the above-mentioned embodiment. Based on the above-mentioned test question generating method, those skilled in the art can understand the specific implementation manner of the test question generating device in the embodiment of the present disclosure and its various variations. Any optional items in the test question generating method embodiment are also applicable to the test question generating device, which will not be described in detail here.
[0131] An embodiment of the present disclosure further provides an electronic device, comprising: a processor and a memory, wherein the memory is used to store computer-executable instructions, and the processor implements the above-mentioned test question generating method when executing the computer-executable instructions.
[0132] The above-mentioned processor can be a general-purpose processor, including a central processing unit CPU, a network processor (NP), etc.; it can also be a digital signal processor DSP, an application-specific integrated circuit ASIC, a field programmable gate array FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.
[0133] The memory may include a random access memory (RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.
[0134] The embodiment of the present disclosure also provides a computer-readable storage medium having computer-executable instructions stored thereon. When the computer-executable instructions are executed by a processor, the above-mentioned test question generating method is implemented.
[0135] The above embodiments are only exemplary embodiments of the present disclosure and are not intended to limit the present disclosure. The protection scope of the present disclosure is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present disclosure within the essence and protection scope of the present disclosure, and such modifications or equivalent substitutions shall also be deemed to fall within the protection scope of the present disclosure.
Claims
1. A test question generation method, comprising: Performing first natural language processing on the question information of the first test question to determine the knowledge points tested in the first test question and the difficulty coefficient of the first test question; Performing a second natural language processing on the stem information of the first test question to obtain entity information of the first test question; wherein the entity information is a core word contained in the first test question; Generate at least one second test question based on the knowledge point, the entity information and the difficulty coefficient, including: determining the associated words corresponding to the core words based on the text features of the core words and the syntactic features of the stem information; generating stem information of at least one second test question that conforms to the grammatical structure based on the knowledge point, the core word and the associated words; inputting the knowledge point, the entity information and the difficulty coefficient into a preset text generation model including the difficulty coefficient to obtain at least one second test question with the same difficulty coefficient as the first test question; wherein, the second test question has the same question type as the first test question.
2. The method according to claim 1, wherein: Performing a first natural language processing on the question information of the first test question to determine the knowledge points tested in the first test question includes: Determine keywords of the subject information; Compare the keyword with knowledge points of a preset category to determine the semantic similarity between the keyword and the knowledge points of the preset category; When the semantic similarity between the keyword and the knowledge point of the preset category is greater than a preset semantic threshold, the knowledge point of the preset category with the greatest semantic similarity is determined as the knowledge point tested in the first test question.
3. The method according to claim 2, wherein: Determine the keywords of the subject information, including: Performing word segmentation processing on the title information of the first test question to determine candidate keywords; Determine synonyms of the candidate keywords, and use the candidate keywords and the synonyms as keywords of the topic information.
4. The method according to claim 2, wherein: Determine the keywords of the subject information, including: Determining the question type of the first test question; Based on the question type of the first test question, determining the information category included in the question information of the first test question; The keywords of the subject information are determined according to the information category.
5. The method according to claim 4, wherein: When the first test question is a multiple-choice question, the question information of the first test question includes question stem information and candidate answer information, and the determining of keywords of the question information according to the information category includes: respectively determining keywords in the stem information and candidate answer information of the first test question, and determining keywords in the stem information that match the keywords in the candidate answer information as keywords in the first test question; or, When the first test question is a composition question or an analysis question, the question information of the first test question includes question stem information and question information, and the determining of keywords of the question information according to the information category includes: Keywords in the stem information and question information of the first test question are determined respectively, and keywords in the stem information that match the keywords in the question information are determined as keywords of the first test question.
6. The method according to claim 1, wherein: Determining the associated words corresponding to the core words based on the text features of the core words and the syntactic features of the question stem information includes: Determine the semantics expressed by the core word in the question stem information based on the text features of the core word and the syntactic features of the question stem information; Based on the semantics expressed by the core word in the question stem information, the core word is semantically expanded to obtain a first associated word; and / or Based on the text features of the core word and the syntactic features of the question stem information, a context-related word corresponding to the core word in the question stem information is determined, and the context-related word is semantically expanded to obtain a second related word.
7. The method according to claim 1, wherein: The text features of the core word include at least one of the category of the core word, the part of speech, the position of the core word in the question stem information, and the dependency relationship with other core words.
8. A test question generating device, comprising: a determination module configured to perform a first natural language processing on the question information of the first test question to determine the knowledge points tested in the first test question and the difficulty coefficient of the first test question; an acquisition module, configured to perform a second natural language processing on the stem information of the first test question to acquire entity information of the first test question; wherein the entity information is a core word contained in the first test question; A generation module is configured to generate at least one second test question based on the knowledge point, the entity information and the difficulty coefficient, including: determining the associated words corresponding to the core words based on the text features of the core words and the syntactic features of the stem information; generating stem information of at least one second test question that conforms to the grammatical structure based on the knowledge point, the core word and the associated words; inputting the knowledge point, the entity information and the difficulty coefficient into a preset text generation model including the difficulty coefficient to obtain at least one second test question with the same difficulty coefficient as the first test question; wherein the second test question is of the same question type as the first test question.
