Homework grading method based on image recognition and large language model

Abstract

The present invention discloses a homework grading method based on image recognition and a large language model, relates to the field of teaching, solves the problem of insufficient practicality of the homework grading method, comprises step S1: obtaining a plurality of homework texts and a plurality of answer texts, and obtaining preliminary collection data of homework, step S2: dividing the plurality of homework texts into a first type of homework text and a second type of homework text, and respectively grading the first type of homework text and the second type of homework text to obtain homework grading data, step S3: performing answer accuracy analysis on homework submitted by target students to obtain a first homework quality assessment coefficient and a second homework quality assessment coefficient to obtain accuracy analysis data, and step S4: performing completion quality assessment on the homework submitted by the target students in the current batch by analyzing the accuracy analysis data. The present invention can expand the function of the homework grading method and further improve practicality.

Description

Technical Field

[0001] The invention belongs to the field of teaching and relates to image recognition technology, in particular to a homework marking method based on image recognition and a large language model. Background Art

[0002] The existing homework grading method has the following specific defects when performing homework grading:

[0003] 1. The existing homework grading method requires repeated text comparison between the standard answers and the same batch of homework completed by different students, which leads to low efficiency of homework grading;

[0004] 2. The existing homework grading method is limited to the grading of students' homework. It is unable to combine the historical grading records and the current batch grading records to evaluate the quality of the homework submitted by students, resulting in the problem of poor practicality of the homework grading method.

[0005] To this end, we propose a homework grading method based on image recognition and large language models. Summary of the invention

[0006] In view of the deficiencies in the prior art, the present invention aims to improve the practicability of the homework marking method.

[0007] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solution: a homework grading method based on image recognition and a large language model, comprising the following specific steps:

[0008] Step S1: obtaining a target task image and a task answer image, splitting the target task image and the task answer image into a plurality of task texts and a plurality of answer texts respectively, and obtaining preliminary task collection data;

[0009] Step S2: Divide multiple homework texts into first-type homework texts and second-type homework texts according to the preliminary collection data of homework, correct the first-type homework texts by using text comparison, create a homework correction large language model to correct the second-type homework texts, obtain the correctness of the answer corresponding to each homework text according to the correction result, and obtain homework correction data;

[0010] Step S3: Analyze the correctness of answers to the homework submitted by the target student based on the preliminary collection data and the homework correction data, and obtain the first homework quality evaluation coefficient and the second homework quality evaluation coefficient according to the analysis results to obtain correctness analysis data;

[0011] Step S4: Evaluate the completion quality of the current batch of homework submitted by the target students by analyzing the correctness analysis data.

[0012] Furthermore, step S1 also includes the following specific steps:

[0013] Step S11: selecting a student from among the students who submitted homework in the current batch as a target student, acquiring an image of the homework submitted by the target student, and obtaining a target homework image;

[0014] Step S12: acquiring an image of the standard answer corresponding to the homework submitted by the target student to obtain a homework answer image;

[0015] Step S13: using an image text recognition algorithm to perform text recognition on the homework answer image to obtain the homework answer text;

[0016] Step S14: using an image text recognition algorithm to perform text recognition on the target job image to obtain the target job text;

[0017] Step S15: split the target homework text into T1 homework text to Ta homework text according to the homework title sequence number in the target homework text;

[0018] Step S16: splitting the homework answers corresponding to the homework texts T1 to Ta in the homework answer text to obtain the answer texts T1 to Ta;

[0019] Step S17: Define T1 job text to Ta job text and T1 answer text to Ta answer text as preliminary job collection data.

[0020] Furthermore, step S2 further includes the following specific steps:

[0021] Step S21: obtaining preliminary collected data of the operation, and obtaining T1 operation text to Ta operation text and T1 answer text to Ta answer text respectively according to the preliminary collected data of the operation;

[0022] Step S22: marking the objective question texts in the homework text T1 to the homework text Ta as the first type of homework text, and marking the subjective question texts in the homework text T1 to the homework text Ta as the second type of homework text;

[0023] Step S23: obtaining the results of grading the first type of homework texts, and obtaining the correctness of the answers corresponding to each first type of homework text;

[0024] Step S24: obtaining the results of grading the second type of homework texts, and obtaining the correctness of the answers corresponding to every two first type of homework texts;

[0025] Step S25: Name the answer correctness corresponding to the homework text T1 to the homework text Ta as T1 answer correctness to Ta answer correctness, and obtain homework correction data.

[0026] Furthermore, step S23 further includes the following specific steps:

[0027] Step S231: If the Ti homework text is the first type of homework text, and the Ti answer text is a single field text, the Ti homework text and the Ti answer text are compared. If the text comparison is consistent, the correctness of the answer corresponding to the Ti homework text is 100%;

[0028] Step S232: If the Ti homework text is the first type of homework text, but the Ti answer text is not a single field text, count the number of character segments contained in the Ti answer text to obtain the answer field quantity value, perform field comparison on the Ti homework text and the Ti answer text, mark the fields that match the comparison as correct answer fields, count the number of correct answer fields to obtain the correct field quantity value, calculate the ratio of the correct answer field to the answer field quantity value, obtain the field correct ratio Zbi, and then judge that the answer accuracy corresponding to the Ti homework text is Zbi%.

[0029] Furthermore, step S24 further includes the following specific steps:

[0030] Step S241: when the Tj assignment text is the second type assignment text, the Tj answer text is divided into a plurality of answer key fields, and the divided plurality of answer key fields are named G1 answer field to Gb answer field;

[0031] Step S242: setting the answer accuracy contribution ratios for the G1 answer field to the Gb answer field respectively, and obtaining the answer accuracy contribution ratios for the G1 answer field to the Gb answer field;

[0032] Step S243: Use the G1 answer field to the Gb answer field to train the existing large language model to obtain the Tj homework correction large language model;

[0033] Step S244: Use the Tj homework correction large language model to correct the Tj homework text and obtain the correctness of the answer corresponding to the Tj homework text.

[0034] Furthermore, step S243 further includes the following specific steps:

[0035] Step S2431: Obtain the student homework completion texts corresponding to several Tj answer texts, obtain multiple homework completion texts, and obtain the answer correctness corresponding to each homework completion text respectively;

[0036] Step S2432: selecting a sample homework completion text from multiple homework completion texts, and obtaining the correctness of the answer corresponding to the sample homework completion text;

[0037] Step S2433: Obtain the answer correctness corresponding to each piece of completed assignment text respectively;

[0038] Step S2434: Use the answer correctness of each piece of completed assignment text to label its corresponding completed assignment text, obtaining completed assignment text labeling data;

[0039] Step S2435: Divide the text labeling data into a completed assignment text training set and a completed assignment text test set according to the image training and testing ratio;

[0040] Step S2436: Obtain a large language model through an existing artificial intelligence platform, and use the completed assignment text training set and the completed assignment text test set to train and test the large language model, obtaining a large language model for Tj assignment marking.

[0041] Furthermore, in step S2432, the following specific steps are further included:

[0042] Step S24321: Use a text comparison algorithm to perform character comparison between the sample completed assignment text and the answer fields from G1 to Gb. If the G1 answer field exists in the sample completed assignment text, assign the value 1 to the field existence degree corresponding to the G1 answer field; if the G1 answer field does not exist in the sample completed assignment text, assign the value 0 to the field existence degree corresponding to the G1 answer field, obtaining the G1 field existence degree;

[0043] Step S24322: Perform character comparison between the sample completed assignment text and the answer fields from G2 to Gb. If the G2 answer field exists in the sample completed assignment text, assign the value 1 to the field existence degree corresponding to the G2 answer field; if the G2 answer field does not exist in the sample completed assignment text, assign the value 1 to the field existence degree corresponding to the G2 answer field, obtaining the G2 field existence degree;

[0044] Step S24323: And so on, assign the field existence degrees corresponding to the answer fields from G3 to Gb respectively, obtaining the G3 field existence degree to the Gb field existence degree;

[0045] Step S24324: Calculate the answer correctness corresponding to the sample completed assignment text through the G1 answer correctness contribution ratio to the Gb answer correctness contribution ratio and the G1 field existence degree to the Gb field existence degree;

[0046] Calculate the answer correctness corresponding to the sample completed assignment text, and the specific formula is as follows:

[0047] ;

[0048] Among them, Zqd is the answer accuracy corresponding to the sample homework completion text, Gzqi is the contribution ratio of Gi answer accuracy, Gxdi is the existence of Gi field, and b is the quantity value corresponding to the key field of the answer.

[0049] Furthermore, step S3 further includes the following specific steps:

[0050] Step S31: Obtain preliminary collected data of the operation, obtain preliminary collected data of the operation, and obtain T1 operation text to Ta operation text respectively according to the preliminary collected data of the operation;

[0051] Step S32: setting the teaching knowledge points P1 to Pc from the homework text T1 to the homework text Ta according to the teaching knowledge points corresponding to the homework content;

[0052] Step S33: obtaining homework correction data, and obtaining T1 answer accuracy to Ta answer accuracy respectively according to the homework correction data;

[0053] Step S34: Calculate the average of the answer correctness corresponding to the teaching knowledge point P1 to the teaching knowledge point Pc according to the answer correctness of T1 to Ta, and obtain the answer correctness of the knowledge point P1 to the knowledge point Pc;

[0054] Step S35: Compare and analyze the correctness of the answers to the knowledge points P1 to Pc and the historical correctness of the knowledge points to obtain a first homework quality evaluation coefficient;

[0055] Step S36: Compare and analyze the correctness of the answers to the knowledge points P1 to Pc with the average correctness of the answers to the assignments submitted in the same batch to obtain a second assignment quality evaluation coefficient;

[0056] Step S37: defining the first operation quality assessment coefficient and the second operation quality assessment coefficient as correctness analysis data;

[0057] Step S36 also includes the following specific steps:

[0058] Step S361: in the current batch of submitted homework, the batch average answer accuracy of the homework text corresponding to the P1 teaching knowledge point is obtained, and the batch accuracy of the P1 knowledge point is obtained. The batch average answer accuracy of the homework text corresponding to the P2 teaching knowledge point is obtained, and the batch accuracy of the P2 knowledge point is obtained. By analogy, the batch average answer accuracy of the homework text corresponding to the Pc teaching knowledge point is obtained, and the batch accuracy of the Pc knowledge point is obtained.

[0059] Step S362: calculating the second job quality evaluation coefficient by converting the correctness of the answer to the knowledge point P1 to the correctness of the answer to the knowledge point Pc and the correctness of the batch of knowledge points P1 to the correctness of the batch of knowledge points Pc;

[0060] The quality evaluation coefficient of the second operation is calculated, and the specific formula is as follows:

[0061] ;

[0062] Among them, Zyz2 is the quality evaluation coefficient of the second homework, Zqdi is the correctness of the answer to the Pi knowledge point, Pczi is the batch correctness of the Pi knowledge point, and c is the number of teaching knowledge points corresponding to the T1 homework text to the Ta homework text.

[0063] Furthermore, step S35 further includes the following specific steps:

[0064] Step S351: obtaining the target student's historical homework modification records, selecting a number of historical batches of homework from the target student's historical homework modification records, and naming the obtained number of historical batches of homework as the first historical batch of homework to the dth historical batch of homework in the order of homework submission time;

[0065] Step S352: Obtain the correctness of the assignments corresponding to the teaching knowledge points P1 to Pc in the first historical batch of assignments, obtain the correctness of the assignments corresponding to the teaching knowledge points P1 to Pc in the second historical batch of assignments, obtain the correctness of the assignments corresponding to the teaching knowledge points P1 to Pc in the second historical batch of assignments, obtain the correctness of the assignments corresponding to the teaching knowledge points P1 to Pc in the second historical batch of assignments, and so on, obtain the correctness of the assignments corresponding to the teaching knowledge points P1 to Pc in the dth historical batch of assignments, and obtain the correctness of the assignments of the dth historical P1 to the dth historical Pc;

[0066] Step S353: Calculate the average of the correctness of the first historical P1 operation to the correctness of the dth historical P1 operation to obtain the correctness of the historical operation of the P1 knowledge point, calculate the average of the correctness of the first historical P2 operation to the correctness of the dth historical P2 operation to obtain the correctness of the historical operation of the P2 knowledge point, and so on, calculate the average of the correctness of the first historical Pc operation to the correctness of the dth historical Pc operation to obtain the correctness of the historical operation of the Pd knowledge point;

[0067] Step S354: calculating the first homework quality evaluation coefficient by combining the historical homework correctness of the knowledge point P1 to the historical homework correctness of the knowledge point Pd and the answer correctness of the knowledge point P1 to the answer correctness of the knowledge point Pc;

[0068] The quality evaluation coefficient of the first operation is calculated, and the specific formula is as follows:

[0069] ;

[0070] Among them, Zyz1 is the quality evaluation coefficient of the first homework, Zqdi is the correctness of the answer to the Pi knowledge point, Lszi is the correctness of the historical homework of the Pi knowledge point, and c is the number of teaching knowledge points corresponding to the homework text from T1 to Ta.

[0071] Furthermore, step S4 further includes the following specific steps:

[0072] Step S41: Acquire correctness analysis data, and acquire a first operation quality assessment coefficient and a second operation quality assessment coefficient according to the correctness analysis data;

[0073] Step S42: Calculating the first operation quality assessment coefficient and the second operation quality assessment coefficient to obtain an operation quality comprehensive assessment coefficient;

[0074] The comprehensive evaluation coefficient of operation quality is calculated, and the specific formula is as follows:

[0075] ;

[0076] Among them, Zpg is the comprehensive evaluation coefficient of operation quality, Zyz1 is the first operation quality evaluation coefficient, and Zyz2 is the second operation quality evaluation coefficient;

[0077] Step S43: obtaining a threshold value of a comprehensive evaluation coefficient of homework quality, performing a numerical comparison between the comprehensive evaluation coefficient of homework quality and the threshold value of the comprehensive evaluation coefficient of homework quality, and performing a quality qualification assessment on the target student according to the numerical comparison result;

[0078] Step S44: obtaining a first operation quality assessment coefficient threshold and a second operation quality assessment coefficient threshold, and calculating the first operation quality assessment coefficient threshold and the second operation quality assessment coefficient threshold to obtain an operation quality comprehensive assessment coefficient threshold;

[0079] When the comprehensive evaluation coefficient of homework quality is greater than or equal to the comprehensive evaluation coefficient threshold of homework quality, it is judged that the quality evaluation of the homework submitted by the target students in the current batch is qualified;

[0080] Get the comprehensive evaluation coefficient threshold of homework quality. When the comprehensive evaluation coefficient of homework quality is less than the comprehensive evaluation coefficient threshold, it is judged that the quality evaluation of the homework submitted by the target students in the current batch is unqualified.

[0081] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0082] 1. The present invention can effectively improve the efficiency of homework grading by creating a large language model for homework grading to uniformly grade different homeworks in the same batch.

[0083] 2. The present invention improves the functional diversity of the homework grading method by combining historical grading records and current batch grading records to evaluate the quality of the homework submitted by the students.

[0084] 3. The present invention evaluates student assignments by analyzing the first assignment quality evaluation coefficient and the second assignment quality evaluation coefficient, thereby improving the accuracy of the evaluation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0086] Figure 1 is the overall system block diagram of the present invention;

[0087] Figure 2 A schematic diagram of the correspondence between the homework text and the answer text of the present invention;

[0088] Figure 3 Schematic diagram of knowledge point matching in the present invention. DETAILED DESCRIPTION

[0089] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0090] Embodiment 1

[0091] See also Figure 1 The present invention provides a technical solution: a homework grading method based on image recognition and a large language model, comprising the following specific steps:

[0092] Step S1: obtaining a target task image and a task answer image, splitting the target task image and the task answer image into a plurality of task texts and a plurality of answer texts respectively, and obtaining preliminary task collection data;

[0093] Step S1 also includes the following specific steps:

[0094] A student is selected from the students who submitted homework in the current batch as a target student, and an image of the homework submitted by the target student is acquired to obtain a target homework image;

[0095] Acquire the image of the standard answer corresponding to the homework submitted by the target student to obtain the homework answer image;

[0096] It should be noted here that:

[0097] In this application, the answers to the homework submitted by the current batch of students involved here are all included in the standard answers, there are no open-ended homework questions with several different answers, and the homework submitted by the current batch of students all have clear corresponding teaching knowledge points;

[0098] Use the image text recognition algorithm to perform text recognition on the homework answer image to obtain the homework answer text;

[0099] Use the image text recognition algorithm to perform text recognition on the target job image to obtain the target job text;

[0100] Split the target homework text into T1 homework text to Ta homework text according to the homework title sequence number in the target homework text;

[0101] What is explained here is:

[0102] In the present application, the character T involved here is an identifier corresponding to the job text, a is the corresponding quantity value of the job text split from the target job text, and a is an integer greater than 0.

[0103] In the homework answer text, the homework answers corresponding to the T1 homework text to the Ta homework text are split to obtain the T1 answer text to the Ta answer text;

[0104] It should be noted here that:

[0105] See also Figure 2 In this application, the T1 answer text to Ta answer text involved here corresponds one to one with the T1 assignment text to Ta assignment text.

[0106] Define T1 assignment text to Ta assignment text and T1 answer text to Ta answer text as preliminary collection data for the assignment;

[0107] Step S2: Divide multiple homework texts into first-type homework texts and second-type homework texts according to the preliminary collection data of homework, correct the first-type homework texts by using text comparison, create a homework correction large language model to correct the second-type homework texts, obtain the correctness of the answer corresponding to each homework text according to the correction result, and obtain homework correction data;

[0108] Step S2 also includes the following specific steps:

[0109] Obtain preliminary collected data of the assignment, and obtain T1 assignment text to Ta assignment text and T1 answer text to Ta answer text respectively according to the preliminary collected data of the assignment;

[0110] Mark the objective question texts from T1 homework text to Ta homework text as the first type of homework text, and mark the subjective question texts from T1 homework text to Ta homework text as the second type of homework text;

[0111] It should be noted here that:

[0112] In this application, the T1 homework text to the Ta homework text involved here only include subjective questions and objective questions;

[0113] The objective questions involved here include single-choice questions, multiple-choice questions, fill-in-the-blank questions and judgment questions.

[0114] The second type of homework text is graded and obtained to obtain the correctness of the answer corresponding to each second type of homework text;

[0115] The details are as follows:

[0116] If the Ti homework text is the first type of homework text, and the Ti answer text is a single-field text, the Ti homework text and the Ti answer text are compared. If the text comparison is consistent, the correctness of the answer corresponding to the Ti homework text is 100%;

[0117] If the Ti homework text is the first type of homework text, but the Ti answer text is not a single-field text, count the number of character segments contained in the Ti answer text to obtain the answer field quantity value, perform field comparison on the Ti homework text and the Ti answer text, mark the fields that match the comparison as correct answer fields, count the number of correct answer fields to obtain the correct field quantity value, calculate the ratio of the correct answer field quantity value to the answer field quantity value, and obtain the field correct ratio Zbi, then judge that the answer correctness corresponding to the Ti homework text is Zbi%;

[0118] It should be noted here that:

[0119] In the present application, the Ti job text involved here can be any second type job text from T1 job text to Ta job text.

[0120] Obtaining the results of grading the first type of homework texts, and obtaining the correctness of the answers corresponding to each first type of homework text;

[0121] The specific steps are as follows:

[0122] When the Tj assignment text is the second type assignment text, the Tj answer text is divided into a plurality of answer key fields, and the divided plurality of answer key fields are named G1 answer field to Gb answer field;

[0123] Set the answer accuracy contribution ratios for the G1 answer field to the Gb answer field respectively, and obtain the answer accuracy contribution ratios for G1 to Gb;

[0124] It should be noted here that:

[0125] In the present application, G referred to here is the identifier corresponding to the key field of the answer, b is the quantity value corresponding to the key field of the answer, and b is an integer greater than 0.

[0126] In the present application, the answer correctness contribution ratio involved here is an indicator to measure the correctness of the answer field. For example, if the homework questions corresponding to the G1 answer field to the Gb answer field are used as examination questions, then the answer correctness contribution ratio corresponding to the G1 answer field to the Gb answer field can be quantified as the ratio of the scores corresponding to the score points of the G1 answer field to the Gb answer field to the total score of the test question.

[0127] In the present application, the answer monitoring fields involved here can be key words or short sentences in the answers to subjective questions in liberal arts type assignments, and the answer monitoring fields involved here can be formulas, parameter calculation results, and phased problem-solving steps in science type assignments.

[0128] Use the G1 answer field to the Gb answer field to train the existing large language model to obtain the Tj homework correction large language model;

[0129] The details are as follows:

[0130] Obtain the student homework completion texts corresponding to several Tj answer texts, obtain multiple homework completion texts, and obtain the answer correctness corresponding to each homework completion text respectively;

[0131] The details are as follows:

[0132] A sample homework completion text is selected from multiple homework completion texts, and a text comparison algorithm is used to compare characters of the sample homework completion text with the G1 answer field to the Gb answer field. If the G1 answer field exists in the sample homework completion text, a value of 1 is used to assign a field existence degree corresponding to the G1 answer field. If the G1 answer field does not exist in the sample homework completion text, a value of 0 is used to assign a field existence degree corresponding to the G1 answer field to obtain the G1 field existence degree.

[0133] It should be noted here that:

[0134] In the present application, the text comparison algorithm involved here may be a Boolean model.

[0135] Perform character comparison between the sample homework completion text and the G2 answer field to the Gb answer field. If the G2 answer field exists in the sample homework completion text, assign a value of 1 to the field existence degree corresponding to the G2 answer field. If the G2 answer field does not exist in the sample homework completion text, assign a value of 1 to the field existence degree corresponding to the G2 answer field to obtain the G2 field existence degree.

[0136] Similarly, the field existences corresponding to the G3 answer field to the Gb answer field are assigned values ​​respectively, and the G3 field existences to the Gb field existences are obtained;

[0137] The answer correctness corresponding to the sample homework completion text is obtained by calculating the contribution ratio of G1 answer correctness to Gb answer correctness and the G1 field existence to Gb field existence;

[0138] The correctness of the answers corresponding to the completed text of the sample homework is calculated. The specific formula is as follows:

[0139] ;

[0140] Among them, Zqd is the answer correctness corresponding to the sample homework completion text, Gzqi is the contribution ratio of Gi answer correctness, Gxdi is the Gi field existence, and b is the number value corresponding to the key field of the answer;

[0141] Repeat the process of obtaining the correctness of the answers corresponding to the sample homework completion texts, and obtain the correctness of the answers corresponding to each homework completion text respectively;

[0142] Use the correctness of the answer of each homework completion text to mark the corresponding homework completion text, and obtain homework completion text marking data;

[0143] The text labelled data is divided into a job completion text training set and a job completion text test set according to the image training and testing ratio;

[0144] It should be noted here that:

[0145] In this application, the image training-test ratio is specifically set to 7:3, that is, the ratio of the number of medical job completion texts in the job completion text training set and the job completion text test set is 7:3;

[0146] Obtain a large language model through the existing artificial intelligence platform and train the large language model using the job completion text training set;

[0147] Use the homework completion text test set to test the large language model and obtain the recognition accuracy. When the recognition accuracy is greater than or equal to the target recognition accuracy, the training of the large language model is completed and the Tj homework correction large language model is obtained. When the recognition accuracy is less than the target recognition accuracy, continue to use the homework completion text training set to train the large language model until the recognition accuracy is greater than or equal to the target recognition accuracy.

[0148] It should be noted here that:

[0149] The target recognition accuracy involved here is specifically set to 95% in this application.

[0150] Use the Tj homework correction large language model to correct the Tj homework text and obtain the correctness of the answer corresponding to the Tj homework text;

[0151] The answer correctness corresponding to homework text T1 to homework text Ta is named as T1 answer correctness to Ta answer correctness, and homework correction data is obtained.

[0152] Step S3: Analyze the correctness of answers to the homework submitted by the target student based on the preliminary collection data and the homework correction data, and obtain the first homework quality evaluation coefficient and the second homework quality evaluation coefficient according to the analysis results to obtain correctness analysis data;

[0153] Step S3 also includes the following specific steps:

[0154] Obtain preliminary collected data for the operation, obtain preliminary collected data for the operation, and obtain T1 operation text to Ta operation text respectively according to the preliminary collected data for the operation;

[0155] Set the teaching knowledge points P1 to Pc from the homework text T1 to the homework text Ta according to the teaching knowledge points corresponding to the homework content;

[0156] It should be noted here that:

[0157] In this application, c here refers to the number of teaching knowledge points corresponding to the homework text T1 to the homework text Ta;

[0158] See also Figure 3 ,In this application, a single homework text can only match one teaching ,knowledge point, but a single teaching knowledge point can be matched by multiple ,homework texts.

[0159] Obtain homework correction data, and obtain the correctness of answers T1 to Ta respectively according to the homework correction data;

[0160] According to the correctness of the answer of T1 to the correctness of the answer of Ta, the correctness of the answer corresponding to the teaching knowledge point of P1 to the teaching knowledge point of Pc is averaged to obtain the correctness of the answer of the knowledge point of P1 to the correctness of the answer of the knowledge point of Pc;

[0161] Compare and analyze the correctness of answers to knowledge points P1 to Pc and the historical correctness of knowledge points to obtain the first homework quality evaluation coefficient;

[0162] The details are as follows:

[0163] Obtain the target student's historical homework modification records, select a number of historical batches of homework from the target student's historical homework modification records, and name the obtained historical batches of homework as the first historical batch of homework to the dth historical batch of homework in the order of homework submission time;

[0164] It should be noted here that:

[0165] In this application, the homework contents corresponding to the first historical batch homework to the dth historical batch homework involved here all include the examination of the teaching knowledge points P1 to Pc;

[0166] In the present application, d referred to here is the quantity value corresponding to the historical batch job, and d is an integer greater than 0.

[0167] Obtain the correctness of the assignments corresponding to the P1 teaching knowledge point to the Pc teaching knowledge point in the first historical batch of assignments, and obtain the correctness of the first historical P1 assignment to the first historical Pc assignment; obtain the correctness of the assignments corresponding to the P1 teaching knowledge point to the Pc teaching knowledge point in the second historical batch of assignments, and obtain the correctness of the second historical P1 assignment to the second historical Pc assignment; and so on, obtain the correctness of the assignments corresponding to the P1 teaching knowledge point to the Pc teaching knowledge point in the dth historical batch of assignments, and obtain the correctness of the dth historical P1 assignment to the dth historical Pc assignment;

[0168] The accuracy of the first historical P1 operation to the dth historical P1 operation is averaged to obtain the accuracy of the historical operation of the P1 knowledge point. The accuracy of the first historical P2 operation to the dth historical P2 operation is averaged to obtain the accuracy of the historical operation of the P2 knowledge point. Similarly, the accuracy of the first historical Pc operation to the dth historical Pc operation is averaged to obtain the accuracy of the historical operation of the Pd knowledge point.

[0169] The first homework quality evaluation coefficient is obtained by calculating the accuracy of the historical homework of the P1 knowledge point to the accuracy of the historical homework of the Pd knowledge point and the accuracy of the answer of the P1 knowledge point to the accuracy of the answer of the Pc knowledge point;

[0170] The quality evaluation coefficient of the first operation is calculated, and the specific formula is as follows:

[0171] ;

[0172] Among them, Zyz1 is the first homework quality evaluation coefficient, Zqdi is the correctness of the answer to the Pi knowledge point, Lszi is the correctness of the historical homework of the Pi knowledge point, and c is the number of teaching knowledge points corresponding to the homework text from T1 to Ta;

[0173] It should be noted here that:

[0174] In the present application, the Pi knowledge point answer accuracy involved here can be any one of the knowledge point answer accuracy from the P1 knowledge point answer accuracy to the Pc knowledge point answer accuracy, and the Pi knowledge point historical assignment accuracy involved here can be any one of the knowledge point historical assignment accuracy from the P1 knowledge point historical assignment accuracy to the Pd knowledge point historical assignment accuracy.

[0175] Compare and analyze the correctness of answers to knowledge points P1 to Pc with the average correctness of answers submitted in the same batch to obtain the second homework quality evaluation coefficient;

[0176] The details are as follows:

[0177] In the current batch of submitted homework, the batch average answer correctness of the homework text corresponding to the P1 teaching knowledge point is obtained, and the batch correctness of the P1 knowledge point is obtained. The batch average answer correctness of the homework text corresponding to the P2 teaching knowledge point is obtained, and the batch correctness of the P2 knowledge point is obtained. And so on, the batch average answer correctness of the homework text corresponding to the Pc teaching knowledge point is obtained, and the batch correctness of the Pc knowledge point is obtained;

[0178] The second task quality evaluation coefficient is obtained by calculating the correctness of the answer to the knowledge point P1 to the correctness of the answer to the knowledge point Pc and the correctness of the batch of knowledge points P1 to the batch of knowledge points Pc;

[0179] The second operation quality evaluation coefficient is calculated, and the specific formula is as follows:

[0180] ;

[0181] Among them, Zyz2 is the quality evaluation coefficient of the second homework, Zqdi is the correctness of the answer to the Pi knowledge point, Pczi is the batch correctness of the Pi knowledge point, and c is the number of teaching knowledge points corresponding to the homework text of T1 to the homework text of Ta;

[0182] It should be noted here that:

[0183] In the present application, the Pi knowledge point answer correctness involved here may be any one of the knowledge point answer correctness from the P1 knowledge point answer correctness to the Pc knowledge point answer correctness, and the Pi knowledge point batch correctness involved here may be any one of the knowledge point batch correctness from the P1 knowledge point batch correctness to the Pc knowledge point batch correctness;

[0184] defining a first operation quality assessment coefficient and a second operation quality assessment coefficient as correctness analysis data;

[0185] Step S4: evaluating the completion quality of the current batch of homework submitted by the target students by analyzing the correctness analysis data;

[0186] Step S4 also includes the following specific steps:

[0187] Acquire correctness analysis data, and acquire a first operation quality assessment coefficient and a second operation quality assessment coefficient according to the correctness analysis data;

[0188] The first operation quality assessment coefficient and the second operation quality assessment coefficient are calculated to obtain a comprehensive operation quality assessment coefficient;

[0189] The comprehensive evaluation coefficient of operation quality is calculated, and the specific formula is as follows:

[0190] ;

[0191] Among them, Zpg is the comprehensive evaluation coefficient of operation quality, Zyz1 is the first operation quality evaluation coefficient, and Zyz2 is the second operation quality evaluation coefficient;

[0192] Obtaining a threshold value of a comprehensive evaluation coefficient of homework quality, comparing the comprehensive evaluation coefficient of homework quality with the threshold value of the comprehensive evaluation coefficient of homework quality, and conducting a quality qualification assessment on the target students according to the numerical comparison result;

[0193] Obtain a first operation quality assessment coefficient threshold and a second operation quality assessment coefficient threshold, and calculate the first operation quality assessment coefficient threshold and the second operation quality assessment coefficient threshold to obtain an operation quality comprehensive assessment coefficient threshold;

[0194] It should be noted here that:

[0195] In the present application, the first job quality assessment coefficient threshold and the second job quality assessment coefficient threshold involved here are the minimum first job quality assessment coefficient and the minimum second job quality assessment coefficient corresponding to the job text that passes the job quality assessment.

[0196] The threshold of the comprehensive evaluation coefficient of job quality is calculated. The specific formula is as follows:

[0197] ;

[0198] Among them, Zpgy is the comprehensive evaluation coefficient of operation quality, Zyzy1 is the first operation quality evaluation coefficient, and Zyzy2 is the second operation quality evaluation coefficient;

[0199] When the comprehensive evaluation coefficient of homework quality is greater than or equal to the comprehensive evaluation coefficient threshold of homework quality, it is judged that the quality evaluation of the homework submitted by the target students in the current batch is qualified;

[0200] Get the comprehensive evaluation coefficient threshold of homework quality. When the comprehensive evaluation coefficient of homework quality is less than the comprehensive evaluation coefficient threshold, it is judged that the quality evaluation of the homework submitted by the target students in the current batch is unqualified.

[0201] In this application, if corresponding calculation formulas appear, the above calculation formulas are all dimensionless and take their numerical calculations. The weight coefficients, proportional coefficients and other coefficients in the formulas are set to a result value obtained by quantifying each parameter. The size of the weight coefficient and the proportional coefficient can be determined as long as it does not affect the proportional relationship between the parameter and the result value.

[0202] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the specific implementation methods of the present invention. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A homework grading method based on image recognition and large language model, characterized in that: include: Step S1: obtaining a target task image and a task answer image, splitting the target task image and the task answer image into a plurality of task texts and a plurality of answer texts respectively, and obtaining preliminary task collection data; Step S2: Divide multiple homework texts into first-type homework texts and second-type homework texts according to the preliminary collection data of homework, correct the first-type homework texts by using text comparison, create a homework correction large language model to correct the second-type homework texts, obtain the correctness of the answer corresponding to each homework text according to the correction result, and obtain homework correction data; Step S3: Analyze the correctness of answers to the homework submitted by the target student based on the preliminary collection data and the homework correction data, and obtain the first homework quality evaluation coefficient and the second homework quality evaluation coefficient according to the analysis results to obtain correctness analysis data; Step S3 also includes the following specific steps: Step S31: Obtain preliminary collected data of the operation, obtain preliminary collected data of the operation, and obtain T1 operation text to Ta operation text respectively according to the preliminary collected data of the operation; Step S32: setting the teaching knowledge points P1 to Pc from the homework text T1 to the homework text Ta according to the teaching knowledge points corresponding to the homework content; Step S33: obtaining homework correction data, and obtaining T1 answer accuracy to Ta answer accuracy respectively according to the homework correction data; Step S34: averaging the answer correctness corresponding to the teaching knowledge points P1 to Pc according to the answer correctness T1 to Ta, and obtaining the answer correctness of the knowledge point P1 to the knowledge point Pc; Step S35: Compare and analyze the correctness of the answers to the knowledge points P1 to Pc and the historical correctness of the knowledge points to obtain a first homework quality evaluation coefficient; Step S36: Compare and analyze the correctness of the answers to the knowledge points P1 to Pc with the average correctness of the answers to the assignments submitted in the same batch to obtain a second assignment quality evaluation coefficient; Step S37: defining the first operation quality assessment coefficient and the second operation quality assessment coefficient as correctness analysis data; Step S36 also includes the following specific steps: Step S361: in the current batch of submitted homework, the batch average answer accuracy of the homework text corresponding to the P1 teaching knowledge point is obtained to obtain the batch accuracy of the P1 knowledge point. Similarly, the batch average answer accuracy of the homework text corresponding to the Pc teaching knowledge point is obtained to obtain the batch accuracy of the Pc knowledge point. Step S362: calculating the second job quality evaluation coefficient by converting the correctness of the answer to the knowledge point P1 to the correctness of the answer to the knowledge point Pc and the correctness of the batch of knowledge points P1 to the correctness of the batch of knowledge points Pc; The quality evaluation coefficient of the second operation is calculated, and the specific formula is as follows: ； Among them, Zyz2 is the quality evaluation coefficient of the second homework, Zqdi is the correctness of the answer to the Pi knowledge point, Pczi is the correctness of the Pi knowledge point batch, and c is the number of teaching knowledge points; Step S4: Evaluate the completion quality of the current batch of homework submitted by the target students by analyzing the correctness analysis data.

2. The homework grading method based on image recognition and large language model according to claim 1 is characterized in that: Step S1 also includes the following specific steps: Step S11: selecting a student from among the students who submitted homework in the current batch as a target student, acquiring an image of the homework submitted by the target student, and obtaining a target homework image; Step S12: acquiring an image of the standard answer to the homework submitted by the target student to obtain a homework answer image; Step S13: Perform text recognition on the homework answer image to obtain the homework answer text; Step S14: performing text recognition on the target job image to obtain the target job text; Step S15: splitting the target job text into T1 job text to Ta job text; Step S16: In the homework answer text, the homework answers corresponding to the T1 homework text to the Ta homework text are split to obtain the T1 answer text to the Ta answer text; Step S17: Define T1 job text to Ta job text and T1 answer text to Ta answer text as preliminary job collection data.

3. The homework grading method based on image recognition and large language model according to claim 1 is characterized in that: Step S2 also includes the following specific steps: Step S21: obtaining preliminary collected data of the operation, and obtaining T1 operation text to Ta operation text and T1 answer text to Ta answer text respectively according to the preliminary collected data of the operation; Step S22: marking the objective question texts in the homework text T1 to the homework text Ta as the first type of homework text, and marking the subjective question texts in the homework text T1 to the homework text Ta as the second type of homework text; Step S23: obtaining the results of grading the first type of homework texts, and obtaining the correctness of the answers corresponding to each first type of homework text; Step S24: obtaining the correction results of the second type of homework texts, and obtaining the correctness of the answers corresponding to each second type of homework text; Step S25: Name the answer correctness corresponding to the homework text T1 to the homework text Ta as T1 answer correctness to Ta answer correctness, and obtain homework correction data.

4. The homework grading method based on image recognition and large language model according to claim 3 is characterized in that: Step S23 also includes the following specific steps: Step S231: If the Ti homework text is a first type of homework text and the Ti answer text is a single field text, the Ti homework text and the Ti answer text are compared. If the text comparison is consistent, the correctness of the answer corresponding to the Ti homework text is 100%; Step S232: If the Ti homework text is a first type of homework text and the Ti answer text is not a single field text, count the number of character segments contained in the Ti answer text to obtain the answer field quantity value, perform field comparison on the Ti homework text and the Ti answer text, mark the fields that match the comparison as correct answer fields, count the number of correct answer fields to obtain the correct field quantity value, calculate the ratio of the correct answer field to the answer field quantity value, obtain the field correct ratio Zbi, and then judge that the answer accuracy corresponding to the Ti homework text is Zbi%.

5. The homework grading method based on image recognition and large language model according to claim 3 is characterized in that: Step S24 also includes the following specific steps: Step S241: when the Tj assignment text is the second type assignment text, the Tj answer text is divided into the G1 answer field to the Gb answer field; Step S242: setting the answer accuracy contribution ratios for the G1 answer field to the Gb answer field respectively, and obtaining the answer accuracy contribution ratios for G1 to Gb; The answer correctness contribution ratio is the ratio of the score assigned to the score point corresponding to the answer field to the total score of the test question; Step S243: Use the G1 answer field to the Gb answer field to train the existing large language model to obtain the Tj homework correction large language model; Step S244: Use the Tj homework correction large language model to correct the Tj homework text and obtain the correctness of the answer corresponding to the Tj homework text.

6. The homework grading method based on image recognition and large language model according to claim 5, characterized in that: Step S243 also includes the following specific steps: Step S2431: obtaining multiple homework completion texts, and obtaining the correctness of the answer corresponding to each homework completion text; Step S2432: selecting a sample homework completion text from multiple homework completion texts, and obtaining the correctness of the answer corresponding to the sample homework completion text; Step S2433: Obtaining the correctness of the answer corresponding to each homework completion text; Step S2434: Use the correctness of the answer to mark the corresponding homework completion text to obtain homework completion text marking data; Step S2435: dividing the text labeling data into a job completion text training set and a job completion text test set according to the image training test ratio; Step S2436: Obtain a large language model through the existing artificial intelligence platform, use the homework completion text training set and the homework completion text test set to train and test the large language model, and obtain the Tj homework correction large language model.

7. The homework grading method based on image recognition and large language model according to claim 6 is characterized in that: Step S2432 also includes the following specific steps: Use a text comparison algorithm to compare the characters of the sample homework completion text with the G1 answer field to the Gb answer field. If the G1 answer field exists in the sample homework completion text, assign a value of 1 to the field existence degree corresponding to the G1 answer field. If the G1 answer field does not exist in the sample homework completion text, assign a value of 0 to the field existence degree corresponding to the G1 answer field to obtain the G1 field existence degree. Similarly, assign values ​​to the field existences corresponding to the G2 answer field to the Gb answer field, respectively, to obtain the G2 field existences to the Gb field existences; The answer correctness corresponding to the sample homework completion text is obtained by calculating the contribution ratio of G1 answer correctness to Gb answer correctness and the G1 field existence to Gb field existence; The correctness of the answers corresponding to the completed text of the sample homework is calculated. The specific formula is as follows: ； Among them, Zqd is the answer accuracy corresponding to the sample homework completion text, Gzqi is the contribution ratio of Gi answer accuracy, Gxdi is the existence of Gi field, and b is the quantity value corresponding to the key field of the answer.

8. The homework grading method based on image recognition and large language model according to claim 1, characterized in that: Step S35 also includes the following specific steps: Step S351: Obtain the target student's historical homework modification records, and select the first to dth historical batches of homework from the target student's historical homework modification records; Step S352: Obtain the correctness of the assignments corresponding to the P1 teaching knowledge point to the Pc teaching knowledge point in the first historical batch of assignments, and obtain the correctness of the first historical P1 assignment to the first historical Pc assignment. Similarly, obtain the correctness of the assignments corresponding to the P1 teaching knowledge point to the Pc teaching knowledge point in the dth historical batch of assignments, and obtain the correctness of the dth historical P1 assignment to the dth historical Pc assignment. Step S353: Calculate the average of the correctness of the first historical P1 operation to the correctness of the dth historical P1 operation to obtain the correctness of the historical operation of the P1 knowledge point. Similarly, calculate the average of the correctness of the first historical Pc operation to the correctness of the dth historical Pc operation to obtain the correctness of the historical operation of the Pd knowledge point. Step S354: calculating the first homework quality evaluation coefficient by combining the historical homework correctness of the knowledge point P1 to the historical homework correctness of the knowledge point Pd and the answer correctness of the knowledge point P1 to the answer correctness of the knowledge point Pc; The quality evaluation coefficient of the first operation is calculated, and the specific formula is as follows: ； Among them, Zyz1 is the quality evaluation coefficient of the first homework, Zqdi is the correctness of the answer to the Pi knowledge point, Lszi is the correctness of the historical homework of the Pi knowledge point, and c is the number of teaching knowledge points.

9. The homework grading method based on image recognition and large language model according to claim 1, characterized in that: Step S4 also includes the following specific steps: Acquire correctness analysis data, and acquire a first operation quality assessment coefficient and a second operation quality assessment coefficient according to the correctness analysis data; The first operation quality assessment coefficient Zyz1 and the second operation quality assessment coefficient Zyz2 are calculated to obtain the operation quality comprehensive assessment coefficient Zpg; Obtain the comprehensive evaluation coefficient threshold of job quality; When the comprehensive evaluation coefficient of homework quality is greater than or equal to the comprehensive evaluation coefficient threshold of homework quality, it is judged that the quality evaluation of the homework submitted by the target students in the current batch is qualified; When the comprehensive evaluation coefficient of homework quality is less than the comprehensive evaluation coefficient threshold of homework quality, it is judged that the quality evaluation of the homework submitted by the target students in the current batch is unqualified.

Images