KR20240017495A - Autonomous grading chatbot using artificial intelligence - Google Patents

Abstract

A server that communicates with a user terminal over a network to perform automatic scoring through a chatbot includes: a processor including one or more cores; and a memory, wherein the processor includes: an automatic scoring engine that provides automatic scoring based on an artificial intelligence (AI) optical character recognition (OCR) model for textbook images received from the user terminal; A learning conversation processing engine that provides a learning consultation conversation based on an AI natural language processing model to the user terminal; And a learning path recommendation algorithm based on a user model modeled for the user based on the user's learning history of the user terminal and the conversation log with the user may be executed.

Description

Automatic grading chatbot using artificial intelligence {AUTONOMOUS GRADING CHATBOT USING ARTIFICIAL INTELLIGENCE}

This disclosure relates to an automatic scoring chatbot using artificial intelligence.

Chatbot is an abbreviation for chat robot and refers to an artificial intelligence communication program that provides appropriate answers to questions or various related information through text conversations with people based on artificial intelligence. These chatbots are interactive services through messenger applications and are provided through preset rules or artificial intelligence technology.

Recently, these chatbots are being used in various fields of industry, and the fields in which chatbots are used are gradually expanding. In particular, chatbots are widely used to provide customer consultation or customer service. The size of the global chatbot market is expected to grow from $2.57 billion in 2020 to $9.42 billion in 2024. In addition, it is analyzed that the domestic chatbot market is growing at an average annual rate of about 51%.

For example, Korea Patent Publication No. 10-2019-0080599 (Patent Document 1) discloses a method of providing a semi-automatic conversation using a chatbot and a counselor performed by a server.

Meanwhile, recently, non-face-to-face education has been emphasized due to the impact of COVID-19, and online education is being conducted in various ways, such as video or real-time video learning. However, as a result, learning management, which was previously burdened by school teachers, academy instructors, or visiting home tutors, can now be done at home. In other words, due to non-face-to-face education, the burden of managing home learning has increased.

In the educational publishing industry, the size of the workbook and reference book market is estimated at approximately 2.5 trillion. However, after students have studied based on published textbooks, that is, published reference books, the burden of having to manage their learning at home or on their own has further increased due to COVID-19. In addition, it was confirmed that sales of home-based study materials decreased significantly by nearly 13.8% due to COVID-19.

Accordingly, the need for an educational chatbot that can manage students' learning non-face-to-face after the sale of textbooks and that can provide automatic grading and learning consultation conversations has emerged.

Patent Document 1: Public Patent Publication No. 10-2019-0080599 (published on July 8, 2019)

This disclosure is intended to solve the above-described problem. When a student photographs a textbook he or she has solved and uploads the photographed textbook photo, the student's answers included in the textbook photo are automatically graded and the grading results and the student's study schedule are recorded. This is about an automatic grading chatbot using artificial intelligence that provides learning consultation conversations based on learning.

In one embodiment of the present disclosure, a server that communicates with a user terminal over a network to perform automatic scoring through an artificial intelligence (AI) chatbot includes: a processor including one or more cores; and a memory, wherein the processor includes: an automatic scoring engine that provides automatic scoring based on an AI optical character recognition (OCR) model for textbook images received from the user terminal; A learning conversation processing engine that provides a learning consultation conversation based on an AI natural language processing model to the user terminal; And a learning path recommendation algorithm based on a user model modeled for the user based on the user's learning history of the user terminal and the conversation log with the user may be executed.

A computer-readable medium storing a computer program for performing automatic grading through a chatbot according to another embodiment of the present disclosure, when the computer program is executed by a processor, causes the processor to: display a teaching material image received from the user terminal; Provides automatic grading based on an artificial intelligence (AI) optical character recognition (OCR) model; providing a learning consultation conversation based on an AI natural language processing model to the user terminal; And may cause the user terminal to execute a learning path recommendation algorithm based on a user model modeled for the user based on the user's learning history and conversation log with the user.

According to an automatic grading chatbot using artificial intelligence according to an embodiment of the present disclosure, images of textbooks taken by a user are scored, and a learning consultation conversation can be provided to the user based on the grading results. As a result, the results of solving existing offline textbooks can be digitized, and the user's grading stress can be reduced.

Additionally, according to the present disclosure, when linking an online learning tool and a textbook, automatic grading can be performed online simply by taking an image of the textbook without the need to input answers one by one. Automatic grading can be provided to elementary school students who find it difficult to grade on their own, or to instructors, teachers, or parents who feel burdened by grading.

Furthermore, according to the present disclosure, it is possible for the user's learning schedule to be managed to improve learning capabilities and habits, based on the automatic scoring results. Additionally, according to the present disclosure, an optimal learning path can be recommended based on the user's learning history and learning consultation conversation log.

In addition, according to the present disclosure, a learning consultation conversation can be provided, providing the student with the role of a management teacher or friend, thereby helping to relieve learning-related stress or problems.

As a result, the user's learning efficiency can be increased and student follow-up management can be realized at a low cost even after purchasing textbooks, thereby improving student learning satisfaction.

The effects of the present disclosure are not limited to the effects described above, and effects not mentioned can be clearly understood by those skilled in the art from this specification and the attached drawings.

1 shows a block diagram of a system for providing an artificial intelligence automatic grading tutorbot according to some embodiments of the present disclosure.
Figure 2 shows a flowchart between an artificial intelligence automatic grading tutorbot and a user terminal according to some embodiments of the present disclosure.
Figure 3 shows a schematic diagram of the operation interface of an artificial intelligence automatic grading tutorbot according to some embodiments of the present disclosure.
Figure 4 shows a flowchart of a method by which an artificial intelligence automatic grading tutorbot recognizes printed and cursive fonts from textbook images according to some embodiments of the present disclosure.
Figure 5 shows a schematic diagram to explain how an artificial intelligence automatic grading tutorbot prints printed and cursive fonts from textbook images according to some embodiments of the present disclosure.
Figure 6 shows a flowchart between an artificial intelligence automatic grading tutorbot and a user terminal according to some embodiments of the present disclosure.
Figure 7 shows a schematic diagram of the operation interface of an artificial intelligence automatic grading tutorbot according to some embodiments of the present disclosure.
Figure 8 shows a schematic diagram of the operation interface of an artificial intelligence automatic grading tutorbot according to some embodiments of the present disclosure.
Figure 9 shows a flowchart of a method for building an OCR engine for an artificial intelligence automatic grading tutorbot according to some embodiments of the present disclosure.
Figure 10 shows a flowchart of a method for building a natural language processing engine for an artificial intelligence automatic grading tutorbot according to some embodiments of the present disclosure.
Figure 11 shows a flowchart of a method for building a learning path recommendation algorithm for an artificial intelligence automatic grading tutorbot according to some embodiments of the present disclosure.
Figure 12 shows a block diagram of a system for providing an artificial intelligence automatic grading tutorbot according to some embodiments of the present disclosure.

Hereinafter, various embodiments of the present invention will be described, but of course, the technical idea of the present invention is not limited thereto and may be modified and implemented in various ways by those skilled in the art. In addition, the terms used in this specification are used to describe the concept of the present invention and embodiments thereof, and are not intended to limit the present invention to only the dictionary or phraseological meaning of the terms. For example, herein, singular expressions may include plural expressions, unless the context clearly dictates otherwise. In addition, in this specification, terms such as "include" or "have" mean the presence of a certain configuration, step, operation, processing, or a combination thereof, and unless the context clearly indicates otherwise, other configurations, steps, operations, It does not exclude the presence or addition of processing.

Terms such as first, second, A, B, etc. may be used to describe various components, but the components should not be limited by the terms. The above terms are used only for the purpose of distinguishing one component from another. For example, a first component may be referred to as a second component, and similarly, the second component may be referred to as a first component without departing from the scope of the present invention. The term “and/or” includes any combination of a plurality of related stated items or any of a plurality of related stated items.

In embodiments of the present disclosure, terms such as “module”, “unit”, “part”, etc. are terms to refer to components that perform at least one function or operation, and these components are either hardware or software. It may be implemented or may be implemented through a combination of hardware and software. In addition, a plurality of "modules", "units", "parts", etc. are integrated into at least one module or chip, except in cases where each needs to be implemented with individual specific hardware, and is integrated into at least one processor. It can be implemented as:

Additionally, in an embodiment of the present disclosure, when a part is connected to another part, this includes not only direct connection but also indirect connection through other media. In addition, the meaning that a part includes a certain component does not mean that other components are excluded, but that it may further include other components, unless specifically stated to the contrary.

Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the technical field to which the present invention pertains. Terms defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related technology, and unless explicitly defined in the present application, should not be interpreted in an ideal or excessively formal sense. No.

The AI automatic scoring tutorbot according to some embodiments of the present disclosure receives an image of a textbook (or workbook) solved by a user (e.g., a student), automatically scores the questions included in the textbook image, Users can provide grading results. In addition, the AI automatic grading tutor bot can provide learning consultation conversations to the user based on grading results, the user's learning history of the user terminal, and the user's utterances. The learning consultation conversation provided by the AI automatic grading tutorbot may include solving incorrectly answered questions, recommending similar problems similar to incorrectly answered questions, encouraging learning, suggesting additional learning methods, and managing study schedules.

The AI automatic grading tutorbot according to some embodiments of the present disclosure may be implemented as a chatbot. For example, an AI automatic grading tutorbot can infer the intention of the utterance from the utterance in the form of a text message entered by the user into the chatbot and provide the user with an utterance in the form of a text message corresponding to the intention. The AI automatic grading tutorbot can exchange text messages with users by displaying a text messenger-type user interface (UI) on the user terminal. The AI automatic scoring tutorbot can generate utterances in the form of text messages containing specific intentions, and provide the generated utterances to users through the text messenger UI. Accordingly, the user can communicate with the AI automatic grading tutorbot in the form of a visual conversation (e.g., in the form of a text messenger).

1 shows a block diagram of an AI automatic grading tutorbot providing system according to some embodiments of the present disclosure. Referring to FIG. 1, the AI automatic grading tutorbot providing system includes a user terminal 10 and AI automatic grading. Tutorbot (100) may be included.

The AI automatic grading tutorbot 100 may include an automatic grading engine 110, a learning conversation processing engine 120, and a learning path recommendation algorithm 130.

The automatic grading engine 110 may provide automatic grading for textbook images received from the user terminal 10 based on an AI Optical Character Recognition (OCR) model. For example, the automatic scoring engine 110 may receive a textbook image captured by a user from the user terminal 10 and score one or more questions in the received textbook image.

The automatic scoring engine 110 may include an OCR engine 111, a scoring module 112, and a textbook and content database (DB) 113.

The OCR engine 111 can recognize printed and cursive fonts in the textbook image provided from the user terminal 10 based on the AI OCR model. For example, the OCR engine 111 can recognize printed text such as page numbers, question (problem) numbers, and answer box areas within textbook images based on the AI OCR model. The OCR engine 111 can recognize handwriting, such as handwriting in the answer box in the textbook image, based on the AI OCR model.

The scoring module 112 can score the user's answer written in the textbook image based on the printed text and cursive text recognized by the OCR engine 111. For example, the scoring module 112 may score questions in a textbook image provided from the user terminal 10 based on the recognition results of the OCR engine 111 and data stored in the textbook and content DB 113.

The scoring module 112 can search the textbook and content DB 113 for data including questions corresponding to the printed text recognized by the OCR engine 111 and the correct answers and solutions paired therewith. The scoring module 112 can score the answers indicated by the user in the textbook by comparing/contrasting the searched correct answer with the user-entered answer corresponding to the handwriting recognized by the OCR engine 111.

In some embodiments, the automatic scoring engine 110 may output a UI to the user terminal 10 to allow the user to confirm the recognition result of the OCR engine 111. The UI for checking the recognition result as a user may include a query as to whether the user can modify the recognition result.

In some embodiments, the automatic scoring engine 110 may receive a recognition result modified by the user from the user terminal 10 in response to a UI for confirming the recognition result with the user. The automatic scoring engine 110 may re-score the user's answer based on the corrected recognition result.

The textbook and content DB 113 can store data related to textbooks or content for which the AI automatic grading tutorbot 100 can provide automatic grading. For example, the textbook and content DB 113 may store textbook questions that can be graded by the AI automatic scoring tutorbot 100 and the correct answers and solution data for each question.

The learning conversation processing engine 120 receives an utterance in text form (e.g., learning-related natural language) from the user terminal 10, processes (or analyzes) the utterance, and generates a response corresponding to the received utterance. can do. In some embodiments, the learning conversation processing engine 120 may provide a learning consultation conversation based on an AI natural language processing model to a user terminal.

The learning conversation processing engine 120 may include a natural language processing engine 121, a response generation module 122, and a learning history DB 123.

The natural language processing engine 121 may analyze the user's utterance received from the user terminal 10 based on an AI natural language processing model. For example, the natural language processing engine 121 may extract keywords from the user's utterance, and based on the extracted keywords, infer the intention of the user's utterance. For example, the natural language processing engine 121 can infer (or understand) the user's emotions or the user's learning state from the user's utterance.

The response generation module 122 may generate a response corresponding to the utterance received from the user terminal 10 based on the analysis result of the natural language processing engine 121. For example, the response generation module 122 may use the intention of the user's utterance inferred by the natural language processing engine 121, the user's learning history and learning conversation log stored in the user's learning history DB 123, and learning path recommendation. Based on the algorithm 130, etc., an utterance to be provided to the user terminal 10 can be generated.

In some embodiments, response generation module 112 may generate utterances based on an artificial intelligence model. For example, the response generation module 112 may use the intention of the user's utterance inferred by the natural language processing engine 121, the user's learning history and learning conversation log stored in the user's learning history DB 123, and learning path recommendation. It may be based on an artificial intelligence model that uses the algorithm 130 as its input and the response utterance as its output. The artificial intelligence model is in the form of a neural network including one or more layers and can be implemented as various neural network models such as RNN (Recurrent Neural Network).

In some embodiments, response generation module 112 may include various types of learning dialogue scenarios and utterances corresponding to each scenario. Alternatively, the learning dialogue processing engine 120 may further include a database for storing these scenarios and utterances corresponding to each scenario. The response generation module 112 may determine a scenario that matches the intention of the input user utterance based on the processing result of the natural language processing engine 121. The response generation module 112 may select an utterance corresponding to the determined scenario as a response, or may modify an utterance corresponding to the determined scenario and determine the modified utterance as a response. In some embodiments, the utterance corresponding to each scenario may be an utterance generated in the past based on an artificial intelligence model.

The learning history DB 123 can store the user's learning path, past learning history, and a conversation log between the AI automatic scoring tutorbot 100 and the user terminal 10. For example, the learning history DB 123 is based on scoring results provided to the user terminal 10 in the past, teaching material images received from the user terminal 10 in the past, and teaching material images received from the user terminal 10 in the past. The learning history of the user of the user terminal 10, the user's learning plan, the pre-built learning plan for each textbook, the utterances received from the user terminal 10, and the AI automatic scoring tutorbot 100 is used by the user. Utterances provided to the terminal 10 can be stored.

The learning path recommendation algorithm 130 provides a clinic (or solution) for incorrectly answered questions and may include an algorithm for recommending a learning path to the user. For example, the learning path recommendation algorithm 130 may include an algorithm for providing the user with a solution (explanation) for an item determined by the automatic scoring engine 110 that the user entered an incorrect answer. The learning path recommendation algorithm (130) is an algorithm for searching the textbook and content DB (113) for problems similar to the question determined by the automatic scoring engine (110) to have been entered by the user as an incorrect answer, and presenting similar problems found to the user. may include.

The learning path recommendation algorithm 130 may include an algorithm for storing the user's learning path. For example, the learning path recommendation algorithm 130 may include an algorithm for inferring which part of the textbook the user has learned based on the textbook image received from the user terminal 10 and storing the inference result. For example, the learning path recommendation algorithm 130 selects the page number, question number, and correctness of the textbook learned by the user based on the printed and cursive font recognized by the OCR engine 111 from the textbook image received from the user terminal 10. It may include an algorithm for storing whether or not. The learning path recommendation algorithm 130 may include an algorithm for inferring which part of the textbook the user has learned based on the page number, question number, and whether or not the textbook is correct. The learning path recommendation algorithm 130 may include an algorithm for storing the inference result of whether the user is learning or not.

The learning path recommendation algorithm 130 may include an algorithm for managing the user's learning. For example, the learning path recommendation algorithm 130 may include an algorithm for comparing a user's learning path and a learning recommendation roadmap. The learning path recommendation algorithm 130 may include an algorithm for recommending a subsequent learning method to the user based on a comparison result between the user's learning path and the learning recommendation roadmap. In some embodiments, the learning recommendation roadmap may include learning order, learning amount, and learning conversation scenario.

The learning path recommendation algorithm 130 may include an algorithm for recommending an optimal learning path to the user. For example, the learning path recommendation algorithm 130 models the user based on the user's learning history collected through the automatic scoring engine 110 and the conversation log with the user collected through the learning conversation processing engine 120. It may include an artificial intelligence training algorithm for

In some embodiments, the learning path recommendation algorithm 130 collects the user's learning history from the printed and cursive fonts recognized from the textbook image, stores the user's utterances received from the user terminal, and stores the collected learning history and stored It may include an algorithm that updates the user model modeled for the user based on the utterances.

The learning path recommendation algorithm 130 includes an algorithm for selecting a learning scenario determined to be optimal for the current user among various learning scenarios based on user modeling and suggesting a recommended learning path to the user based on the selected learning scenario. can do.

In some embodiments, automatic scoring engine 110, learning dialogue processing engine 120, and learning path recommendation algorithm 130 may be implemented in the form of a computer program or program code stored on a computer-readable storage medium. In these embodiments, AI automatic grading tutorbot 100, including automatic grading engine 110, learning dialogue processing engine 120, and learning path recommendation algorithm 130, may be executed by one or more processors.

FIG. 2 shows a flowchart between the AI automatic grading tutorbot 100 and the user terminal 10 according to some embodiments of the present disclosure.

1 and 2, the user of the user terminal 10 can capture a textbook image (201). For example, the user can take a photo of a textbook (or worksheet or study sheet) containing questions and corresponding answers through the user terminal 10.

The user terminal 10 can upload the photographed textbook image to the AI automatic grading tutorbot 100 (202). For example, the user terminal 10 may transmit a textbook image to the AI automatic grading tutorbot 100 through the UI provided by the AI automatic grading tutorbot 100. The textbook image transmitted from the user terminal 10 may be transmitted to the AI automatic grading tutorbot 100 through a wired/wireless network.

The AI automatic grading tutorbot (100) can recognize printed and cursive fonts from textbook images based on an AI OCR model (203). For example, the OCR engine 111 of the AI automatic grading tutorbot 100 can recognize printed and cursive fonts from the textbook image based on the AI OCR model by performing OCR processing on the textbook image. The operating method of the OCR engine 110 will be described in detail later with reference to FIG. 4.

The AI automatic grading tutorbot (100) can grade the user's answers in the textbook image (204). For example, the grading module 112 of the AI automatic grading tutorbot 100 may determine whether the answer entered by the user is incorrect based on the recognition result of the OCR engine 111.

The AI automatic scoring tutorbot 100 can output a UI to check the recognition result to the user terminal 10. For example, the AI automatic scoring tutorbot 100 may request the user terminal 10 to confirm and modify the recognition result of the OCR engine 111.

In some embodiments, the AI automatic grading tutorbot 100 uses the OCR engine 111 to recognize results, such as questions recognized from textbook images, answers recognized as written by the user, and scoring. The results can be provided to the user terminal 10 along with an utterance asking for the need for correction.

Using the UI of the AI automatic scoring tutorbot 100, the user terminal 10 can input a recognition result confirmation response (206). For example, the user terminal 10 may transmit a response (eg, an utterance containing such intention) to the AI automatic scoring tutor bot 100 confirming that there is no problem with the recognition result. For another example, the user terminal 10 may transmit a response to the AI automatic grading tutorbot 100 to correct the recognition result of the AI automatic grading tutorbot 100. In this example, the response transmitted from the user terminal 10 to the AI automatic scoring tutor bot 100 may include a corrected question number or a corrected user answer.

The AI automatic scoring tutorbot 100 may generate a scoring result utterance based on the scoring result (207) and output the generated scoring result utterance to the user terminal 10 (208). For example, the learning conversation processing engine 120 of the AI automatic grading tutorbot 100 responds if there is an incorrect answer in the answer entered by the user based on the automatic grading result by the automatic grading engine 110. A scoring result utterance can be created to provide a clinic. In this example, the scoring result utterance may include a solution to an incorrect answer, a problem similar to the incorrect problem, or a recommended follow-up learning method. Solutions to incorrect answers, problems similar to the incorrect problem, and subsequent recommended learning methods may be selected based on the learning path recommendation algorithm 130.

For another example, the learning conversation processing engine 120 of the AI automatic scoring tutorbot 100 generates an utterance that praises the user or encourages the user's further learning if there are no incorrect answers in the answer entered by the user. , this can be output to the user terminal 10.

Figure 3 shows a schematic diagram of the operation interface of the AI automatic grading tutorbot 100 according to some embodiments of the present disclosure. More specifically, FIG. 3 may show an example of a user interface of the AI automatic grading tutorbot 100.

Referring to Figures 1 to 3, a user can upload a textbook image through the user terminal 10 (301). AI automatic grading tutorbot (100) can receive uploaded textbook images. The AI automatic grading tutorbot 100 can recognize the page number, question number, and user's answer from the uploaded textbook image based on the AI OCR model. AI automatic grading tutorbot (100) can grade recognized answers. The AI automatic scoring tutor bot (100) outputs a UI to check the recognition result to the user terminal (10) (302). For example, the UI for checking the recognition result may include an image indicating whether or not each question is correct in the uploaded textbook image and an utterance inquiring about errors in the recognition result.

In the illustrated embodiment, a UI for checking the recognition result is provided as a combination of an image and a speech, but the present disclosure is not limited to this embodiment. For example, the UI for checking the recognition result may include an utterance in the form of a text message including the recognition result and grading result of the AI automatic grading tutorbot 100. The UI for checking the recognition result may include an utterance directly asking whether to modify the recognition result or scoring result.

The user can check the grading results of the AI automatic grading tutorbot (100) and input a confirmation response to the AI automatic grading tutorbot (100) (303). For example, the confirmation response may be a text message generated directly by the user.

The AI automatic scoring tutor bot 100 may generate a scoring result utterance in response to the user's confirmation response and output the scored result utterance to the user terminal 10 (304). In the illustrated embodiment, the AI automatic grading tutorbot 100 infers the user's intention not to modify the recognition result and grading result from the user's confirmation response, and based on this inference result, based on the existing grading result. You can create an utterance as a result of scoring. The scoring result utterance may include solutions to questions judged to be incorrect and recommendations for similar problems.

FIG. 4 illustrates a flowchart of a method by which the AI automatic grading tutorbot 100 recognizes printed and cursive fonts from textbook images according to some embodiments of the present disclosure.

Referring to FIGS. 1 to 4, the OCR engine 111 of the AI automatic grading tutorbot 100 can recognize the page number print font from the textbook image received from the user terminal 10 based on the AI OCR model ( 401). The OCR engine (111) can cut textbook images for each question (402). For example, the OCR engine 111 may divide a textbook image into one or more regions. One or more areas may each correspond to one or more questions included in the textbook image.

The OCR engine 111 can recognize the correct answer box print for each question based on the AI OCR model (403). For example, in a textbook image, each question may be printed in a square-shaped answer box. The OCR engine 111 recognizes the square-shaped print included in the first area corresponding to the first question among the plurality of divided areas, thereby recognizing the answer box print for the first question based on the AI OCR model. You can.

The OCR engine 111 can recognize the handwriting in the recognized answer box for each question based on the AI OCR model (404). For example, the OCR engine 111 may recognize handwriting written in the recognized answer box print for the first question based on an AI OCR model. The OCR engine 111 may determine the recognized handwriting as corresponding to the answer the user wrote to the first question.

Figure 5 shows a schematic diagram to explain how the AI automatic grading tutorbot 100 prints printed and cursive fonts from textbook images according to some embodiments of the present disclosure. The operating method of the OCR engine 111 shown in FIG. 4 is described in more detail with reference to FIG. 5.

Referring to FIGS. 1 to 5, the OCR engine 111 can recognize the page number print font 501 within the textbook image based on the AI OCR model. For example, the OCR engine 111 can scan the lower part of a textbook image and extract a number image from the images included in the scanned lower part, thereby recognizing the page number print font based on the AI OCR model. In the illustrated embodiment, the OCR engine 111 may extract a numeric image included in the lower left part of the textbook image. The OCR engine 111 may infer that the extracted numeric image corresponds to the integer “88”. The OCR engine 111 may determine the number image included in the lower left corner of the textbook image corresponding to the integer “88” as a printed page number within the textbook image. Accordingly, the AI automatic grading tutorbot 100 can recognize that the page number of the textbook image received from the user terminal 10 is “88.”

Next, the OCR engine 111 can segment the textbook image by question. For example, the OCR engine 111 may divide a textbook image into multiple regions, and each region may correspond to any one of the five questions listed on page “88” of the textbook to which the textbook image corresponds. there is. In the illustrated embodiment, the first area 502 may correspond to the first question, that is, question “01”, among the five questions listed on page “88” of the textbook.

And, the OCR engine 111 can recognize the answer box print for each question based on the AI OCR model. In some embodiments of the present disclosure, in a textbook for which the AI automatic grading tutorbot 100 can provide automatic grading, a rectangular answer box may be printed for each question. For example, in the illustrated embodiment, a first answer box 503 for the first question may be printed in the first area 502.

The shape and location of the answer box print are not limited to the illustrated embodiment. For example, in the illustrated embodiment, the answer box print font is provided in a square shape, but in some embodiments, the answer box print font has a shape that is distinct from the part where the question is printed, and can be identified by the OCR engine 111. It may be in the form of a symbol. For example, the answer box print may be circular or triangular in shape. As another example, the answer box typography could be an underscore symbol.

Through the use of answer box printing, the accuracy of recognition can be improved based on the AI OCR model of the answers entered by the user. In the illustrated embodiment, if the correct answer box typography is not present, OCR engine 111 may recognize a number other than the number “3” as a cursive letter written by the user as the correct answer. For example, the OCR engine 111 may incorrectly recognize numbers written in a mathematical formula written by the user to solve a problem as the correct answer written by the user. By first recognizing the answer box print font based on the AI OCR model and then recognizing the cursive text in the answer box print font based on the AI OCR model, the accuracy of recognition can be improved based on the AI OCR model of the correct answer written in the user's handwriting. there is.

In the illustrated embodiment, for the first question, the OCR engine 111 may recognize a square image corresponding to the first answer box print 503 from the first area 502 based on the AI OCR model. . The OCR engine 111 may determine the recognized square image as the first answer box print font 503.

In a similar way, the OCR engine 111 can recognize the answer box printouts for the remaining questions on page “88” of the textbook, respectively, based on the AI OCR model.

Thereafter, for the first question, the OCR engine 111 may recognize the cursive font 504 in the first answer box print font 503 based on the AI OCR model. For example, in the depicted embodiment, the OCR engine 111 generates a number from an image corresponding to the number “3” handwritten by the user (i.e., a handwritten “3”) within the first answer box typography 503. “3” can be recognized based on the AI OCR model.

As a result, based on the recognition results based on the printed AI OCR model of the OCR engine 111 and the recognition results based on the cursive AI OCR model, the AI automatic grading tutorbot 100 recognizes the text “01” on page “88” of the textbook. It may be determined that the user selected option “3” as the answer to question number 3.

In a similar way, the OCR engine 111 can recognize the cursive font in each answer box for the remaining questions on page “88” of the textbook based on the AI OCR model.

Afterwards, the scoring module 112 compares the recognition result with the information on the corresponding page stored in the textbook and content DB 113 based on the AI OCR model of the OCR engine 111, and determines the user's answer on page “88” of the textbook. You can grade.

With reference to FIGS. 2 to 5 , the operation method of the automatic grading engine 110 of the AI automatic grading tutorbot 100 has been described. However, the order of the operation method of the automatic scoring engine 110 is not limited to the described order. For example, the automatic grading engine 110 can recognize the page number print font from a textbook image based on an AI OCR model and retrieve information on the corresponding page from the textbook and content DB 113. The information on the page may include the number of questions included in the page, the number of each question, the correct answer to each question, and information about the solution to each question. Afterwards, the automatic scoring engine 110 can cut the textbook image for each question, recognize the answer box print fonts based on the AI OCR model, and recognize the cursive font in each answer box print font based on the AI OCR model. And, the automatic scoring engine 110 can score the user's answer by comparing/contrasting the recognition result, that is, the correct answer entered by the user with the information on the corresponding page retrieved from the textbook and content DB 113.

For another example, the automatic scoring engine 110 may output a UI for the user to check and modify the recognition result before performing automatic scoring based on the recognition result based on the AI OCR model. Thereafter, the automatic scoring engine 110 may perform automatic scoring based on the recognition result modified by the user.

In some embodiments, the OCR engine 111 may preprocess textbook images before performing recognition based on an AI OCR model. For example, the OCR engine 111 can adjust the sharpness, brightness, saturation, etc. of the textbook image by using an algorithm to improve the image quality of the textbook image uploaded by the user through the user terminal 10. Accordingly, even if the quality of the textbook image is low due to the user photographing the textbook in the living environment, the accuracy of the OCR engine 111 can be improved.

FIG. 6 shows a flowchart between the AI automatic grading tutorbot 100 and the user terminal 10 according to some embodiments of the present disclosure.

Referring to FIGS. 1 and 6 , the AI automatic grading tutorbot 100 may generate a first management utterance for managing the user's learning schedule (601). The AI automatic scoring tutorbot 100 can output the generated first management utterance to the user terminal 10.

For example, the AI automatic grading tutorbot 100 may generate a first management utterance to encourage the user's learning based on a preset learning schedule (or schedule, roadmap). In other words, in order to encourage the user's learning, the AI automatic scoring tutorbot 100 first generates a first management utterance without a preceding utterance of the user terminal 10, and then sends the first management utterance to the user terminal 10 ) can be output. Accordingly, the AI automatic grading tutor bot 100 may be recognized by the user as first having a learning consultation conversation with the user.

In response to the first management utterance, the user terminal 10 may input the first response utterance into the AI automatic grading tutorbot 100 (603).

The AI automatic scoring tutorbot 100 can infer the intention of the first response utterance (604). For example, the natural language processing engine 121 of the AI automatic scoring tutorbot 100 may analyze the content of the first response utterance and infer the user's intention inherent in the first response utterance.

The natural language processing engine 121 may extract one or more keywords from the first response utterance and infer the user's intention based on the one or more extracted keywords. In some embodiments, natural language processing engine 121 may infer the user's intent based further on previous learning conversation history.

The AI automatic scoring tutorbot 100 may generate a second management utterance (605) based on the inferred intention of the first response utterance and the learning path recommendation algorithm (130). The AI automatic scoring tutorbot 100 can output the generated second management utterance to the user terminal 10 (606). For example, the second management utterance may be in the form of asking the user a question to understand the detailed intention of the first response utterance, suggesting a follow-up plan to the user, or encouraging or comforting the user. , but is not limited to this.

In some embodiments, the AI automatic grading tutorbot 100 may infer the intention of ending the conversation from the first response utterance. In these embodiments, the AI automatic grading tutorbot 100 may end the conversation without generating and outputting a second management utterance.

In some embodiments, the AI automatic grading tutorbot 100 may not be able to infer user intent from the first response utterance. For example, learning conversation processing engine 120 may determine that inference of user intent has failed in response to failing to extract a keyword from the first response utterance. For another example, the learning conversation processing engine 120 may determine that the inference of the user's intention has failed if the probability that the user's intention inferred based on the keywords extracted from the first response utterance is accurate is lower than a preset threshold. You can.

Based on determining that the inference of the user intention has failed, the learning conversation processing engine 120 may call a fallback block and provide the fallback block to the user terminal 10 instead of the second management utterance. The fallback block may include various response messages tailored to the user's eye level based on the user's age, etc. For example, the fallback block may include a message that induces the user to re-enter the utterance. As another example, a fallback block may contain one or more messages with intent that the user is likely to enter based on the user's learning history, etc. (e.g., “I'm done, but I left the book somewhere else,” “Today.” Buttons for selecting one of the following (“I’ve already solved all the parts.”, “I understand. I’ll solve it now.”, etc.) can be provided to the user terminal (10).

In some embodiments, the learning conversation processing engine 120 of the AI automatic grading tutorbot 100 may be trained in advance on the speech habits of the user of the AI automatic grading tutorbot 100. For example, the learning conversation processing engine 120 may be trained in advance based on natural language conversation data of elementary school students and middle school students.

In some embodiments, the AI automatic grading tutorbot 100 may recognize various speech scenarios in advance. For example, the learning conversation processing engine 120 of the AI automatic grading tutorbot 100 may be trained in advance based on learning conversation scenarios. The learning dialogue processing engine 120 may select one of the pre-trained learning dialogue scenarios based on the inferred intention of the first response utterance. Learning dialogue processing engine 120 may generate a second management utterance based on the selected learning dialogue scenario.

FIG. 7 shows a schematic diagram of the operation interface of the AI automatic grading tutorbot 100 according to some embodiments of the present disclosure.

Referring to FIGS. 1, 6, and 7, the AI automatic grading tutorbot 100 generates a management utterance 701 for managing the user's learning schedule, and sends the management utterance 701 to the user terminal 10. ) can be output. The management utterance 701 may include the user's name and information about the amount of textbooks the user should learn at the time of output of the management utterance 701 according to the user's learning schedule.

The AI automatic scoring tutorbot 100 may receive a response utterance 702 corresponding to the management utterance 701 from the user terminal 10. The AI automatic scoring tutorbot 100 infers the user's intention inherent in the response utterance 702, generates a management utterance 703 corresponding to the response utterance 702 based on the inference result, and manages the management utterance 703. ) can be output to the user terminal (10).

For example, the natural language processing engine 121 of the AI automatic scoring tutorbot 100 extracts one or more keywords from the response utterance 702 and determines the intent of the response utterance 702 based on the one or more extracted keywords. can be inferred. The natural language processing engine 121 may infer the user's emotion or learning state inherent in the response utterance 702, or an unexpected situation related to learning, from one or more extracted keywords. For example, the natural language processing engine 121 may select a learning conversation scenario corresponding to one or more keywords extracted from various pre-trained learning conversation scenarios. The natural language processing engine 121 may infer the intent of the response utterance 702 based on the selected learning conversation scenario. Based on the inference results of the natural language processing engine 121, the response generation module 122 may generate the management utterance 703.

In the illustrated embodiment, the natural language processing engine 121 may extract keywords such as “book,” “I left it behind,” etc. from the response utterance 702. The natural language processing engine 121 can infer from the extracted keywords the intention of the response utterance 702 that the user does not possess the textbook at the current time. Based on the inference results, the AI automatic grading tutor bot 100 recognizes the need for coordination of the learning schedule and can adjust the stored learning schedule through the learning recommendation path algorithm 130. For example, the AI automatic grading tutorbot 100 can adjust the learning schedule so that today's learning amount is included in tomorrow's learning amount through the learning recommendation path algorithm 130. The response generation module 122 may generate a management utterance 703 for managing the learning schedule by delaying the learning schedule to a future time, and output the generated management utterance 703 to the user terminal 710.

In the illustrated embodiment, the user may transmit a response utterance 704 corresponding to the management utterance 703 to the AI automatic grading tutorbot 100. The natural language processing engine 121 can extract the keyword “yes” from the response utterance 704. Natural language processing engine 121 further bases the keywords extracted from response utterance 704 and previous learning conversation history, such as response utterance 702 and management utterances 701, 703, into response utterance 704. The intention to end the conversation inherent in can be inferred. Accordingly, the AI automatic grading tutorbot 100 can end the conversation without generating and outputting a response to the response utterance 704.

Based on adjusting the learning schedule in response to the response utterance 702, the AI automatic grading tutorbot 100 generates a management utterance 705 for managing the user's learning schedule, and sends it to the user terminal 10. Management utterance 705 can be output. The management utterance 705 may include the intention to encourage learning not only for the amount of learning at the time of creation of the management utterance 705, but also for the amount of learning at the time of creation of the management utterance 701, based on the coordinated learning schedule. there is.

In the illustrated embodiment, the management speech 705 may be output to the user terminal 10 at the same time as the management speech 701. For example, the AI automatic scoring tutorbot 100 may generate management utterances for managing the learning schedule at a preset time or a time set by the user and output the generated management utterances to the user terminal 10. . In other words, the AI automatic scoring tutor bot 100 can periodically generate management utterances and output them to the user terminal 10. In the illustrated embodiment, the management utterance generation cycle of the AI automatic scoring tutorbot 100 is 24 hours, but the present disclosure is not limited to this. The management utterance generation cycle of the AI automatic scoring tutorbot 100 may be set in advance or adjusted by the user.

FIG. 8 shows a schematic diagram of the operation interface of the AI automatic grading tutorbot 100 according to some embodiments of the present disclosure.

Referring to Figures 1 and 6 to 8, Figure 7 may illustrate an embodiment in which a conversation is initiated by the AI automatic grading tutorbot 100, and in contrast, Figure 8 shows an example in which a conversation is initiated by a user. An example can be shown.

More specifically, in the embodiment shown in FIG. 8, the AI automatic grading tutorbot 100 may receive the first utterance 801 from the user. The AI automatic scoring tutorbot 100 can infer the user's intention inherent in the first utterance 801. For example, the natural language processing engine 121 of the AI automatic grading tutorbot 100 can extract keywords such as “study” and “I didn’t get it” from the first utterance 801. The natural language processing engine 121 may infer the user's intention that the first utterance 801 did not learn well today based on the extracted keywords.

In response to the first utterance 801, the AI automatic grading tutorbot 100 may generate a second utterance 802 and output the second utterance 802 to the user terminal 10. In the illustrated embodiment, the AI automatic grading tutorbot 100 may generate a question to induce the user's detailed intention as the second utterance 802. For example, the AI automatic scoring tutor bot 100 asks questions about why learning was not done well in order to analyze the reason for not learning well about the intention of the first utterance that learning was not done well today. It is possible to generate a second utterance 802 including.

The AI automatic grading tutorbot 100 can receive the third utterance 803 corresponding to the second utterance 802. The AI automatic scoring tutorbot 100 can infer the user's intention inherent in the third utterance 803. For example, the natural language processing engine 121 of the AI automatic scoring tutorbot 100 can extract keywords such as “I don’t know” and “there were a lot” from the third utterance 803. The natural language processing engine 121 can infer the intention of the third utterance 803 that the learning content was not well understood based on the extracted keywords.

In response to the third utterance 803, the AI automatic grading tutorbot 100 may generate a fourth utterance 804, and output the generated fourth utterance 804 to the user terminal 10. The AI automatic scoring tutorbot 100 selects a follow-up processing plan to provide to the user based on the intention of the inferred third utterance 803 and the learning path recommendation algorithm 130, and selects an utterance corresponding to the selected follow-up processing plan. can be produced as the fourth utterance 804.

In some embodiments, follow-up measures may include encouraging learning, providing additional learning methods, encouraging using previous grading results, and coordinating learning schedules. The learning path recommendation algorithm 130 may include an algorithm that selects one of the plurality of subsequent processing methods based on the inferred intention of the third utterance 803.

In the illustrated embodiment, the AI automatic grading tutorbot 100 uses previous grading results to encourage the user and add additional information based on the intent of the inferred third utterance 803 and the learning path recommendation algorithm 130. Providing a learning method can be selected as a follow-up method. The AI automatic scoring tutor bot 100 may decide to provide review as an additional learning method based on the user modeling and learning history of the learning path recommendation algorithm 130. The AI automatic scoring tutorbot 100 can generate a fourth utterance 804 corresponding to the selected follow-up processing plan.

FIG. 9 shows a flowchart of a method of building the OCR engine 111 of the AI automatic grading tutorbot 100 according to some embodiments of the present disclosure.

In some embodiments, the method shown in FIG. 9 may be performed in advance by an external computing device before the OCR engine 111 is mounted on the AI automatic grading tutorbot 100. In some embodiments, the method shown in FIG. 9 may be performed by the AI automatic grading tutorbot 100.

First, textbook data and student-solved question paper data can be collected (901). For example, image data of a textbook (or original image data of a question paper) and image data of a question paper in which a student (or learner, user, etc.) writes answers by hand may be collected as training data.

Next, for each collected image data, the image area subject to recognition may be labeled (902). For example, in textbook image data, the area at the bottom of the textbook for recognizing page number print, image areas corresponding to each question, the area where the answer box print font is located in one image area, and the cursive text in the answer box print font. The located area, etc. may be labeled. As another example, in image data of a question paper in which a student writes answers by hand, the cursive area within the answer box print may be labeled.

Based on the labeled image, an AI model (e.g., AI OCR model) of OCR engine 111 may be fitted (903). For example, the AI model of the OCR engine 111 can be trained based on a supervised learning algorithm, using labeled image data as training data.

In some embodiments, the AI model of the OCR engine 111 may be built as any one or a combination of various neural networks that have an image dataset as input data. For example, the AI model of the OCR engine 111 may be implemented as a combination of a neural network model that detects the location of text from a document with high text density by learning a handwriting OCR dataset and a neural network model that recognizes the detected text. . In other words, the AI model of the OCR engine 111 can be implemented as a combination of a neural network model for detecting text and a model for recognizing the detected text. In some embodiments, the AI model of the OCR engine 111 may be implemented as neural network models such as CRAFT (Character Region Awareness for Text Detector), Swin-Transformer, HRCenterNet, ResNet, etc., or a combination thereof.

The AI model of the fitted OCR engine 111 may be updated based on image data newly collected from the user according to the operation of the AI automatic scoring tutorbot 100 (904). For example, referring to FIG. 2, the AI automatic grading tutorbot 100 may request feedback (confirmation response) on the recognition result of the OCR engine 111 from the user (205). The AI automatic scoring tutorbot 100 converts the AI model of the fitted OCR engine 111 into the OCR engine based on feedback provided from the user (e.g., recognition result confirmation response input in response to the UI at 206). It can be retrained using the recognition results of (111).

In some embodiments, updating 904 of the AI model of the fitted OCR engine 111 may be performed periodically by the AI automatic grading tutorbot 100. For example, the AI automatic grading tutorbot 100 receives a textbook image with answers from the user and performs automatic grading whenever the number of times exceeds a certain number of times (for example, each time automatic grading is performed). The AI model of the OCR engine 111 can be updated. For another example, the AI automatic grading tutorbot 100 may periodically update the AI model of the OCR engine 111 at a preset interval (e.g., once a week, etc.).

FIG. 10 shows a flowchart of a method for building the natural language processing engine 121 of the AI automatic grading tutorbot 100 according to some embodiments of the present disclosure.

In some embodiments, the method shown in FIG. 10 may be performed in advance by an external computing device before the natural language processing engine 121 is mounted on the AI automatic grading tutorbot 100. In some embodiments, the method shown in FIG. 10 may be performed by the AI automatic grading tutorbot 100.

First, learning phrases and response data for each user conversation intention can be collected (1001). For example, conversation logs between students or between students and teachers can be collected as training data.

Next, for each collected conversation data, keywords and intentions to be recognized may be labeled (1002). For example, for each collected utterance, one or more keywords that the natural language processing engine 121 must recognize may be extracted, and intentions corresponding to the extracted keywords may be labeled.

Based on the labeled conversation data, an AI model (e.g., a natural language processing AI model) of the natural language processing engine 121 may be fitted (1003). For example, the AI model of the natural language processing engine 121 can be trained (or learned) based on a supervised learning algorithm, using an utterance dataset containing each utterance and the keywords and intent labeled thereto as training data. there is.

The AI model of the natural language processing engine 121 may be updated based on conversation data newly collected from the user according to the operation of the AI automatic grading tutorbot 100 (1004). For example, a conversation log between the AI automatic grading tutorbot 100 and the user terminal 10, for example, the AI automatic grading tutorbot 100 and the user terminal 10 shown in FIGS. 3, 7, and 8. ) Based on the conversation log between the AI models of the natural language processing engine 121, the AI model can be retrained.

In some embodiments, the update 1004 of the AI model of the natural language processing engine 121 may be periodically performed by the AI automatic grading tutorbot 100. For example, the AI automatic scoring tutorbot 100 may update the AI model of the natural language processing engine 121 whenever the number of utterances input from the user exceeds a certain number. For another example, the AI automatic grading tutorbot 100 may periodically update the AI model of the OCR engine 111 at a preset interval (e.g., once a week, etc.).

FIG. 11 shows a flowchart of a method for building a learning recommendation path algorithm 130 of the AI automatic grading tutorbot 100 according to some embodiments of the present disclosure.

In some embodiments, some of the methods shown in FIG. 11 may be performed in advance by an external computing device before the learning recommendation path algorithm 130 is mounted on the AI automatic grading tutorbot 100.

In some embodiments, the method shown in FIG. 11 may be performed by the AI automatic grading tutorbot 100.

First, user modeling may be performed based on the user learning history and learning conversation log (1101). For example, for each user, the user's previous learning history data and speech data between the user and other students or between the user and the teacher may be collected as modeling data. Based on the modeling data, a user model (or user profile) can be built for each user. Each user model may include information about the user's learning speed, learning tendency, degree of deviation from the learning path, learning amount completion rate, correct answer rate for problem solving, degree of fit with the learning roadmap, etc.

Next, a learning recommendation rule can be designed for each user model (1102). For example, for each user model, the level of detail of the solution to be provided for incorrectly answered questions, the average amount of learning, the learning order, and the tendency of learning conversation scenarios to be provided to users may be determined.

Thereafter, each user of the AI automatic grading tutorbot 100 may be provided with incorrect answer clinics, learning management utterances, and subsequent learning course recommendations based on the corresponding user model for which learning recommendation rules are designed. The AI automatic scoring tutorbot (100) uses the learning history collected through the OCR engine (111), the user's conversation log collected through the natural language processing engine (121), and the information provided to each user from the AI automatic scoring tutorbot (100). The user's reaction in response to the incorrect answer clinic, learning management speech, and subsequent learning course recommendation, and the user's subsequent learning history can be tracked, and the user's reaction and learning history tracking results can be collected as user data. And the learning history tracking results may include whether the user's learning achievement has improved following the service provided by the AI automatic scoring tutor bot 100, the amount of increase or decrease in the amount of learning, and whether deviation from the learning path has been reduced.

The AI automatic scoring tutorbot 100 can fit the AI model of the learning path recommendation algorithm 130 based on the collected user data (1103). For example, the AI model of the learning path recommendation algorithm 130 may label each user model and the corresponding user data, and train the AI model of the learning path recommendation algorithm 130 based on the labeled data. The AI model of the learning path recommendation algorithm 130 may be based on RNN, etc.

The AI model of the fitted learning path recommendation algorithm 130 may be updated based on data newly collected from the user according to the operation of the AI automatic grading tutorbot 100 (1104). For example, the AI automatic scoring tutorbot 100 can retrain the AI model of the learning path recommendation algorithm 130 using the learning history and conversation log collected from each user.

In some embodiments, the update 1104 of the AI model of the learning path recommendation algorithm 130 may be performed periodically by the AI automatic grading tutorbot 100. For example, the AI automatic grading tutorbot 100 receives a textbook image with answers from the user and performs automatic grading whenever the number of times exceeds a certain number of times (for example, each time automatic grading is performed). The AI model of the learning path recommendation algorithm 130 can be updated. For another example, the AI automatic grading tutorbot 100 may periodically update the AI model of the learning path recommendation algorithm 130 at a preset interval (e.g., once a week, etc.). For another example, the AI automatic scoring tutorbot 100 may update the AI model of the learning path recommendation algorithm 130 whenever the number of utterances input from the user exceeds a certain number.

12 shows a block diagram of a communication system according to some embodiments of the present disclosure.

Referring to FIG. 12 , the communication system may include a user terminal 1210, a network 1220, and a server 1230.

In some embodiments, a communication system may include one or more user terminals (e.g., a sending terminal and a receiving terminal), one or more servers (e.g., an intermediate server and a client server), and one or more networks. there is. In these embodiments, each user terminal and each server may be connected through different networks.

The user terminal 1210 may transmit or receive a communication signal from the server 1230 via the network device 1220. In some embodiments, an application may be installed on the user terminal 1210. For example, the user terminal 1210 may be installed with applications including various application program interfaces (APIs) or embedded software to provide an automatic grading and learning management platform. These applications may run on one or more processors within user terminal 1210.

The user terminal 1210 can transmit or receive data or signals to the server 1230 through an installed application. In some embodiments, user terminal 1210 may include a smartphone, tablet computer, wearable device, laptop computer, etc.

The network 1220 may include a wired or wireless data communication network for transmitting and receiving data such as text data, image data, or video data between the user terminal 1210 and the server 1230.

In some embodiments, the network 1220 may be an optical cable-based wired communication system that provides data transmission and reception services using optical cables.

In some embodiments, the network 1220 is an IP (Internet Protocol) network that provides large data transmission and reception services through the Internet Protocol (IP), or an All IP network that integrates different IP networks. You can.

In some embodiments, the network 1220 is a mobile communication network including a wired network and a Wibro (Wireless Broadband) network, a mobile communication network including WCDMA, a High Speed Downlink Packet Access (HSDPA) network, and a Long Term Evolution (LTE) network. It may be one of a mobile communication network including LTE advanced (LTE-A), a satellite communication network, and a Wi-Fi network, or a combination of one or more of them.

The server 1230 may be a computing device capable of transmitting and receiving data to and from the user terminal 1210 through the network 1220. In some embodiments, the server 1230 may be implemented as a computing device capable of transmitting and receiving data, such as a personal computer (PC), a laptop computer, or a tablet personal computer (Tablet PC).

In some embodiments, server 1230 is a cloud server, an IP Multimedia Subsystem (IMS) server, a Telephony Application server, an Instant Messaging (IM) server, a Media Gateway Control Function (MGCF) server, and an MSG. (Messaging Gateway) server and CSCF (Call Session Control Function) server.

Server 1230 may include a processor 1231, memory 1232, and communication device 1233.

The processor 1231 may perform a function as a central cleaning device of the server 1230. The processor 1231 may control the operation of the server 1230 by executing software, firmware, program codes, or instructions loaded into the memory 1232.

In some embodiments, processor 1231 may drive AI automatic grading tutorbot 100 of FIG. 1 by executing program code stored in memory 1232. In other words, the AI automatic grading tutorbot 100 of FIG. 1 may be implemented as program code, instructions, or a mobile application program that is loaded into the memory 1232 and executed by the processor 1231.

In some embodiments, processor 1231 may be a collection of one or more processors. In some embodiments, processor 1231 may include one or more cores.

Memory 1232 may store data, software, firmware, program codes, or instructions to be processed or processed by processor 1231. For example, software, firmware, program codes, or instructions for controlling the server 1230 by the processor 1231 may be loaded into the memory 1232. The memory 1232 may include Dynamic Random Access Memory (DRAM) or Static RAM (SRAM).

In some embodiments, memory 1232 may be a non-transitory computer-readable medium that stores program code of AI automatic grading tutorbot 100, executable by processor 1231. The number of memories 1232 may be one or more.

The communication device 1233 can communicate with a device external to the server 1230. For example, under the control of processor 1232, communication device 1233 transmits data to user terminal 1210 via network 1220, based on various wired or wireless protocols, and ) can receive data from.

In some embodiments, the server 1230 connects the user terminal 1210 and the network 1220 to perform automatic grading through an artificial intelligence chatbot (e.g., AI automatic grading tutorbot 100 in FIG. 1). It may be a server that communicates through.

The sequence of processing described in the method or process disclosed in this disclosure is described as an example. Therefore, the order of each step can be adjusted within the spirit of the present invention as needed. Additionally, the devices and systems disclosed in this specification may include means capable of performing the functions described in this specification, and, if necessary, may be implemented as independent devices or systems, or may exist in a linked or integrated form with other systems. .

The techniques described in this disclosure may be implemented, at least in part, in hardware, software, firmware, or any combination thereof. They may be implemented, for example, as one or more processors, DSP, ASIC, FPGA, or equivalent integrated or discrete logic circuitry, or any combination of at least one or more of these. Such hardware, software, and firmware may be implemented within one or multiple systems or devices to support the processes and functions disclosed herein, or may be implemented in a linked or integrated form with other systems or devices. . Additionally, the components described herein may be implemented separately or together with separate but interoperable logic devices. Each function and process separately described in this specification is only described as such to emphasize each function, and such functions do not have to be realized in separate hardware, firmware, or software components, and do not need to be implemented in common or It may also be integrated within a combination of separate hardware and/or software. The term “processor” or “processing circuit” may generally refer to any of the above-described logic circuits, alone or in combination with other logic circuits, or any other equivalent circuit. A control unit comprising hardware may perform one or more of the techniques of the present invention.

Those skilled in the art will understand that the various illustrative logical blocks, modules, processors, means, circuits and algorithm steps described in connection with the embodiments disclosed herein may be used in electronic hardware, (for convenience) It will be understood that the implementation may be implemented by various forms of program or design code (referred to herein as “software”) or a combination of both. To clearly illustrate this interoperability of hardware and software, various illustrative components, blocks, modules, circuits and steps have been described above generally with respect to their functionality. Whether this functionality is implemented as hardware or software depends on the specific application and design constraints imposed on the overall system. A person skilled in the art of this disclosure may implement the described functionality in various ways for each specific application, but such implementation decisions should not be construed as departing from the scope of this disclosure.

The various embodiments presented in this disclosure may be implemented as a method, device, or article of manufacture using standard programming and/or engineering techniques. The term “article of manufacture” includes a computer program, carrier, or media accessible from any computer-readable device. Here, the medium may include a storage medium and a transmission medium. For example, computer-readable storage media include magnetic storage devices (e.g., hard disks, floppy disks, magnetic strips, etc.), optical disks (e.g., CDs, DVDs, etc.), smart cards, and flash. Includes, but is not limited to, memory devices (e.g., EEPROM, cards, sticks, key drives, etc.). Additionally, various storage media presented herein include one or more devices and/or other machine-readable media for storing information. Transmission media also includes, but is not limited to, wireless channels and various other media capable of conveying instruction(s) and/or data.

Embodiments described in this disclosure can also be practiced in distributed computing environments where certain tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote memory storage devices.

Additionally, the techniques described in this disclosure may also be implemented or stored in a computer-readable storage medium containing instructions. And instructions stored in a computer-readable medium can cause methods and processing related to the instructions to be performed by a processor in each device. Computer-readable storage media may include RAM, ROM, PROM, EPROM, EEPROM, flash memory, hard disk, CD-ROM, magnetic media, optical media, or other storage media.

Although embodiments of the present invention have been described above with reference to the attached drawings, those skilled in the art will understand that the present invention can be implemented in other specific forms without changing its technical idea or essential features. You will be able to understand it. Therefore, the embodiments described above should be understood in all respects as illustrative and not restrictive.

100: Artificial Intelligence (AI) automatic grading tutorbot
110: Automatic scoring engine
120: Learning conversation processing engine
130: Learning path recommendation algorithm

Claims (17)

In a server that communicates with a user terminal through a network to perform automatic scoring through an artificial intelligence (AI) chatbot,
A processor including one or more cores; and
contains memory,
The processor:
An automatic grading engine that provides automatic grading based on an AI optical character recognition (OCR) model for textbook images received from the user terminal;
A learning conversation processing engine that provides a learning consultation conversation based on an AI natural language processing model to the user terminal; and
A server that executes a learning path recommendation algorithm based on a user model modeled for the user based on the learning history of the user of the user terminal and the conversation log with the user. According to claim 1,
The automatic grading engine includes an OCR engine that recognizes printed and cursive fonts from the textbook image based on an AI optical character recognition (OCR) model;
a scoring module that scores the answer of the user of the user terminal described in the textbook image based on the recognized printed font and the cursive font; and
A server including a textbook and content database (DB) that stores the questions included in the textbook image and the correct answers and solution data for each question. According to claim 2,
When executed by the processor, using the AI OCR model, the OCR engine:
From the textbook image, recognize the page number print font;
dividing the textbook image into one or more regions, wherein the one or more regions each correspond to one or more questions included in the textbook image;
recognize a first answer box printout from a first area corresponding to a first question among the one or more areas;
recognize the cursive written in the recognized first answer box print; and
A server that determines the recognized handwriting as corresponding to an answer written by the user to the first question. According to claim 2,
When executed by the processor, the scoring module:
searching the textbook and content DB for questions corresponding to the printed text recognized from the textbook image and correct answers and solutions to the questions;
A server that compares the searched correct answer with the user's answer corresponding to the cursive recognized from the textbook image. According to claim 2,
When executed by the processor, the automatic scoring engine outputs a user interface (UI) for providing recognition results of the OCR engine to the user terminal,
Singi UI is a server that includes a query about whether to modify the recognition result. According to claim 5,
When executed by the processor, the automatic scoring engine:
In response to the UI, receive a recognition result modified by the user from the user terminal; and
A server that re-scores the user's answer based on the modified recognition result. According to claim 5,
When executed by the processor, the automatic grading engine is configured to: the printed font and the cursive font recognized from the textbook image, the result of the automatic grading, and the UI for confirming with the user the recognition result of the AI OCR engine. A server that updates the AI OCR model based on the user's response. According to claim 1,
When executed by the processor, the learning dialogue processing engine, in response to the user's answer described in the textbook image being scored as an incorrect answer, provides a solution for the incorrect answer, a problem similar to the incorrect problem, or a subsequent recommended learning method. Generate a scoring result utterance containing at least one of
A server wherein the solution to the incorrect answer, a problem similar to the incorrect problem, and the subsequent recommended learning method are selected based on a learning path recommendation algorithm. According to claim 1,
When executed by the processor, the learning conversation processing engine:
a natural language processing engine that analyzes the first utterance received from the user terminal based on an AI natural language processing model;
a response generation module for generating a second utterance to be output to the user terminal based on the analysis result of the natural language processing engine; and
A server comprising a learning history DB that stores the user's learning history and a learning conversation log with the user. According to clause 9,
When executed by the processor, based on the AI natural language processing model, the natural language processing engine:
extract one or more keywords from the first utterance; and
A server that analyzes the content of the first utterance from the one or more extracted keywords. According to clause 9,
When executed by the processor, based on the AI natural language processing model, the response generation module includes the content of the first utterance included in the analysis result of the natural language processing engine, the user's learning history stored in the learning history DB, and A server that generates the second utterance based on a learning conversation log with the user and the learning path recommendation algorithm. According to claim 11,
When executed by the processor, the response generation module is based on the content of the first utterance, the learning history of the user stored in the learning history DB and the learning conversation log with the user, and the learning path recommendation algorithm, determine a learning conversation scenario matching the content of the first utterance; and
A server generating the second utterance to correspond to the determined scenario. According to clause 9,
When executed by said processor:
The natural language processing engine receives a third utterance corresponding to the second utterance from the user, analyzes the content of the received third utterance, and determines to end the conversation based on the analyzed content of the third utterance. do, and
The response generation module determines to end the conversation without generating an utterance corresponding to the third utterance. According to claim 1,
When executed by the processor, the learning conversation processing engine:
generate a first utterance to manage the user's learning schedule;
Initiating the learning consultation conversation with the user by outputting the first utterance to the user terminal;
receive a second utterance from the user in response to the first management utterance;
Analyze the content of the second utterance based on an AI natural language processing model; and
Outputting a third utterance corresponding to the content of the analyzed second utterance to the user terminal,
The first utterance includes an utterance to encourage learning of the user based on the user's learning schedule. According to claim 14,
The learning conversation processing engine updates the natural language processing AI model based on the first to third utterances. According to claim 14,
The third utterance includes a subsequent processing plan selected based on the analyzed content of the second utterance and the learning path recommendation algorithm,
The learning path recommendation algorithm includes an algorithm for selecting one of a plurality of follow-up processing methods based on the user's learning history, conversation log, learning schedule, etc.,
The plurality of follow-up processing methods include encouraging learning, providing additional learning methods, encouraging using previous grading results, and coordinating a learning schedule. A computer-readable medium storing a computer program for performing automatic scoring through a chatbot, wherein the computer program, when executed by a processor, causes the processor to:
Provide automatic scoring for textbook images received from the user terminal based on an artificial intelligence (AI) optical character recognition (OCR) model;
providing a learning consultation conversation based on an AI natural language processing model to the user terminal; and
A computer-readable medium that causes the user terminal to execute a learning path recommendation algorithm based on a user model modeled for the user based on the user's learning history and conversation logs with the user.

Images