Method and device for generating question solving model for solving coordinate system question

Abstract

The embodiment of the invention provides a method and device for generating a question solving model used for solving a coordinate system question, and the method comprises the steps: obtaining the solving information of a plurality of questions to be solved obtained through a large language model; the to-be-answered questions are questions related to the coordinate system; the answering information comprises video data used for generating an answering video of the to-be-answered question; obtaining training data based on the plurality of questions to be answered and the corresponding answering information containing the video data; training a problem solving model by using the training data to obtain a trained problem solving model; the trained problem solving model can generate video data for explaining a problem solving video about a coordinate system problem. According to the embodiment of the invention, the video data of the question solving video of the question related to the coordinate system can be quickly generated by using the trained question solving model, so that the question solving video can be quickly obtained, the condition of manually compiling the video data is avoided, and the generation efficiency of the question solving video is improved.

Description

Technical Field

[0001] The embodiments of the present specification relate to the field of computer technology, and more particularly to a method, apparatus, device and medium for generating a problem-solving model for solving coordinate system problems. Background Art

[0002] With the development of artificial intelligence and big data technology, the use of computer technology to assist teaching has become a trend. In particular, using animations and videos to explain topics can greatly improve students' learning interest and comprehension ability. However, most of the existing teaching videos are manually produced and have low work efficiency.

[0003] Therefore, it is necessary to provide a method for quickly generating problem-solving videos. Summary of the invention

[0004] In view of this, an embodiment of this specification provides a method for generating a problem-solving model for solving a coordinate system problem. One or more embodiments of this specification also relate to a device for generating a problem-solving model for solving a coordinate system problem, a computing device, a computer-readable storage medium, and a computer program to solve the technical defects existing in the prior art.

[0005] According to a first aspect of an embodiment of this specification, there is provided a method for generating a problem-solving model for solving a coordinate system problem, comprising: Obtaining answer information of a plurality of unanswered questions obtained by using a large language model; the unanswered questions are questions related to a coordinate system; the answer information includes video data for generating a video of answering the unanswered questions; Based on the plurality of questions to be answered and the corresponding answer information including the video data, obtaining training data; The training data is used to train a problem-solving model to obtain a trained problem-solving model; the trained problem-solving model can generate video data for explaining a problem-solving video about a coordinate system problem.

[0006] According to a second aspect of an embodiment of this specification, there is provided a device for generating a problem-solving model for solving a coordinate system problem, comprising: The answer information acquisition module is configured to acquire answer information of a plurality of unanswered questions obtained by using a large language model; the unanswered questions are questions related to a coordinate system; the answer information includes video data for generating a video of solving the unanswered questions; A training data acquisition module is configured to obtain training data based on the plurality of questions to be answered and the corresponding answer information including the video data; The model training module is configured to train the problem-solving model using the training data to obtain a trained problem-solving model; the trained problem-solving model can generate video data for explaining a problem-solving video about coordinate system problems.

[0007] According to a third aspect of an embodiment of this specification, a computing device is provided, including: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the above method are implemented.

[0008] According to a fourth aspect of the embodiments of this specification, a computer-readable storage medium is provided, which stores computer-executable instructions, and the instructions implement the steps of the above method when executed by a processor.

[0009] According to a fifth aspect of the embodiments of this specification, a computer product is provided, including a computer program / instruction, which implements the steps of the above method when executed by a processor.

[0010] An embodiment of the present specification achieves the following beneficial effects: obtaining answer information of several unanswered questions obtained by using a large language model, wherein the unanswered questions are questions related to the coordinate system, and the answer information includes video data for generating a problem-solving video of the unanswered questions; obtaining training data based on several unanswered questions and the corresponding answer information containing video data; using the training data to train the problem-solving model to obtain a trained problem-solving model. Thus, the trained problem-solving model can be used to generate video data for a problem-solving video explaining a problem about a coordinate system, and then the problem-solving video can be obtained more quickly, avoiding the situation of manually writing video data, reducing manual operations, and thus improving the generation efficiency of the problem-solving video. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 is a flow chart of a method for generating a problem-solving model for solving a coordinate system problem provided by an embodiment of the present specification; Figure 2 is a schematic diagram of an image frame of a problem-solving video provided by an embodiment of the present specification; Figure 3 is a schematic diagram of an image frame of another problem-solving video provided by an embodiment of the present specification; Figure 4 It is a schematic diagram of a process of generating a problem-solving model for solving a coordinate system problem provided by an embodiment of the present specification; Figure 5 It is a structural schematic diagram of a device for generating a problem-solving model for solving a coordinate system problem provided by an embodiment of the present specification; Figure 6 It is a structural block diagram of a computing device provided by an embodiment of this specification. DETAILED DESCRIPTION

[0012] Many specific details are described in the following description to facilitate a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the connotation of this specification, so this specification is not limited to the specific implementation disclosed below.

[0013] The terms used in one or more embodiments of this specification are only for the purpose of describing specific embodiments, and are not intended to limit one or more embodiments of this specification. The singular forms of "a", "said" and "the" used in one or more embodiments of this specification and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0014] It should be understood that although the terms first, second, etc. may be used to describe various information in one or more embodiments of this specification, this information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of this specification, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0015] In addition, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0016] First, the terms involved in one or more embodiments of this specification are explained.

[0017] Large Language Model (LLM): Also known as a large language model or large model, it is a deep learning model that uses autoregression to pre-train on a large corpus in the field of natural language processing. It can understand and generate natural language text. The large language model can predict the next word or sentence by learning the statistical laws and semantic information of natural language text. As the input data set and parameter space continue to expand, the capabilities of the large language model will also expand accordingly. It can be applied to a variety of fields such as machine learning, machine translation, speech recognition, and image processing.

[0018] Questions related to coordinate systems: Questions related to coordinate systems are a classic type of questions in elementary school mathematics. These questions refer to math questions that are solved using coordinate systems, especially questions related to position and graphics. For example, "Xiao Min sees Xiao Fang in the direction of 60 degrees north by west. In which direction does Xiao Fang see Xiao Min?" Retrieval-Augmented Generation (RAG): is a model that combines language models and information retrieval technology. It generates answers or content by referencing information from external knowledge bases. It has strong interpretability and customization capabilities and is suitable for multiple natural language processing tasks such as question-answering systems, document generation, and intelligent assistants. The advantage of RAG is that it is highly versatile and can achieve instant knowledge updates, thereby providing more efficient and accurate information services.

[0019] GPT (Generative Pre-Trained): is a deep learning model for text generation based on the Internet and available data for training. It has been trained with a large amount of Internet text and is a large language model that can generate coherent sentences or paragraphs in a human style.

[0020] Prompt: refers to the command or instruction provided by the user when interacting with the large language model. It can be a question, keyword, context information, etc., which is used to indicate the action that the large language model needs to perform or the output it generates. Users can provide clear prompts to guide the large language model to generate the expected response, thereby improving the effect and quality of the interaction.

[0021] Prompt Template: It is a structured text used to generate prompt words. Generating prompt words based on the prompt template can help users interact with the large language model. The large language model can better understand the user's intention and generate more accurate responses that better meet the user's needs.

[0022] Supervised Fine-Tuning (SFT) of Large Models: SFT is a machine learning strategy. SFT mainly involves further parameter tuning of pre-trained models using labeled datasets that contain the correct answers or labels. During the SFT process, the parameters of some layers (or all layers) of the model will be updated based on this new task so that the model can perform this specific task better.

[0023] Manim (Mathematical Animation Engine): is an open source Python library for creating high-quality mathematical animations. Manim allows users to precisely control the details of animations through programming, which is very suitable for education and popular science. Manim provides many internal classes such as Dot, VGroup, Brace, etc., which are very helpful for solving queuing problems by drawing.

[0024] Few-Shot Learning: Few-Shot Learning is a term in the field of machine learning that refers to the ability to train a model with only a small amount of labeled data. This method is particularly suitable for scenarios where it is costly or infeasible to obtain a large amount of labeled data. It is one of the key methods to solve the problem of data scarcity and has promoted the application of machine learning technology in a wider range of fields, especially when data acquisition is limited.

[0025] In this specification, a method for generating a problem-solving model for solving a coordinate system problem, an apparatus for generating a problem-solving model for solving a coordinate system problem, a computing device, a computer-readable storage medium, and a computer program product are provided, which are described in detail one by one in the following embodiments.

[0026] See also Figure 1 , Figure 1 A flowchart of a method for generating a problem-solving model for solving a coordinate system problem provided according to an embodiment of the present specification is shown. From a program perspective, the execution subject of the process can be a program installed on a server or a training platform. From a hardware perspective, the execution subject of the process can be a server or a training platform capable of model training. The method can specifically include the following steps.

[0027] Step 102: Obtaining answer information of a plurality of unanswered questions obtained by using a large language model; the unanswered questions are questions related to a coordinate system; the answer information includes video data for generating a video of answering the unanswered questions.

[0028] In actual applications, a number of questions to be answered can be obtained first, and then the answer information of the questions to be answered can be obtained. There are many ways to obtain the questions to be answered, such as being written based on expert experience; being generated based on a large language model; or being obtained from a known question bank using a retrieval-augmented generation model (RAG), etc., which are not specifically limited here.

[0029] In the embodiment of this specification, the problem to be solved may be one or more problems related to coordinate systems in mathematical applications.

[0030] Problems related to coordinate systems can refer to mathematical problems that are solved using coordinate systems. For example, "The auditorium is 30° east of the stadium by south, so in which direction is the stadium?" For this problem, the first coordinate system can be established with the stadium as the origin. Taking the coordinate system with north at the top, south at the bottom, west at the left, and east at the right as an example, you can take the stadium as the starting point and draw a line segment at 30° east of the stadium to indicate the location of the auditorium. Then, you can establish a second coordinate system with the auditorium as the origin. Based on the coordinate system with north at the top, south at the bottom, west at the left, and east at the right, you can know that the stadium is 30° west of the auditorium.

[0031] In the embodiments of this specification, the video data of the problem-solving video can be the code of the problem-solving video of the problem to be solved, or can be other forms of digital information. Among them, the problem-solving video can be a video used to solve the problem to be solved.

[0032] In the embodiments of this specification, a large language model is used to generate answer information for a number of questions to be answered, and then training data can be obtained based on the answer information generated by the large language model, without having to manually generate training data, thereby improving the efficiency of generating training data.

[0033] Step 104: Obtain training data based on the plurality of questions to be answered and the corresponding answer information including the video data.

[0034] The training data of the embodiments of this specification may include questions to be answered and answer information. For example, the questions to be answered and the answer information may be concatenated to generate training data. Specifically, the questions to be answered may be placed before the answer information, or after the answer information.

[0035] In the embodiments of this specification, the training data may include multiple training data. For example, multiple questions to be answered may be obtained at one time, and multiple training data may be generated using the multiple questions to be answered according to the above method of generating training data.

[0036] For another example, you can obtain unanswered questions multiple times, and according to the above method of generating training data, you can generate multiple training data based on the unanswered questions obtained multiple times.

[0037] Step 106: Use the training data to train a problem-solving model to obtain a trained problem-solving model; the trained problem-solving model can generate video data for explaining a problem-solving video about coordinate system problems.

[0038] The video data generated by the trained problem-solving model for explaining a problem-solving video about a coordinate system problem may include the code of the problem-solving video about the coordinate system problem.

[0039] The code of the problem-solving video can be compiled to generate a problem-solving video. In practical applications, the code of the problem-solving video output by the trained problem-solving model can be compiled using compilation software such as Manim to generate a problem-solving video.

[0040] Generally speaking, the common method used by those skilled in the art is to find ways to collect videos that can be used as training data, and then use these videos to train the problem-solving model; but it is not easy to think of using information in other formats to generate training data. The large language model can process the data to obtain information in text format, and it cannot directly generate videos. Therefore, it is even more difficult for those skilled in the art to think of using a large language model to generate training data for training the problem-solving model. However, in the present application, a large language model can be used to process the questions to be answered to generate information in text format. This text format information may include video data of the problem-solving video of the question to be answered, and then the problem-solving model is trained using the training data containing the video data of the problem-solving video. It is no longer necessary to manually record the corresponding problem-solving video based on the question to be answered, which reduces the cost of generating training data and improves the training efficiency of the problem-solving model after training.

[0041] On the other hand, since the trained problem-solving model is obtained by training with questions related to the coordinate system, it is targeted at solving questions related to the coordinate system, and can more accurately generate the code of the problem-solving video for such questions, thereby ensuring the quality of the generated problem-solving video for questions related to the coordinate system.

[0042] For example, when users such as online teaching institutions or book publishing institutions need to generate problem-solving videos for a large number of questions in the database, they can directly use the trained problem-solving model to output the video data of the problem-solving video for questions related to the coordinate system, and then generate problem-solving videos based on the video data, avoiding the tedious process of manually writing scripts and manually making animations, and improving the efficiency of problem-solving video production.

[0043] In addition, when users such as teachers, students or parents use teaching software for online learning and need to obtain a solution video of a problem related to the coordinate system, the teaching software can be loaded with the above-mentioned trained problem-solving model or can call the above-mentioned trained problem-solving model. After the user enters a problem related to the coordinate system in the teaching software, the teaching software can use the trained problem-solving model to generate video data of the solution video of the above-mentioned problem. The teaching software can also use video compilation software to compile the video data, and then obtain the solution video of the above-mentioned problem and display it, thereby avoiding long waiting times for users, improving problem-solving efficiency, and thus improving user experience.

[0044] based on Figure 1 The method, the examples of this specification also provide some specific implementation plans of the method, which are described below.

[0045] In the embodiments of this specification, the large language model used can be a model of the GPT series, such as GPT-3.5, GPT-4, and GPT-4o. Of course, other series of models can also be used, such as the Tongyi Qianwen model, the Ant Bailing model, etc., which are not limited here.

[0046] In prompt engineering, prompts can refer to text or sentences used to guide the large language model to generate specific responses. Prompts can help the large language model better understand user needs, thereby facilitating the large language model to generate content that better meets user needs.

[0047] In the embodiment of this specification, the answer information of the question to be answered can be obtained by constructing the prompt words of the large language model. Among them, the prompt words of the embodiment of this specification can specifically be text used to guide the large language model to generate the answer information of the question to be answered. The method can also include: Acquire several reference examples of different types related to the coordinate system; the different types include at least one of orientation type and number pair type; the reference examples include example questions and video data for generating a solution video of the example questions.

[0048] Based on the reference example question, a prompt word template including the reference example question is generated.

[0049] For any question to be answered among the plurality of questions to be answered, the question to be answered is added to the prompt word template to obtain a prompt word for the question to be answered.

[0050] The prompt word is input into the large language model to obtain answer information corresponding to any of the questions to be answered.

[0051] In the embodiments of this specification, the reference examples can be selected from a reference example set. For example, a semantic matching model can be used to select reference examples related to the coordinate system from the reference example set. Alternatively, a retrieval enhancement generation model can be used to select reference examples related to the coordinate system from the reference example set. In addition, reference examples can be selected from the reference example set by keyword matching or other matching methods, etc.

[0052] It is understandable that the reference examples can also be written based on expert experience. Of course, it is also possible to extract the code of the existing problem-solving video of the coordinate system-related problem by reverse engineering, and then use the obtained problem and the code of the problem-solving video as reference examples, etc.

[0053] The embodiments of this specification may select a preset number of reference examples, such as one reference example, three reference examples, ten reference examples, etc., which are not limited here.

[0054] In the embodiments of this specification, by using Few-Shot Learning in prompt engineering to guide a large language model to generate answer information, a large amount of answer information can be obtained, thereby facilitating obtaining a large amount of training data.

[0055] In the embodiments of the present specification, the reference example questions may include the example question title and video data for generating a problem-solving video of the example question title, wherein the video data of the problem-solving video may be the code of the problem-solving video.

[0056] Furthermore, in addition to the example questions and the video data of the problem-solving videos, the reference examples may also include other content, so that the answer information output by the large language model also includes the above-mentioned other content, thereby assisting teaching.

[0057] Optionally, the reference example question also includes at least one of step-by-step analysis information for answering the example question step by step, knowledge point information corresponding to the example question, and answer information corresponding to the example question.

[0058] And / or, the answer information obtained using the large language model also includes at least one of step-by-step analysis information for answering the question to be answered step by step, knowledge point information corresponding to the question to be answered, and answer information corresponding to the question to be answered.

[0059] Among them, the step-by-step analysis information can represent the distributed explanation of the question, and the distributed explanation can include a line segment diagram corresponding to each step; the knowledge point information can represent the knowledge point tested by the question; and the answer information can represent the answer to the question.

[0060] For ease of understanding, this specification provides an example of a reference question, which can be specifically shown as follows:

topic

[0061] A. 60° east by south B. 30° west by north C. 60° west by north

Knowledge Points

[0062]

Step-by-step analysis

[0063] The second step is that the auditorium is 30° east-southeast of the stadium. According to the principle that north is at the top, south is at the bottom, west is on the left, and east is on the right, we take the stadium as the starting point and draw a line segment in the direction of 30° east-southeast to indicate the location of the auditorium.

[0064] In the third step, finally, we establish a coordinate system with the auditorium as the center. According to the properties of the image and relative position, we can know that the stadium is 30° west-northwest of the auditorium.

[0065]

Answer

Code

topic

Knowledge Points

[0066]

Step-by-step analysis

[0067] The second step is to observe the positions of these three points. We can find that the y coordinates of A and B are the same, and the x coordinates of A and C are the same, which means that AB is horizontal and AC is vertical, so triangle ABC is a right triangle with the right angle at point A.

[0068] The third step is to connect AB, BC and AC on the coordinate system respectively, and then calculate the lengths of AB and AC. It can be calculated that the length of AB is 5-2=3 units, and the length of AC is 4-2=2 units.

[0069] Step 4. According to the calculation formula for the area of ​​a triangle, the area is equal to one half multiplied by the base multiplied by the height. We can know that the area of ​​triangle ABC is 1 / 2 * AB * AC = 1 / 2 * 3 * 2 = 3 unit areas.

[0070] Step 5. If the side length of a square represents 10 cm, then the area of ​​the triangle is 3 * 10 2 = 300 square centimeters.

[0071]

Answer

[0072] (2) If the side length of a square represents 10 cm, the area of ​​this triangle is 300 square cm.

[0073] ...python {{example_code}} ... In the embodiments of this specification, the reference example may include one or more of the step-by-step analysis information, knowledge point information, and answer information of the example question, and then the prompt words for the question to be answered may include one or more of the step-by-step analysis information, knowledge point information, and answer information of the example question. Under the guidance of the prompt words, the answer information of the question to be answered generated by the large language model will also include information corresponding to the information included in the reference example. This can meet the relevant requirements of step-by-step explanation, overall summary, etc. in the teaching process, and improve the teaching quality.

[0074] In the embodiments of this specification, a large language model is used to automatically generate information including step-by-step analysis information, knowledge point information, answer information, problem-solving video code, etc., thereby reducing the tedious process of manually writing the above content information, improving the efficiency of obtaining solution information for questions to be answered, and improving the efficiency of obtaining training data, thereby improving the training efficiency of the problem-solving model.

[0075] In the embodiment of this specification, the prompt word may be generated based on a prompt word template.

[0076] A prompt word template is a structured text used to generate prompt words. Generating prompt words based on the prompt word template can help the large language model interact with users, so that the large language model can better understand user intentions and generate more accurate responses that better meet user needs.

[0077] In the embodiment of this specification, the prompt word template may include a system prompt word. The system prompt word may be description information for this task. The description information may specifically include one or more of the role information of the large language model, task information indicating the completion of the large language model, task requirement information, and output data format information.

[0078] The role information of the large language model may be the role information that the large language model needs to play for this task. For example, if the large language model is required to play a math teacher, the role information of the large language model may be "You are a math teacher". The task information indicating that the large language model is to complete may be the task information that the large language model needs to complete for this task. The task requirement information may be the requirement information for this task. The task requirement information may include various information of the video data of the problem-solving video, such as at least one of code type information, voice format information, and voice playback format information.

[0079] In the embodiment of the present specification, the prompt word template may further include a first area to be filled and a second area to be filled. The first area to be filled may be used to fill in reference examples, and the second area to be filled may be used to fill in questions to be answered.

[0080] The server or training platform can add system prompt words to the prompt word template, add reference examples to the first area to be filled, add questions to be answered to the second area to be filled, etc., and the prompt words generated using the prompt word template can include system prompt words, reference examples, and questions to be answered.

[0081] In order to ensure the accuracy of the obtained answer information, the embodiments of this specification can also obtain the prompt words of the large prediction model based on different types of reference examples. In this way, for any question to be answered, the large language model can generate answer information more accurately based on the reference examples of the type corresponding to the question to be answered. Optionally, the generation of the prompt word template containing the reference example based on the reference example may specifically include: A first number of first target reference examples belonging to the orientation type are selected from the plurality of reference examples.

[0082] Based on the first number of first target reference examples, a first prompt word template including the first number of first target reference examples is generated.

[0083] Adding any of the questions to be answered to the prompt word template to obtain a prompt word for the any of the questions to be answered may specifically include: If any of the questions to be answered is a question of the orientation type, then any of the questions to be answered is added to the first prompt word template to obtain a prompt word for the any of the questions to be answered.

[0084] In practical applications, the reference examples of the orientation type may be reference examples related to orientation, such as the reference example mentioned above: "The auditorium is 30° east-south of the stadium, so in which direction is the stadium?"

[0085] In order to facilitate understanding of this solution, the embodiments of this specification also provide specific content about the above-mentioned first prompt word template. The following is an example of a first prompt word template: You are a math teacher. Please use manim to write a code to make a tutorial video for the following topic.

[0086] The requirements are as follows: *Create a `ProblemSolving` class, inherit `MyScene`, and write the `self.solve` method.

[0087] *First, use `audio_text` to generate a spoken explanation. Then use `self.voiceover` to play the synthesized audio. Note that latex cannot be used in audio_text, and the formula needs to be converted into Chinese.

[0088] *Call the `Axes` class built into `manim` to create a coordinate system, and call `MyAngle`, `Line`, `Arrow` and other classes to mark the angle relationship.

[0089] * Call `self.create_text` to create a `Text` object, do not use the `Text` built-in in `manim`.

[0090] *Use `self.play_text_left` to display the formula. It needs to be combined with text description. Please keep the text description as concise as possible.

[0091] *When a geographic direction appears in the question, call `calculate_direction_vector` to calculate the direction vector.

[0092] *Only output [Question], [Knowledge Point], [Step-by-step Analysis], [Answer] and [Code], do not output other content.

[0093] Reference Example 1: ... Reference Example 2: ... Enter a question, and generate [Question], [Knowledge Point], [Step-by-step Analysis], [Answer] and [Code] according to the above requirements. As in the above example, "You are a math teacher" can represent the role information of the large language model; "Please use manim to write a piece of code to make a teaching video for the following question" can represent the task information instructing the large language model to complete; "Use `self.play_text_left` to display the formula, which needs to be combined with text descriptions. Please keep the text descriptions as concise as possible" can represent the requirements of the task. "*Create a `ProblemSolving` class, inherit `MyScene`, and write a `self.solve` method." can represent the code type information of the explanation data; "*First use `audio_text` to generate a spoken explanation. Then use `self.voiceover` to play the synthesized audio." can represent the voice format information and voice playback format information in the explanation data.

[0094] In the embodiment of the present specification, the first prompt word template may also include: instruction information for instructing the large language model to establish a coordinate system and mark the angle relationship. For example, in the above example, "*Call the `Axes` class built into `manim` to create a coordinate system, and call `MyAngle`, `Line`, `Arrow` and other classes to mark the angle relationship." Among them, reference example 1 and reference example 2 can be reference examples of orientation type.

[0095] Optionally, generating a prompt word template including the reference example question based on the reference example question may specifically include: A second number of second target reference examples belonging to the number pair type are selected from the plurality of reference examples.

[0096] Based on the second number of second target reference examples, a second prompt word template including the second number of second target reference examples is generated.

[0097] Adding any of the questions to be answered to the prompt word template to obtain a prompt word for the any of the questions to be answered may specifically include: If any of the questions to be answered is a question of the number pair type, then the any of the questions to be answered is added to the second prompt word template to obtain a prompt word for the any of the questions to be answered.

[0098] In practical applications, number pair type reference examples can be reference examples of the relative positional relationship between numbers. For example, "On a square grid, there is a triangle with three vertices at A(2,2), B(5,2), and C(2,4). If the side length of a square represents 10 centimeters, what is the area of ​​this triangle in square centimeters?"

[0099] In order to facilitate understanding of this solution, the embodiments of this specification also provide specific content about the second prompt word template. The following is an example of a second prompt word template: You are a math teacher. Please use manim to write a code to make a tutorial video for the following topic.

[0100] The requirements are as follows: *Create a `ProblemSolving` class, inherit `MyScene`, and write the `self.solve` method.

[0101] *First, use `audio_text` to generate a spoken explanation. Then use `self.voiceover` to play the synthesized audio. Note that latex cannot be used in audio_text, and the formula needs to be converted into Chinese.

[0102] *Call `manim`'s built-in graphics object `Axes` to create a coordinate system, and add arrows, tick marks, and numbers to the coordinate system.

[0103] *Call `Dot` to create dots to represent the location of the object. Different colors are used to distinguish different objects in the figure.

[0104] * Call `self.create_text` to create a `Text` object, do not use the `Text` built-in in `manim`.

[0105] *Use `self.play_text_left` to display the formula. It needs to be combined with text description. Please keep the text description as concise as possible.

[0106] *Only output [Question], [Knowledge Point], [Step-by-step Analysis], [Answer] and [Code], do not output other content.

[0107] Reference Example 3: ... Reference Example 4: ... Enter a question, and please generate [Question], [Knowledge Points], [Step-by-step Analysis], [Answer] and [Code] according to the above requirements.

[0108] Among them, reference example 3 and reference example 4 can be reference examples of number pair type.

[0109] The second prompt word template provided in the embodiment of this specification is partially identical to the first prompt word template provided, except that the reference examples used are different, and the second prompt word template may include instruction information for instructing the large language model to establish a coordinate system and mark the coordinate system parameters, wherein the coordinate system parameters include at least one of arrows, tick marks, numbers, and information for representing the shape and color of the object in the question. The instruction information for instructing the large language model to establish a coordinate system and mark the coordinate system parameters may be "*Call `manim` built-in graphic object `Axes` to create a coordinate system, and add arrows, tick marks and numbers to the coordinate system.", and "*Call `Dot` to create dots to represent the position of objects, and use different colors to distinguish different objects appearing in the figure." in the above example.

[0110] In practical applications, if the question to be answered is a question of the direction type, the question of the direction type can be added to the first prompt word template to obtain the prompt word of the large language model. If the question to be answered is a question of the number pair type, the question of the number pair type can be added to the second prompt word template to obtain the prompt word of the large language model.

[0111] In the embodiment of this specification, in order to ensure the quality of training data, the answer information of the video data that meets the problem-solving requirements and the corresponding questions to be solved can be selected to generate training data. The method of generating a problem-solving model for solving coordinate system problems in the embodiment of this specification can also include: For any of the plurality of questions to be answered, video data contained in the answer information corresponding to the question to be answered is compiled using video compilation software to obtain a problem-solving video.

[0112] Determine whether the problem-solving video meets the problem-solving requirements.

[0113] The obtaining of training data based on the plurality of questions to be answered and the corresponding answer information including the video data may specifically include: If the problem-solving video meets the problem-solving requirements, any of the questions to be answered and the answer information corresponding to any of the questions to be answered are used as training data.

[0114] In practical applications, the answer information may include a variety of different information, such as step-by-step analysis information of the question to be answered, knowledge point information corresponding to the question to be answered, and answer information corresponding to the question to be answered, etc. Therefore, when it is necessary to generate a solution video for the question to be answered, the video data in the answer information can be filtered out from the answer information.

[0115] Optionally, compiling the video data contained in the answer information corresponding to any of the unanswered questions using video compilation software to obtain the answer video may specifically include: The video code contained in the answer information is determined by using a regular expression.

[0116] Save the video code as a Python file.

[0117] Use the manim command to compile the Python file to obtain the problem-solving video.

[0118] In the embodiments of this specification, a regular expression is a tool for describing a string pattern, and a regular expression can be used to search, match and replace a specific string in a text. The regular expression in the embodiments of this specification is a regular expression that can extract video data from the answer information. For example, a regular expression can contain a code identifier character, such as Python, which can start with the code identifier character and extract the information after the code identifier character. Commonly used regular expressions in Python include if...print, etc.

[0119] In the embodiments of this specification, the video compilation software used may be Manim software. The manim (Mathematical Animation Engine) command is a Python library for making animations; using manim to compile Python files can generate a target explanation video.

[0120] It is understandable that, in addition to using Manim software to compile the explanation video, other software such as HandBrake software and FFmpeg software can also be used to compile the explanation video. This is not limited here.

[0121] Furthermore, the embodiments of this specification can also optimize the video compilation software, and use the optimized video compilation software to compile the video. The optimized video compilation software has a built-in optimized logic module, and for the compilation of video data, it can use a small amount of code to complete the work that required a large amount of code before the optimization. For example, if it is necessary to display the distribution analysis information on the left side of the video and the line segment graph on the right side of the video, the video compilation software before optimization needs to accurately determine the distribution analysis information and the absolute coordinate information of the line segment graph. The optimized video compilation software, based on the optimized logic module, can execute "display the distribution analysis information on the left side and the line segment graph on the right side", and can automatically find the specific position on the left side and the specific position on the right side, and then display the analysis and parsing information on the left side and the line segment graph on the right side, without having to accurately determine the distribution analysis information and the absolute coordinate information of the line segment graph. Therefore, the video compilation efficiency of the optimized video compilation software is higher.

[0122] In the embodiments of this specification, the problem-solving video may include images and voice synchronized with the images.

[0123] Furthermore, the problem-solving video may also include video content that explains the problem-solving process step by step. A portion of the video display page of the problem-solving video is used to display the blackboard content, and another portion of the blackboard content is used to display the line segment graph content. The blackboard content may include information such as step-by-step analysis information, knowledge point information, and answer information.

[0124] Figure 2 FIG. 1 is a schematic diagram of an image frame of a problem-solving video provided by an embodiment of the present specification. Figure 2 As shown, Figure 2 The distribution analysis information is shown on the left side of the problem-solving video, and the line segment graph is shown on the right side of the problem-solving video.

[0125] Figure 3 FIG. 1 is a schematic diagram of another image frame of a problem-solving video provided by an embodiment of the present specification. Figure 3 As shown, Figure 3 The answer information is shown on the left side of the problem-solving video, and the line graph is shown on the right side of the problem-solving video.

[0126] In the embodiment of this specification, the video data of the problem-solving video can be video data including the introduction of the topic, the introduction of the knowledge points, the explanation of the topic and the conclusion. Based on the topic introduction video data, the topic introduction video can be obtained, based on the knowledge point introduction video data, the knowledge point introduction video can be obtained, based on the conclusion video data, the conclusion video can be obtained, and so on.

[0127] The topic introduction video can be a video that introduces the topic. For example, the topic introduction video can include a display of the topic, accompanied by a voice: "Let me explain the following question to you", and then the topic can be broadcasted by voice. The knowledge point introduction video can be a video that introduces the knowledge point information involved in the topic. The knowledge point information involved in the topic can be the distribution analysis information involved in the topic. The ending video can be a video used to end the video explanation, for example, it can be a voice broadcast: "This question has been explained, have you learned it, students?"

[0128] In addition, the video data of the problem-solving video may also include video data introducing the identity of the teaching software that loads or calls the trained problem-solving model. Based on the video data, a video introducing the identity of the teaching software that loads or calls the trained problem-solving model may be obtained. For example, it may be a voice broadcast: "Hi, I am an AI smart math teacher."

[0129] Since the problem-solving video is compiled from the video data of the problem-solving video, by judging whether the problem-solving video meets the problem-solving requirements, it is possible to determine whether the video data of the problem-solving video meets the problem-solving requirements, thereby determining whether the answer information meets the problem-solving requirements.

[0130] In the embodiments of the present specification, the answer information of the problem-solving video that can generate the problem-solving requirements and the corresponding questions to be answered are used as training samples, which can reduce the interference of erroneous training samples on the problem-solving model and improve the accuracy of the problem-solving model after training. Then, the problem-solving model after training can more accurately answer questions related to the coordinate system.

[0131] It can be understood that after determining the solution information that meets the problem-solving requirements based on the problem-solving video, the unsolved questions corresponding to the solution information that meets the problem-solving requirements can be determined based on the solution information that meets the problem-solving requirements, so that positive sample training data can be generated based on the determined solution information and the unsolved questions.

[0132] Of course, after selecting the answer information that meets the problem-solving requirements from all the answer information in the above manner, the video data in the answer information that does not meet the problem-solving requirements can also be modified to obtain correct video data. The correct answer information after modification and the unsolved questions corresponding to the correct answer information after modification are used to generate positive sample training data.

[0133] For ease of understanding, the embodiments of this specification also provide specific content for determining whether a problem-solving video meets the problem-solving requirements.

[0134] Determining whether the problem-solving video meets the problem-solving requirements may specifically include: Determine whether the problem-solving video has at least one of the following conditions: The content of the problem-solving video does not correctly answer the question.

[0135] No coordinate diagram is drawn in the problem-solving video.

[0136] There is image overlap in the coordinate diagram drawn in the problem-solving video.

[0137] The coordinate diagram drawn in the problem-solving video is incomplete.

[0138] The coordinate diagram drawn in the problem-solving video is inconsistent with the explanation content.

[0139] The positions of the points representing number pairs in the coordinate graph drawn in the problem-solving video are inconsistent with those in the problem.

[0140] The positions of the points in the coordinate graph drawn in the problem-solving video are inconsistent with the coordinate values ​​of the points.

[0141] The coordinate diagram drawn in the problem-solving video does not have scales or arrows marked on it.

[0142] The angle or direction of the position marked in the coordinate diagram drawn in the problem-solving video is inconsistent with that in the problem.

[0143] The drawing animation and voice explanation in the problem-solving video are not synchronized.

[0144] If not, it is determined that the problem-solving video meets the problem-solving requirements.

[0145] In the embodiments of this specification, one or more models such as a semantic analysis model and a word vector model can be used to analyze whether the question matches the explanation content of the problem-solving video, to determine whether the explanation process of the problem-solving video is correct, to determine whether the explanation result is correct, etc., to determine whether the problem-solving video correctly answers the question. Specifically, recognition can be performed for each frame of the problem-solving video, or one or more frames of images can be randomly selected from the problem-solving video for recognition, which is not specifically limited here.

[0146] Furthermore, image recognition technology can be used to identify whether there is a coordinate graph in the video frame of the problem-solving video to determine whether a line segment graph is drawn in the problem-solving video. If a coordinate graph exists in one or more image frames in the problem-solving video, it can be considered that a coordinate graph is drawn in the problem-solving video, otherwise, it can be considered that a coordinate graph is not drawn in the problem-solving video.

[0147] In practical applications, in addition to ensuring that a coordinate diagram is drawn in the problem-solving video, it is also necessary to ensure that the coordinate diagram is qualified.

[0148] Specifically, one or more methods including bounding box detection methods, image segmentation methods, and deep learning models can be used to detect whether there is image overlap in the coordinate graph in the problem-solving video, thereby determining whether the coordinate graph is qualified.

[0149] One or more of an edge detection algorithm, a contour detection algorithm, and a hash algorithm may be used to detect the boundary of the image to identify whether the coordinate graph is complete, so as to determine whether the coordinate graph is qualified.

[0150] Furthermore, the semantic recognition model can be used to identify whether the coordinate diagram drawn during the explanation corresponds to the explanation content, to determine whether the coordinate diagram drawn in the problem-solving video is consistent with the explanation content, and thus to determine whether the coordinate diagram is qualified.

[0151] Whether the coordinate graph is qualified can be judged by using methods such as key point matching to identify whether the positions of the points representing number pairs in the coordinate graph drawn in the problem-solving video are consistent with the positions stated in the question, whether the positions of the points in the coordinate graph drawn in the problem-solving video are consistent with the coordinate values ​​of the points, whether scales or arrows are marked in the coordinate graph drawn in the problem-solving video, and whether the angle or direction of the orientation marked in the coordinate graph drawn in the problem-solving video is inconsistent with the question, etc.

[0152] In order to ensure the quality of teaching, it is also necessary to synchronize the drawing animation and voice explanation displayed in the problem-solving video to avoid affecting the teaching experience. Specifically, a model can be pre-trained to determine whether the drawing animation and voice explanation are synchronized, and then the trained model can be used to identify whether the drawing animation and voice explanation are synchronized. Of course, relevant video editing software or audio analysis tools can also be used to analyze the synchronization of video and audio.

[0153] In practical applications, the problem-solving video can be a problem explanation using distribution analysis information, so the last frame of the problem-solving video often contains more content than other frames of the problem-solving video. In order to conveniently and accurately determine whether the problem-solving video meets the problem-solving requirements, the problem-solving video can be determined based on the last frame of the problem-solving video.

[0154] For the specific process of determining whether the last frame image of the problem-solving video contains the above-mentioned situation, please refer to the aforementioned content related to determining whether the problem-solving video meets the problem-solving requirements, which will not be elaborated here.

[0155] In the embodiments of the present specification, the training data used to train the problem-solving model may be data in a predetermined format, so that the problem-solving model can understand the training data and improve the training efficiency and model accuracy of the problem-solving model.

[0156] Optionally, obtaining training data based on the plurality of questions to be answered and corresponding answer information including the video data may specifically include: Based on the training data format, the plurality of questions to be answered and the corresponding answer information containing the video data are generated into training data that conforms to the training data format.

[0157] In the embodiments of the present specification, the questions to be answered and the answer information of the questions to be answered can be spliced ​​in a preset order to generate training data.

[0158] Specifically, the question to be answered can be placed in front of the answer information, and the question to be answered and the answer information are separated by a first preset character. In addition, a second preset character can be set before the question to be answered, so that the problem-solving model can understand the training data and improve the training efficiency and model accuracy of the problem-solving model. Among them, the first preset character and the second preset character can be set according to actual needs, and are not specifically limited here.

[0159] For ease of understanding, the embodiments of this specification provide specific descriptions of the training data format.

[0160] Optionally, the training data format may include a role definition part, a question part and a parsing part.

[0161] The role definition part includes information for defining the role played by the problem-solving model.

[0162] The question part includes information of the question to be answered.

[0163] The analysis part includes the knowledge point information corresponding to the question to be answered, step-by-step analysis information, answer information and video data of the problem-solving video.

[0164] In the embodiment of the present specification, the training data format may include multiple identifiers with different meanings, and the problem-solving model may learn the training data from different perspectives based on the identifiers.

[0165] Specifically, the identifiers may include a first identifier, such as [SYSTEM], which can be used to define the role of the model and describe the model's current task. The first identifier may be located at the beginning of the role definition part. The identifier may also include a second identifier, such as [USER], which may be a question to be answered. The identifier may also include a third identifier, such as [ASSISTANT], which may be the parsed part of the question to be answered. Of course, the above identifiers may also be words or texts in other fields, which are not specifically limited here.

[0166] Furthermore, the problem-solving model can understand the model's current task based on the content after the first identifier, so the first identifier can be used to distinguish different business scenarios. In addition, based on the first identifier, the problem-solving model can also clarify the identity of the model performing this task, for example, the problem-solving model is a math teacher. The second identifier and the third identifier can be understood as a set of dialogues. Among them, the second identifier can be a question asked by the user as a user, and the content after the second identifier can be the content of the user's question. The third identifier can be the answer given by the problem-solving model as an assistant to the user's question, and the content after the third identifier can be the assistant's answer. The above three identifiers are spliced ​​in order and can be used as training data to train the problem-solving model.

[0167] It is understandable that [ASSISTANT] may be the first preset character mentioned above, and [USER] may be the second preset character mentioned above.

[0168] For example, the format and content of the training data can be as follows: [SYSTEM] You are a math teacher. Please use manim to write a code to make a teaching video for the following question.

[0169] [USER]The school auditorium is 30° east-southeast of the stadium. In which direction is the stadium?

[0170] [ASSISTANT] [Knowledge point] xxx, [Step-by-step analysis] xxx, [Answer information] xxx, and [Code information] xxx.

[0171] Usually, we hope that the trained problem-solving model can output accurate answer information for the question after obtaining the question. Specifically, when using the training data to train the problem-solving model, the problem-solving model can output prediction data based on the role definition part and the question part proposed by the user. The problem-solving model can also use the analysis part of the question as reference data to adjust the model parameters, and then obtain the trained problem-solving model.

[0172] In the embodiments of this specification, supervised fine-tuning can be used to train the problem-solving model. During the training process, the problem-solving model can adjust its own hyperparameters according to the training data to obtain a trained problem-solving model.

[0173] The goal of training the problem-solving model is to minimize the difference between the reference data and the predicted data, that is, to reduce the loss between the reference data and the predicted data. Therefore, the parameters of the problem-solving model can be continuously adjusted during the training process to reduce the loss. When the loss no longer decreases or decreases to within the preset range, it can be said that the problem-solving model training is complete.

[0174] To improve training efficiency, the problem-solving model in the embodiments of this specification may be a pre-trained model, such as the YunLLM70B model, the Multimodal-CoT model, or the Mixtral8x7B model, or other types of pre-trained models, which are not specifically limited here.

[0175] In the embodiments of this specification, the obtained training data can be used to train the problem-solving model.

[0176] Assuming that multiple training data are obtained, each piece of training data may include the question to be answered, knowledge point information, step-by-step analysis information, answer information, and the code of the corresponding problem-solving video, etc. In addition, the training data should include various types of questions to be answered and other information corresponding to the various types of questions to be answered, so as to improve the generalization ability of the problem-solving model.

[0177] After obtaining the training data, the training data can also be cleaned and formatted to ensure the consistency and standardization of the data, so as to facilitate the training of the problem-solving model. Specifically, the training data can be segmented, denoised, and the problem-solving video data can be formatted. If the knowledge point information, step-by-step analysis information, answer information, and corresponding problem-solving video data in the training data are generated using a large language model, since the data generated by the large language model is usually standardized and in a unified format, data cleaning and formatting are not required for the training data at this time.

[0178] When the problem-solving model training starts, you can initialize the problem-solving model and set some training parameters. For example, for the YunLLM70B model, you can load the YunLLM70B model and configure the training parameters, such as the learning rate, batch size, and number of training rounds.

[0179] In addition, the problem-solving model can be fine-tuned using training data. Specifically, the problem to be solved can be used as input, and the generation of accurate knowledge point information, step-by-step analysis information, answer information, and video data of the problem-solving video can be used as the goal. During the training process, the problem-solving model continuously adjusts parameters through back propagation to minimize the difference between the generated problem-solving video data and the labeled problem-solving video video data.

[0180] During the problem-solving model training process, the validation set can be used to evaluate the model performance regularly or irregularly to adjust the problem-solving model parameters and training strategies to ensure the stability and accuracy of the problem-solving model.

[0181] After the problem-solving model training is completed, the model parameters and structure of the trained problem-solving model can be saved for subsequent use and deployment.

[0182] In practical applications, model evaluation is an important step in verifying the quality and reliability of video data generated by the trained problem-solving model to generate the problem-solving video. As an implementation method, the embodiment of this specification can evaluate the performance of the trained problem-solving model through the following steps.

[0183] Get a test set. Specifically, 100 questions can be randomly selected from the test question bank, and it is ensured that the test set questions do not overlap with the training data. In addition, the test set usually needs to ensure that the questions cover various types and complexities of questions that can be solved using line graphs, so as to fully evaluate the performance of the problem-solving model.

[0184] The trained problem-solving model is used to generate the corresponding problem-solving video data for each question in the test set. The generation process is similar to the training stage. Specifically, the question can be input into the trained problem-solving model, and the trained problem-solving model can output knowledge point information, step-by-step analysis information, and problem-solving video video data.

[0185] The video data of the generated problem-solving video is input into the video compiling software to compile the problem-solving video.

[0186] Furthermore, the performance of the trained problem-solving model can be evaluated based on the compiled problem-solving video. Specifically, the problem-solving video can be evaluated using a predetermined evaluation standard. The evaluation standard may include the accuracy of the video, such as whether the problem-solving steps and results are correct, the intuitiveness of the video, such as whether the video is clear and easy to understand, the vividness of the video, such as whether the explanation process is vivid and interesting, and the technicality of the video, such as whether the code generation is efficient and the video playback is smooth.

[0187] In addition, the generated video can also be evaluated using the method mentioned above to determine whether the problem-solving video meets the problem-solving requirements. Professional teachers and technical experts can also be organized to manually evaluate the generated video and record the evaluation results. Of course, the several methods mentioned above can also be combined, and no specific restrictions are made here.

[0188] Furthermore, the evaluation results can be statistically analyzed to evaluate the performance of the trained model. Based on the analysis results, the strengths and weaknesses of the trained problem-solving model can be determined, thus providing a basis for further optimizing the problem-solving model.

[0189] Through the above evaluation steps, we ensure that the trained problem-solving model has the ability to generate high-quality video data for problem-solving videos and provide reliable technical support for practical applications.

[0190] Figure 4 FIG. 1 is a schematic diagram of a method for generating a problem-solving model for solving a coordinate system problem provided by an embodiment of the present specification. Figure 4As shown, the method for generating a problem-solving model for solving a coordinate system problem may include: Step 402: Obtain reference examples.

[0191] In the embodiments of this specification, the reference examples may be several reference examples of different types related to the coordinate system.

[0192] The reference examples may be selected from a reference example set, for example, by using a semantic matching model or a retrieval enhancement generation model to select reference examples related to the coordinate system from the reference example set.

[0193] It is understandable that the reference examples can also be written based on expert experience. Of course, it is also possible to extract the code of the existing problem-solving video of the coordinate system-related problem by reverse engineering, and then use the obtained problem and the code of the problem-solving video as reference examples, etc.

[0194] Step 404: Get answer information.

[0195] In the embodiment of this specification, the answer information may be answer information of several unanswered questions obtained by using a large language model.

[0196] The answer information may include video data for generating a video of the problem to be solved, wherein the video data of the video of the problem-solving may be a code of the video of the problem-solving.

[0197] A large language model is used to generate answer information for a number of questions to be answered, and then training data can be obtained based on the answer information generated by the large language model, without having to manually generate training data, which can improve the efficiency of generating training data.

[0198] Step 406: Compile the problem-solving video.

[0199] In practical applications, the code of the problem-solving video in the answer information obtained in step 404 can be compiled using compilation software such as Manim, thereby generating a problem-solving video.

[0200] Step 408: Annotate the problem-solving video.

[0201] In the embodiments of this specification, marking a problem-solving video may indicate whether the problem-solving video meets the problem-solving requirements. Then, the problem-solving videos that do not meet the problem-solving requirements may be removed, and the questions and problem-solving information corresponding to the problem-solving videos that meet the problem-solving requirements may be used as training samples to train the problem-solving model.

[0202] Step 410: Model fine-tuning.

[0203] In the embodiments of this specification, supervised fine-tuning can be used to train the problem-solving model. During the training process, the problem-solving model can adjust its own hyperparameters according to the training data to obtain a trained problem-solving model.

[0204] The video data generated by the trained problem-solving model for explaining a problem-solving video about a coordinate system problem may include the code for the problem-solving video explaining the problem-solving video about the coordinate system problem.

[0205] In the embodiments of this specification, the trained problem-solving model can be used to directly obtain video data of a problem-solving video about a coordinate system problem. Compared with obtaining video data using a large language model, since the trained problem-solving model is obtained after training using reference examples related to the coordinate system, it is targeted at questions related to the coordinate system and can more accurately generate video data for such questions, thereby ensuring the quality of the generated problem-solving video.

[0206] In addition, when using the trained problem-solving model to solve problems, there is no need to preset templates to obtain a large number of prompt words similar to those used by the above-mentioned large language model, which can also improve problem-solving efficiency.

[0207] Corresponding to the above-mentioned method embodiment for generating a problem-solving model for solving a coordinate system problem, this specification also provides an apparatus embodiment for generating a problem-solving model for solving a coordinate system problem. Figure 5 FIG. 1 is a schematic diagram showing a structure of a device for generating a problem-solving model for solving a coordinate system problem provided by an embodiment of the present specification. Figure 5 As shown, the device comprises: The answer information acquisition module 502 is configured to obtain answer information of several unanswered questions obtained by using a large language model; the unanswered questions are questions related to the coordinate system; the answer information includes video data used to generate a solution video of the unanswered questions.

[0208] The training data acquisition module 504 is configured to obtain training data based on the plurality of questions to be answered and the corresponding answer information including the video data.

[0209] The model training module 506 is configured to train the problem-solving model using the training data to obtain a trained problem-solving model. The trained problem-solving model can generate video data for explaining a problem-solving video about a coordinate system problem.

[0210] Optionally, the device may further include: The reference example acquisition module is configured to acquire several reference examples of different types related to the coordinate system; the different types include at least one of the orientation type and the number pair type; the reference example includes the example question and video data for generating a solution video of the example question.

[0211] The template generation module is configured to generate a prompt word template containing the reference example question based on the reference example question.

[0212] The prompt word acquisition module is configured to add any of the plurality of questions to be answered to the prompt word template to obtain a prompt word for the question to be answered.

[0213] The prompt word input module is configured to input the prompt word into the large language model to obtain the answer information corresponding to any of the questions to be answered.

[0214] Optionally, the template generation module may be specifically configured as follows: A first number of first target reference examples belonging to the orientation type are selected from the plurality of reference examples.

[0215] Based on the first number of first target reference examples, a first prompt word template including the first number of first target reference examples is generated.

[0216] The prompt word acquisition module can be specifically configured as follows: If any of the questions to be answered is a question of the orientation type, then any of the questions to be answered is added to the first prompt word template to obtain a prompt word for the any of the questions to be answered.

[0217] Optionally, the template generation module may be specifically configured as follows: A second number of second target reference examples belonging to the number pair type are selected from the plurality of reference examples.

[0218] Based on the second number of second target reference examples, a second prompt word template including the second number of second target reference examples is generated.

[0219] The prompt word acquisition module can be specifically configured as follows: If any of the questions to be answered is a question of the number pair type, then the any of the questions to be answered is added to the second prompt word template to obtain a prompt word for the any of the questions to be answered.

[0220] Optionally, the first prompt word template also includes: instruction information for instructing the large language model to establish a coordinate system and mark angle relationships.

[0221] Optionally, the second prompt word template also includes: instruction information for instructing the large language model to establish a coordinate system and mark coordinate system parameters. The coordinate system parameters include at least one of arrows, scale lines, numbers, and information for representing the shape and color of an object in the question.

[0222] Optionally, the reference example question also includes at least one of step-by-step analysis information for answering the example question step by step, knowledge point information corresponding to the example question, and answer information corresponding to the example question.

[0223] And / or, the answer information obtained using the large language model also includes at least one of step-by-step analysis information for answering the question to be answered step by step, knowledge point information corresponding to the question to be answered, and answer information corresponding to the question to be answered.

[0224] Optionally, the device may further include: The compiling module is configured to compile the video data contained in the answer information corresponding to any of the plurality of questions to be answered using video compiling software to obtain a solution video.

[0225] The judgment module is configured to judge whether the problem-solving video meets the problem-solving requirements.

[0226] The training data acquisition module 504 may be specifically configured as follows: If the problem-solving video meets the problem-solving requirements, any of the questions to be answered and the answer information corresponding to any of the questions to be answered are used as training data.

[0227] Optionally, the judging module may be specifically configured as follows: Determine whether the problem-solving video has at least one of the following conditions: The content of the problem-solving video does not correctly answer the question.

[0228] No coordinate diagram is drawn in the problem-solving video.

[0229] There is image overlap in the coordinate diagram drawn in the problem-solving video.

[0230] The coordinate diagram drawn in the problem-solving video is incomplete.

[0231] The coordinate diagram drawn in the problem-solving video is inconsistent with the explanation content.

[0232] The positions of the points representing number pairs in the coordinate graph drawn in the problem-solving video are inconsistent with those in the problem.

[0233] The positions of the points in the coordinate graph drawn in the problem-solving video are inconsistent with the coordinate values ​​of the points.

[0234] The coordinate diagram drawn in the problem-solving video does not have scales or arrows marked on it.

[0235] The angle or direction of the position marked in the coordinate diagram drawn in the problem-solving video is inconsistent with that in the problem.

[0236] The drawing animation and voice explanation in the problem-solving video are not synchronized.

[0237] If not, it is determined that the problem-solving video meets the problem-solving requirements.

[0238] Optionally, the compilation module may be specifically configured as follows: The video code contained in the answer information is determined by using a regular expression.

[0239] Save the video code as a Python file.

[0240] Use the manim command to compile the Python file to obtain the problem-solving video.

[0241] Optionally, the training data acquisition module 504 may be specifically configured as follows: Based on the training data format, the plurality of questions to be answered and the corresponding answer information containing the video data are generated into training data that conforms to the training data format.

[0242] Optionally, the training data format includes a role definition part, a question part and a parsing part.

[0243] The role definition part includes information for defining the role played by the problem-solving model.

[0244] The question part includes information of the question to be answered.

[0245] The analysis part includes the knowledge point information corresponding to the question to be answered, step-by-step analysis information, answer information and video data of the problem-solving video.

[0246] Figure 6 The block diagram of a computing device 600 according to an embodiment of the present specification is shown. The components of the computing device 600 include but are not limited to a memory 610 and a processor 620. The processor 620 is connected to the memory 610 via a bus 630, and the database 650 is used to store data.

[0247] The computing device 600 also includes an access device 640 that enables the computing device 600 to communicate via one or more networks 660. Examples of these networks include a public switched telephone network (PSTN), a local area network (LAN), a wide area network (WAN), a personal area network (PAN), or a combination of communication networks such as the Internet. The access device 540 may include one or more of any type of network interface (e.g., a network interface card (NIC)) of wired or wireless, such as an IEEE 802.11 wireless local area network (WLAN) wireless interface, a world-wide interoperability for microwave access (Wi-MAX) interface, an Ethernet interface, a universal serial bus (USB) interface, a cellular network interface, a Bluetooth interface, and a near field communication (NFC).

[0248] In one embodiment of the present specification, the above components of the computing device 600 and Figure 6 Other components not shown in the figure may also be connected to each other, for example, via a bus. It should be understood that Figure 6 The computing device structure block diagram shown is only for the purpose of illustration, and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0249] The computing device 600 may be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, a personal digital assistant, a laptop computer, a notebook computer, a netbook, etc.), a mobile phone (e.g., a smart phone), a wearable computing device (e.g., a smart watch, smart glasses, etc.), or other types of mobile devices, or a stationary computing device such as a desktop computer or a personal computer (PC). The computing device 600 may also be a mobile or stationary server.

[0250] The processor 620 is used to execute the following computer executable instructions, which, when executed by the processor, implement the steps of the method for generating a problem-solving model for solving coordinate system problems.

[0251] The above is a schematic scheme of a computing device of this embodiment. It should be noted that the technical scheme of the computing device and the technical scheme of the method of the problem-solving model and / or the technical scheme of the method of solving the problem belong to the same concept, and the details of the technical scheme of the computing device that are not described in detail can be found in the description of the technical scheme of the method of the problem-solving model and / or the technical scheme of the method of solving the problem.

[0252] An embodiment of the present specification also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-mentioned method of generating a problem-solving model for solving coordinate system problems.

[0253] The above is a schematic scheme of a computer-readable storage medium of this embodiment. It should be noted that the technical scheme of the storage medium and the technical scheme of the method for generating a problem-solving model for solving a coordinate system problem belong to the same concept, and the details not described in detail in the technical scheme of the storage medium can be referred to the description of the technical scheme of the method for generating a problem-solving model for solving a coordinate system problem.

[0254] An embodiment of the present specification also provides a computer program, wherein when the computer program is executed in a computer, the computer is caused to execute the steps of the method for generating a problem-solving model for solving a coordinate system problem.

[0255] The above is a schematic scheme of a computer program of this embodiment. It should be noted that the technical scheme of the computer program and the technical scheme of the method of the problem-solving model and / or the technical scheme of the method of solving the problem belong to the same concept, and the details of the technical scheme of the computer program that are not described in detail can be found in the description of the technical scheme of the method of the problem-solving model and / or the technical scheme of the method of solving the problem.

[0256] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0257] The computer instructions include computer program codes, which may be in source code form, object code form, executable files or some intermediate forms, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium may be appropriately increased or decreased according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0258] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the embodiments of this specification are not limited by the order of the actions described, because according to the embodiments of this specification, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the embodiments of this specification.

[0259] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0260] The preferred embodiments of this specification disclosed above are only used to help explain this specification. The optional embodiments do not describe all the details in detail, nor do they limit the invention to only the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of the embodiments of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the embodiments of this specification, so that technicians in the relevant technical field can well understand and use this specification. This specification is only limited by the claims and their full scope and equivalents.

Claims

1. A method for generating a problem-solving model for solving a coordinate system problem, characterized in that: include: Obtaining the answer information of several unanswered questions obtained by using a large language model; The questions to be answered are questions related to the coordinate system; The answer information includes video data used to generate a video of the answer to the question to be answered; Based on the plurality of questions to be answered and the corresponding answer information including the video data, obtaining training data; The training data is used to train a problem-solving model to obtain a trained problem-solving model; the trained problem-solving model can generate video data for explaining a problem-solving video about a coordinate system problem.

2. The method according to claim 1, characterized in that The method further comprises: Acquire a plurality of reference examples of different types related to the coordinate system; the different types include at least one of an orientation type and a number pair type; the reference examples include example questions and video data for generating a video of solving the example questions; Based on the reference example question, generating a prompt word template including the reference example question; For any question to be answered among the plurality of questions to be answered, adding the question to be answered to the prompt word template to obtain a prompt word for the question to be answered; The prompt word is input into the large language model to obtain answer information corresponding to any of the questions to be answered.

3. The method according to claim 2, characterized in that The step of generating a prompt word template including the reference example question based on the reference example question specifically includes: Selecting a first number of first target reference examples belonging to the orientation type from the plurality of reference examples; Based on the first number of first target reference examples, generating a first prompt word template including the first number of first target reference examples; The step of adding any of the questions to be answered to the prompt word template to obtain a prompt word for the any of the questions to be answered specifically includes: If any of the questions to be answered is a question of the orientation type, then any of the questions to be answered is added to the first prompt word template to obtain a prompt word for the any of the questions to be answered.

4. The method according to claim 2, characterized in that: The step of generating a prompt word template including the reference example question based on the reference example question specifically includes: Selecting a second number of second target reference examples belonging to the number pair type from the plurality of reference examples; Based on the second number of second target reference examples, generating a second prompt word template including the second number of second target reference examples; The step of adding any of the questions to be answered to the prompt word template to obtain a prompt word for the any of the questions to be answered specifically includes: If any of the questions to be answered is a question of the number pair type, then the any of the questions to be answered is added to the second prompt word template to obtain a prompt word for the any of the questions to be answered.

5. The method according to claim 3, characterized in that: The first prompt word template also includes: instruction information for instructing the large language model to establish a coordinate system and mark the angle relationship.

6. The method according to claim 4, characterized in that The second prompt word template also includes: instruction information for instructing the large language model to establish a coordinate system and mark coordinate system parameters; the coordinate system parameters include at least one of arrows, scale lines, numbers, and information for representing the shape and color of an object in the question.

7. The method according to claim 2, characterized in that: The reference example question also includes at least one of step-by-step analysis information for answering the example question step by step, knowledge point information corresponding to the example question, and answer information corresponding to the example question; And / or, the answer information obtained using the large language model also includes at least one of step-by-step analysis information for answering the question to be answered step by step, knowledge point information corresponding to the question to be answered, and answer information corresponding to the question to be answered.

8. The method according to claim 1, characterized in that: The method further comprises: For any of the plurality of questions to be answered, compile the video data contained in the answer information corresponding to the question to be answered using video compilation software to obtain a question-solving video; Determine whether the problem-solving video meets the problem-solving requirements; The obtaining of training data based on the plurality of questions to be answered and the corresponding answer information including the video data specifically includes: If the problem-solving video meets the problem-solving requirements, any of the questions to be answered and the answer information corresponding to any of the questions to be answered are used as training data.

9. The method according to claim 8, characterized in that The determining whether the problem-solving video meets the problem-solving requirements specifically includes: Determine whether the problem-solving video has at least one of the following conditions: The content of the problem-solving video fails to correctly answer the question; The coordinate diagram is not drawn in the problem-solving video; There is image overlap in the coordinate diagram drawn in the problem-solving video; The coordinate diagram drawn in the problem-solving video is incomplete; The coordinate diagram drawn in the problem-solving video is inconsistent with the explanation content; The positions of the points representing the number pairs in the coordinate diagram drawn in the problem-solving video are inconsistent with those in the problem; The position of the point in the coordinate graph drawn in the problem-solving video is inconsistent with the coordinate value of the point; The coordinate diagram drawn in the problem-solving video does not have scales or arrows marked on it; The angle or direction of the position marked in the coordinate diagram drawn in the problem-solving video is inconsistent with that in the problem; The drawing animation and voice explanation in the problem-solving video are not synchronized; If not, it is determined that the problem-solving video meets the problem-solving requirements.

10. The method according to claim 8, characterized in that The step of compiling the video data contained in the answer information corresponding to any of the questions to be answered using video compilation software to obtain a question-solving video specifically includes: Using a regular expression, determining the video code contained in the answer information; Save the video code as a Python file; Use the manim command to compile the Python file to obtain the problem-solving video.

11. The method according to claim 1, characterized in that: The obtaining of training data based on the plurality of questions to be answered and the corresponding answer information including the video data specifically includes: Based on the training data format, the plurality of questions to be answered and the corresponding answer information containing the video data are generated into training data that conforms to the training data format.

12. The method according to claim 11, characterized in that The training data format includes a role definition part, a question part and a parsing part; The role definition part includes information for defining the role played by the problem-solving model; The question part includes information of the question to be answered; The analysis part includes the knowledge point information corresponding to the question to be answered, step-by-step analysis information, answer information and video data of the problem-solving video.

13. A device for generating a problem-solving model for solving a coordinate system problem, characterized in that: include: An answer information acquisition module is configured to acquire answer information of a plurality of unanswered questions obtained by using a large language model; The problem to be solved is a problem related to the coordinate system; the solution information includes video data for generating a solution video of the problem to be solved; A training data acquisition module is configured to obtain training data based on the plurality of questions to be answered and the corresponding answer information including the video data; A model training module is configured to train a problem-solving model using the training data to obtain a trained problem-solving model; The trained problem-solving model can generate video data for explaining a problem-solving video about coordinate system problems.

14. A computing device, characterized in that include: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method described in any one of claims 1 to 12 are implemented.

15. A computer-readable storage medium, characterized in that: It stores computer executable instructions, which, when executed by a processor, implement the steps of the method described in any one of claims 1 to 12.

16. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 12 are implemented.

Images