Scientific education method and device based on thinking chain and computer storage medium

Abstract

The invention provides a thinking chain-based scientific education method, a scientific education device and a computer storage medium. The scientific education method comprises the following steps: determining a plurality of first-level candidate thinking chains by utilizing data characteristics of education input data; obtaining a first judgment result of each first-level candidate thinking chain; determining the first-level candidate thinking chain with a qualified judgment result as a first-level teaching thinking chain; obtaining an output result of the first-level teaching thinking chain for the data features; obtaining an evaluation result of each second-level candidate thinking chain; and determining the second-level candidate thinking chain with a qualified judgment result as a second-level teaching thinking chain until the last-level teaching thinking chain outputs a scientific education answer of the education input data. Through the scientific education method, teaching behaviors are disassembled, step-by-step execution and reflex retrieval are carried out, step-by-step errors or unreasonable parts are found, a new teaching path is executed again, the adaptability and continuity of the scientific teaching process are met, and the scientific education effect is improved.

Description

Technical Field

[0001] The present application relates to the field of online education technology, and in particular to a scientific education method, a scientific education device and a computer storage medium based on a thinking chain. Background Art

[0002] The existing large educational models mainly use existing subject problem-solving process data for training and reasoning, gradually enhancing the large model's ability to "solve problems". The O1 model newly launched by foreign Open AI manufacturers creatively solves complex problems step by step by generating a long chain of thoughts internally. This chain of thoughts breaks down large problems into a series of small steps and solves them one by one. Whether it is the existing large educational models in China or the advanced O1 models abroad, they lack the integration of human teachers' teaching experience and knowledge, and cannot achieve truly human-like scientific education and teaching, resulting in poor online education results. Summary of the invention

[0003] In order to solve the above technical problems, the present application proposes a scientific education method, a scientific education device and a computer storage medium based on thought chain.

[0004] In order to solve the above technical problems, the present application proposes a scientific education method based on thinking chain, which includes:

[0005] Obtaining educational input data;

[0006] Determine a number of first-level candidate thinking chains using the data features of the educational input data;

[0007] Using the preset scientific education prior information, obtain the first evaluation result of each first-level candidate thinking chain;

[0008] Determine the first-level candidate thinking chain with qualified evaluation results as the first-level teaching thinking chain;

[0009] Obtaining the output result of the first-level teaching thinking chain for the data feature;

[0010] Using the output results, determining a number of second-level candidate thinking chains;

[0011] Using the preset scientific education prior information, obtain the evaluation results of each second-level candidate thinking chain;

[0012] The second-level candidate thinking chain that is judged to be qualified is determined as the second-level teaching thinking chain, until the last-level teaching thinking chain outputs the scientific education answer to the education input data.

[0013] The method of using the preset scientific education prior information to obtain the first evaluation result of each first-level candidate thinking chain includes:

[0014] Input the preset scientific education prior information and each first-level candidate thinking chain into the self-evaluation model, and extract the teaching suitability evaluation value and logical reasoning evaluation value of the first-level candidate thinking chain;

[0015] Based on the teaching suitability evaluation value and the logical reasoning evaluation value of the first-level candidate thinking chain, output a first evaluation result of the first-level candidate thinking chain;

[0016] The preset scientific education prior information includes subject knowledge data and / or teaching and learning behavior data.

[0017] The step of determining the first-level candidate thinking chain with a qualified evaluation result as the first-level teaching thinking chain includes:

[0018] In response to the existence of at least two first-level candidate thinking chains with qualified evaluation results, the teaching suitability evaluation value and / or the highest logical reasoning evaluation value among the first-level candidate thinking chains with qualified evaluation results are obtained and determined as the first-level teaching thinking chain.

[0019] The method of using the data features of the educational input data to determine a number of first-level candidate thought chains includes:

[0020] Inputting the data features of the educational input data into the language macro model to generate a number of first-level candidate thought chains;

[0021] The scientific education method further comprises:

[0022] Inputting subject knowledge data, teaching and learning behavior data and training samples into the language macro model, and obtaining teaching suitability evaluation values ​​and educational value evaluation values ​​of the training samples;

[0023] The teaching suitability evaluation value and the educational value evaluation value are used to train the language model to update the parameters of the language model.

[0024] After obtaining the teaching suitability evaluation value and the educational value evaluation value of the training sample, the scientific education method further includes:

[0025] Based on the expert guidance instructions, the teaching appropriateness evaluation value and / or the educational value evaluation value are updated.

[0026] The step of training the language model by using the teaching suitability evaluation value and the educational value evaluation value to update parameters of the language model includes:

[0027] Input the teaching suitability evaluation value and the educational value evaluation value into a collaborative reward model to obtain training reward parameters;

[0028] The training reward parameter is used to feed back the language macro model.

[0029] Wherein, after determining the first-level candidate thinking chain with the qualified evaluation result as the first-level teaching thinking chain, the scientific education method further includes:

[0030] Generate a seed instruction set using the first-level teaching thinking chain and the data features;

[0031] The prompt information instructions of the first-level teaching thinking chain are filtered and generated from the seed instruction set.

[0032] Wherein, the educational input data includes: at least two of text modality data, image modality data, and voice modality data;

[0033] After obtaining the educational input data, the scientific education method includes:

[0034] extracting multimodal data features of the educational input data;

[0035] Mapping the multimodal data features to a shared embedding space;

[0036] Extracting data features of the multimodal data features in the shared embedding space.

[0037] In order to solve the above technical problems, the present application also proposes a scientific education device, which includes a memory and a processor coupled to the memory; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the scientific education method as described above.

[0038] In order to solve the above technical problems, the present application also proposes a computer storage medium, which is used to store program data. When the program data is executed by a computer, it is used to implement the above scientific education method.

[0039] Compared with the prior art, the beneficial effects of the present application are as follows: the scientific education device obtains educational input data; uses the data features of the educational input data to determine a number of first-level candidate thinking chains; uses the preset scientific education prior information to obtain the first evaluation result of each first-level candidate thinking chain; determines the first-level candidate thinking chain with a qualified evaluation result as the first-level teaching thinking chain; obtains the output result of the first-level teaching thinking chain for the data features; uses the output result to determine a number of second-level candidate thinking chains; uses the preset scientific education prior information to obtain the evaluation result of each second-level candidate thinking chain; determines the second-level candidate thinking chain with a qualified evaluation result as the second-level teaching thinking chain, until the last-level teaching thinking chain outputs the scientific education answer to the educational input data. Through the above scientific education method, the teaching behavior is disassembled, the search is performed step by step and reflected, the errors or unreasonableness of the steps are found, and the new teaching path is re-executed to meet the appropriateness and coherence of the scientific teaching process and improve the scientific education effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0041] in:

[0042] Figure 1 It is a flowchart of an embodiment of a scientific education method based on thinking chain provided by the present application;

[0043] Figure 2 It is a scene diagram of a specific embodiment of the scientific education method based on the thinking chain provided by the present application;

[0044] Figure 3 It is a schematic diagram of the semantic alignment process of cross-scenario multimodal teaching provided by this application;

[0045] Figure 4 It is a schematic diagram of the multimodal science education semantic extraction process provided by this application;

[0046] Figure 5 It is a schematic diagram of the automatic construction process of the teaching thinking chain provided by this application;

[0047] Figure 6 It is a schematic diagram of the large model optimization process of the “teaching appropriateness-value alignment” synergistic enhancement provided by this application;

[0048] Figure 7 yes Figure 1The specific flow chart of step S12 of the scientific education method based on the thinking chain is shown;

[0049] Figure 8 It is a schematic diagram of the large model optimization process of "teaching appropriateness-logical accuracy" synergistic enhancement provided by this application;

[0050] Fig. 9 yes Figure 1 The specific flow chart of step S13 of the scientific education method based on the thinking chain is shown;

[0051] Fig.10 It is a schematic diagram of the instruction set expansion process based on the teaching thinking chain provided by this application;

[0052] Fig.11 It is a flowchart of another embodiment of the scientific education method based on thinking chain provided by the present application;

[0053] Fig.12 It is a structural schematic diagram of an embodiment of a scientific education device provided by the present application;

[0054] Fig.13 It is a structural diagram of an embodiment of a computer storage medium provided by the present application. DETAILED DESCRIPTION

[0055] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0056] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can, for example, be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0057] Since the large language model has powerful capabilities in language understanding, text generation, knowledge question and answer, logical reasoning, mathematical ability, coding ability and multi-modal interaction, and is trained based on a large amount of unsupervised educational text data, the large model is exposed to various experiences and methods in the scientific education process. Therefore, the large model usually has the ability of primary and secondary school education and teaching, and can be applied in many fields such as primary and secondary school science education and teaching in the future.

[0058] However, the capabilities of general big models alone cannot meet the actual application of various vertical scenarios. With the rapid development of big models, various vertical big models for scenario applications have also emerged, such as medical big models, judicial big models and other special big models.

[0059] The goal of science education is to cultivate students' practical ability, comprehensive thinking ability, innovative ability, and interest and love for science. It is a highly complex teaching activity that strongly relies on the experience of teachers and matches the abilities of learners at different stages. Intelligent assistance for science education and teaching requires comprehensive requirements such as teaching appropriateness, knowledge and logical accuracy, and educational values. However, the complexity of science education teaching scenarios and elements and the resulting general large models have poor teaching appropriateness, inaccurate reasoning, and poor educational effects in science education scenarios. It is urgent to overcome the difficulty of quantitative representation of all complex elements of the teaching process, how to enhance the knowledge of the education large model based on teaching semantics such as subject knowledge, teaching and learning behaviors, how to enhance the complex mathematical and logical reasoning capabilities of domestic multimodal science education large models, and how to achieve automatic alignment of models of cross-scenario teaching semantics and educational values ​​and establish an efficiency-enhancing learning mechanism. This is the fundamental way to ensure that the teaching and learning of science education-specific large models are reliable and efficient.

[0060] Among them, the thought chain is an improved prompt technology that is used to improve the performance of large models on complex reasoning tasks, especially problems involving reasoning and multi-step thinking. The intermediate steps in this series of reasoning are called thought chains.

[0061] In this regard, this application provides a scientific education method based on thinking chain. The domestic multimodal science education model driven by the teaching thinking chain proposed in this application is based on the knowledge in the field of science education (interdisciplinary knowledge semantic system, complex teaching and learning process knowledge, complex mathematical logic reasoning, etc.), and through the cross-scenario multimodal teaching semantic automatic alignment technology, the semantic consistency of different modal teaching data is ensured; through the teaching semantics, a teaching thinking chain that conforms to the teaching appropriateness and logical reasoning is constructed, and then through the reinforcement learning of scientific expert feedback and self-feedback reinforcement learning and other technologies, a domestic multimodal science education model is established.

[0062] Specifically, this application explores the automatic semantic alignment technology of multimodal teaching across scenarios based on interdisciplinary knowledge graphs and multimodal teaching cases to ensure the semantic consistency of teaching data in different modalities; studies the special training technology of domestic multimodal science education big models based on the teaching thinking chain, including knowledge enhancement driven by teaching semantics, collaborative feedback of teaching appropriateness and value alignment, collaborative enhancement of teaching appropriateness and logical accuracy, etc. Research the robust adaptive evaluation technology of scenario-based education big models, explore the evaluation of big model results and thinking chain reasoning processes for scenario-based application effects, and provide a basis for the iterative optimization of big models; thus constructing a special big model for science education based on the teaching thinking chain that integrates "teaching appropriateness-logical accuracy-value alignment".

[0063] Please refer to Figure 1 and Figure 2 , Figure 1 This is a flow chart of an embodiment of a scientific education method based on thinking chain provided by this application. Figure 2 It is a scenario diagram of a specific embodiment of the scientific education method based on thinking chain provided by this application.

[0064] The scientific education method based on thought chain of the present application is applied to the scientific education device, wherein the scientific education device of the present application can be a server, a terminal device, or a system in which the server and the terminal device cooperate with each other. Accordingly, the various parts included in the scientific education device, such as various units, sub-units, modules, and sub-modules, can all be set in the server, can all be set in the terminal device, or can be set in the server and the terminal device respectively.

[0065] Furthermore, the above-mentioned server can be hardware or software. When the server is hardware, it can be implemented as a distributed server cluster consisting of multiple servers, or it can be implemented as a single server. When the server is software, it can be implemented as multiple software or software modules, such as software or software modules used to provide distributed servers, or it can be implemented as a single software or software module, which is not specifically limited here.

[0066] like Figure 1 As shown, the specific steps are as follows:

[0067] Step S11: Obtain education input data.

[0068] In the embodiment of the present application, the scientific education device obtains the education input data input by the user, wherein the education input data can be input data of various modalities, including but not limited to: text modal data, image modal data, voice modal data, etc. Figure 2As shown, the user inputs text modal data, that is, a text question is asked: "You are a sixth-grade primary school student. Please answer this question: Use "84" disinfectant to prepare a liquid to disinfect white clothes. It is required to be diluted at a ratio of 1:200. Now you need to prepare 4020 grams of this liquid. How many grams of "84" disinfectant are needed?"

[0069] Furthermore, the user can also input at least two of the above three modal data at the same time, that is, multimodal education input data. At this time, since the semantic features of different modal data are in different feature spaces, they cannot be compared and fused. Therefore, this application provides a cross-scenario multimodal teaching semantic automatic alignment solution. For details, please refer to Figure 3 , Figure 3 It is a schematic diagram of the cross-scenario multimodal teaching semantic alignment process provided by this application.

[0070] like Figure 3 As shown, the cross-scenario multimodal teaching semantic alignment process of the present application is specifically an intelligent semantic analysis method for multimodal scientific education data, which provides a basis for subsequent data augmentation; designs a multimodal scientific education data evaluation and augmentation method, and solves the deficiencies found in semantic analysis by enhancing and expanding teaching data, thereby enriching teaching data; proposes multidimensional teaching semantic perception and cross-modal alignment technology to ensure the semantic consistency of teaching data of different modalities in scientific disciplines, and provide high-quality data support for improving the subsequent training effect of the scientific education large model.

[0071] Among them, Figure 3 As shown, the semantic alignment process of this application mainly includes the following three parts: intelligent semantic analysis of multimodal scientific education data, evaluation and amplification of multimodal scientific education data, and multidimensional teaching semantic perception and cross-modal alignment.

[0072] For more information on intelligent semantic analysis of multimodal scientific education data, please continue to refer to Figure 4 , Figure 4 It is a schematic diagram of the multimodal science education semantic extraction process provided by this application.

[0073] like Figure 4 As shown in the figure, the scientific education device develops a multimodal scientific education data segmentation and decomposition tool to automatically deconstruct multimodal data. Secondly, the scientific education device extracts key semantic elements from multimodal scientific data. Then, the scientific education device analyzes the quality assessment requirements of multimodal data and defines an evaluation index system suitable for scientific education data, including key indicators such as data integrity, accuracy, consistency, and relevance. Finally, the scientific education device constructs a multimodal data quality assessment model to ensure that the multimodal training data has high reliability and high quality, and realizes intelligent semantic analysis of scientific education data.

[0074] Among them, regarding the evaluation and amplification of multimodal scientific education data, the scientific education device combines the multimodal data quality evaluation model designed in the previous step to evaluate the quality of the multimodal scientific education data input into the large model. The amplification method is to automatically extract and analyze the existing multimodal scientific education data to ensure a comprehensive understanding of the characteristics of the scientific education data.

[0075] Specifically, the scientific education device can use the multimodal data enhancement method in the feature space to implement the data augmentation strategy, thereby rapidly expanding the volume of multimodal data. Among them, the multimodal data enhancement method can specifically create a new enhanced data by mixing the data features of two samples from the same class. Finally, the scientific education device uses the multimodal data quality automated assessment model to conduct quality assessments including data completeness, accuracy, consistency, and relevance, so as to adjust the augmentation strategy and gradually improve the quality and diversity of the data set.

[0076] Among them, regarding multi-dimensional teaching semantic perception and cross-modal alignment, first, the scientific education device constructs a multi-dimensional teaching semantic deep perception model to improve the semantic understanding ability of the scientific education large model, and realizes accurate perception and dynamic analysis of teaching semantics. Secondly, the scientific education device performs feature representation learning on the multi-modal data to be trained, and realizes the formation of a unified semantic representation of data in multiple modalities such as text, image, audio, and video, providing a basis for cross-modal semantic alignment. Then, the scientific education device maps data of different modalities to a shared embedding space, and then calculates the similarity between them in this space. Specifically, the different modal data are mapped to a potential common subspace through the corresponding deep learning network, in which the cross-modal semantic association similarity of different modal data can be directly measured. Finally, the scientific education device realizes the alignment between different modalities at the semantic level, solves the semantic differences between different modal data, and realizes efficient and accurate cross-modal semantic alignment.

[0077] Step S12: Using the data features of the educational input data, determine a number of first-level candidate thinking chains.

[0078] In the embodiment of the present application, after the scientific education device extracts the data features of the education input data input by the user in step S11, it further constructs a teaching thinking chain. The data features of the education input data can be data features directly extracted from single-modal data, or data features obtained by fusion of multi-modal data through semantic alignment.

[0079] Specifically, the scientific education device automatically constructs a teaching thinking chain that conforms to the cognition of scientific education experts based on the automatic alignment of cross-scenario multimodal teaching semantics, hierarchical graphs, subject knowledge, teaching and learning behaviors, learning situations and other scientific education elements through contextual learning technology. Figure 2Continue reading Figure 5 , Figure 5 It is a schematic diagram of the automatic construction process of the teaching thinking chain provided by this application.

[0080] like Figure 5 As shown, the teaching thinking chain constructed by this application is in the form of a multi-level thinking chain. The output of the previous level thinking chain is the input of the next level thinking chain, and the output of the last level thinking chain is the teaching goal, which is the answer to the educational input data input by the user.

[0081] Specifically, the scientific education device will simultaneously conduct "reflection testing" at each step of the teaching behavior. When the planned teaching path does not meet the teaching appropriateness, other teaching paths will be re-executed. Figure 5 As shown, Figure 5 The “×” in the figure indicates a teaching path that does not meet the teaching appropriateness. Only when the entire teaching path meets the teaching appropriateness can the teaching logic be used to output the final teaching goal. The scientific education device can also output the existing problems of the teaching path that does not meet the teaching appropriateness, such as “the knowledge point is beyond the syllabus, inappropriate, adjust the teaching link”, etc.

[0082] Combination Figure 2 As shown, in the first reflective check of the step-by-step explanation, the scientific education device analyzed the teaching appropriateness of the teaching thinking chain in step 1 and found that the problems with the teaching knowledge used in the teaching thinking chain in step 1 were beyond the syllabus.

[0083] The scientific education device repeatedly executes the construction of thinking chains at all levels until finally establishing a single teaching thinking chain construction under a specific teaching and learning scenario.

[0084] Specifically, in order to improve the teaching appropriateness of the teaching thinking chain at all levels, this application introduces the large model optimization technology of "teaching appropriateness-value alignment" synergistic enhancement. Figure 6 and Figure 7 , Figure 6 This is a schematic diagram of the large model optimization process of "teaching appropriateness-value alignment" synergistic enhancement provided by this application. Figure 7 yes Figure 1 The specific flow chart of step S12 of the scientific education method based on the thinking chain is shown.

[0085] In this application, the scientific education device uses the pre-trained language model to generate thought chains at all levels to construct the final teaching thought chain. The training of the language model is as follows: Figure 6 and Figure 7 As shown, the scientific education device realizes the heuristic expansion of the dynamic teaching behavior chain based on the existing subject knowledge graph, and strengthens the teaching semantic driving ability of the large model through a training program that integrates teaching behavior and teaching knowledge.

[0086] like Figure 7 As shown, the specific steps are as follows:

[0087] Step S121: Input the subject knowledge data, teaching and learning behavior data and training samples into the language model to obtain the teaching suitability evaluation value and educational value evaluation value of the training samples.

[0088] In the embodiments of the present application, Figure 6 As shown, the scientific education device inputs the training samples, as well as the pre-input subject knowledge and teaching and learning behaviors into the language model in the form of data, so as to obtain the teaching appropriateness evaluation value and the educational value evaluation value of the thinking chain generated by analyzing the training samples.

[0089] Among them, teaching appropriateness and educational values ​​are two different dimensions of judging the thinking chain, and the specific judging rules can be set with reference to expert experience. Specifically, expert guidance can also occur during the training process of the language model, that is, experts can use scientific education devices to adjust the teaching appropriateness evaluation value and educational value evaluation value output during the training process of the language model, so as to iteratively optimize the language model in the direction required by the user. By introducing expert knowledge, the training effect and accuracy of the language model can also be further improved.

[0090] Furthermore, experts can also use scientific education devices to make real-time adjustments to positive and negative samples in training samples to accelerate the iterative convergence of large language models.

[0091] Step S122: Use the teaching suitability evaluation value and the educational value evaluation value to train the language model to update the parameters of the language model.

[0092] In an embodiment of the present application, the scientific education device inputs the teaching suitability evaluation value and the educational value evaluation value of the training sample into the collaborative reward model, obtains the training reward parameters output by the collaborative reward model, and finally uses the training reward parameters to feed back to the language model for parameter optimization.

[0093] Specifically, the scientific education device makes full use of the data from scientific education experts' feedback on teaching appropriateness and value alignment, constructs multiple collaborative reward models, designs proximal strategy optimization algorithms, and collaboratively optimizes model parameters to further achieve collaborative enhancement of the large model in teaching appropriateness and value alignment.

[0094] Step S13: using the preset scientific education prior information, obtaining the first evaluation result of each first-level candidate thinking chain.

[0095] In an embodiment of the present application, the scientific education device uses preset scientific education prior information, such as expert knowledge, to judge candidate thinking chains at all levels to determine the thinking chain with the best evaluation result or the qualified evaluation result as the teaching thinking chain in the teaching logic.

[0096] Specifically, in order to improve the logical accuracy of the teaching thinking chain at all levels, this application introduces a large model optimization technology that synergistically enhances "teaching appropriateness-logical accuracy". Figure 8 and Fig. 9 , Figure 8 This is a schematic diagram of the large model optimization process of "teaching appropriateness-logical accuracy" synergistic enhancement provided by this application. Fig. 9 yes Figure 1 The specific flow chart of step S13 of the scientific education method based on the thinking chain is shown.

[0097] like Figure 8 As shown, the scientific education device uses a self-evaluation model to evaluate several candidate thinking chains generated by the language model at each level, so as to select the optimal thinking chain at each level through reinforcement learning to construct the final teaching thinking chain.

[0098] like Fig. 9 As shown, the specific steps are as follows:

[0099] Step S131: input the preset scientific education prior information and each first-level candidate thinking chain into the self-evaluation model, and extract the teaching suitability evaluation value and logical reasoning evaluation value of the first-level candidate thinking chain.

[0100] In the embodiment of the present application, the scientific education device generates a step-level multi-path candidate teaching thinking chain (teaching thinking chain candidate set = thinking chain 1, thinking chain 2, ..., thinking chain n) based on the language model of step S12. Each thinking chain continues to execute on the basis of the previous step logic, improving the thinking chain solution space, and each step of the thinking chain produces a result output, recorded as Ln.

[0101] The scientific education device inputs each step of the teaching thinking chain into the self-evaluation model, and through the self-evaluation model based on "teaching appropriateness-logical accuracy", it realizes the evaluation and judgment of the step-level thinking chain:

[0102] Evaluation result output = self-evaluation model {Ln, teaching appropriateness, logical reasoning, subject knowledge, teaching and learning behavior}

[0103] Step S132: Based on the teaching suitability evaluation value and the logical reasoning evaluation value of the first-level candidate thinking chain, output the first evaluation result of the first-level candidate thinking chain.

[0104] In an embodiment of the present application, the scientific education device constructs a collaborative enhancement scheme for the step-level thinking chain, adds the evaluation results of the previous step to the reinforcement learning process, and realizes the collaborative enhancement of the large model in terms of teaching appropriateness and logical accuracy through logically controllable step-by-step teaching thinking chain training and reasoning.

[0105] Specifically, Figure 8 As shown, the self-evaluation model of the science education device evaluates each candidate thinking chain of the same level in terms of teaching appropriateness and logical reasoning. The evaluation results include qualified (√) and unqualified (×).

[0106] Step S14: Determine the first-level candidate thinking chain that is judged to be qualified as the first-level teaching thinking chain.

[0107] In the embodiment of the present application, the scientific education device determines the candidate thinking chain determined by the self-evaluation model as qualified as the teaching thinking chain of the current level. If there are multiple qualified candidate thinking chains, the scientific education device can select the candidate thinking chain with the highest evaluation score as the teaching thinking chain of the current level, and the scientific education device can determine multiple qualified candidate thinking chains as teaching thinking chains on the other hand, and eliminate unqualified teaching paths by whether the teaching thinking chains of subsequent levels meet the teaching suitability, and finally determine the qualified teaching path as the teaching logic.

[0108] Furthermore, since the language big model can continuously expand new thinking chains, the instructions for each level of teaching thinking chains also need to be continuously expanded, where the instructions are prompts used to describe the problem and inform the big model of the output format.

[0109] Please refer to Fig.10 and Fig.11 , Fig.10 It is a schematic diagram of the instruction set expansion process based on the teaching thinking chain provided by this application. Fig.11 It is a flow chart of another embodiment of the scientific education method based on thinking chain provided by this application.

[0110] Step S21: Generate a seed instruction set using the first-level teaching thinking chain and data features.

[0111] In the embodiment of the present application, the scientific education device inputs the teaching thinking chain constructed in the previous step into the instruction set expansion process, and obtains the instruction fine-tuning input-output pair data for the scientific education scenario based on the instruction synthesis technology to construct the seed instruction set:

[0112] Seed instruction set L = instruction synthesis {multi-modal input, teaching thinking chain}

[0113] Finally, the scientific education device uses instruction evolution technology to expand the seed instruction set and filter the low-quality instruction fine-tuning data, outputting a high-quality instruction set for subsequent training.

[0114] Step S22: Filter and generate prompt information instructions of the first-level teaching thinking chain from the seed instruction set.

[0115] In the embodiment of the present application, after the scientific education device determines the teaching thinking chains at each level, it searches for the instruction output with the highest matching degree in the seed quality set according to the determined teaching thinking chains. Figure 2 Prompt for a step-by-step explanation of each step shown.

[0116] Step S15: Obtain the output results of the first-level teaching thinking chain for data features.

[0117] Step S16: Using the output results, determine a number of second-level candidate thought chains.

[0118] Step S17: Using the preset scientific education prior information, obtain the evaluation results of each second-level candidate thinking chain.

[0119] In the embodiment of the present application, the process of determining the teaching thinking chain at each level is the same and will not be repeated here.

[0120] Step S18: Determine the second-level candidate thinking chain that is judged to be qualified as the second-level teaching thinking chain, until the last-level teaching thinking chain outputs a scientific education answer to the education input data.

[0121] In the present application, the scientific education device obtains educational input data; uses the data features of the educational input data to determine a number of first-level candidate thinking chains; uses the preset scientific education prior information to obtain the first evaluation result of each first-level candidate thinking chain; determines the first-level candidate thinking chain with a qualified evaluation result as the first-level teaching thinking chain; obtains the output result of the first-level teaching thinking chain for the data features; uses the output result to determine a number of second-level candidate thinking chains; uses the preset scientific education prior information to obtain the evaluation result of each second-level candidate thinking chain; determines the second-level candidate thinking chain with a qualified evaluation result as the second-level teaching thinking chain, until the last-level teaching thinking chain outputs the scientific education answer to the educational input data. Through the above-mentioned scientific education method, the teaching behavior is disassembled, the search is performed step by step and reflected, the errors or unreasonableness of the steps are found, and the new teaching path is re-executed to meet the appropriateness and coherence of the scientific teaching process and improve the scientific education effect.

[0122] Those skilled in the art will appreciate that, in the above method of specific implementation, the order in which the steps are written does not imply a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of the steps should be determined by their functions and possible internal logic.

[0123] In order to realize the above scientific education method, this application also proposes a scientific education device, please refer to Fig.12 , Fig.12 It is a structural schematic diagram of an embodiment of a scientific education device provided by the present application.

[0124] The science education device 400 of this embodiment includes a processor 41 , a memory 42 , an input / output device 43 , and a bus 44 .

[0125] The processor 41, memory 42, and input / output device 43 are respectively connected to a bus 44. The memory 42 stores program data, and the processor 41 is used to execute the program data to implement the scientific education method described in the above embodiment.

[0126] In the embodiment of the present application, the processor 41 may also be referred to as a CPU (Central Processing Unit). The processor 41 may be an integrated circuit chip having the ability to process signals. The processor 41 may also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gates or transistor logic devices, or discrete hardware components. A general-purpose processor may be a microprocessor or the processor 41 may also be any conventional processor, etc.

[0127] This application also provides a computer storage medium, please continue to refer to Fig.13 , Fig.13 It is a structural diagram of an embodiment of a computer storage medium provided in the present application. The computer storage medium 600 stores a computer program 61. When the computer program 61 is executed by a processor, it is used to implement the scientific education method of the above embodiment.

[0128] When the embodiments of the present application are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk.

[0129] The above description is only an implementation method of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly used in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A scientific education method based on thinking chain, characterized in that: The scientific education methods include: Obtaining educational input data; Determine a number of first-level candidate thinking chains using the data features of the educational input data; Using the preset scientific education prior information, obtain the first evaluation result of each first-level candidate thinking chain; Determine the first-level candidate thinking chain with qualified evaluation results as the first-level teaching thinking chain; Obtaining the output result of the first-level teaching thinking chain for the data feature; Using the output results, determining a number of second-level candidate thinking chains; Using the preset scientific education prior information, obtain the evaluation results of each second-level candidate thinking chain; The second-level candidate thinking chain that is judged to be qualified is determined as the second-level teaching thinking chain, until the last-level teaching thinking chain outputs the scientific education answer to the education input data.

2. The scientific education method according to claim 1, characterized in that: The method of using the preset scientific education prior information to obtain the first evaluation result of each first-level candidate thinking chain includes: Input the preset scientific education prior information and each first-level candidate thinking chain into the self-evaluation model, and extract the teaching suitability evaluation value and logical reasoning evaluation value of the first-level candidate thinking chain; Based on the teaching suitability evaluation value and the logical reasoning evaluation value of the first-level candidate thinking chain, output a first evaluation result of the first-level candidate thinking chain; The preset scientific education prior information includes subject knowledge data and / or teaching and learning behavior data.

3. The scientific education method according to claim 2, characterized in that: The step of determining the first-level candidate thinking chain with a qualified evaluation result as the first-level teaching thinking chain includes: In response to the existence of at least two first-level candidate thinking chains with qualified evaluation results, the teaching suitability evaluation value and / or the highest logical reasoning evaluation value among the first-level candidate thinking chains with qualified evaluation results are obtained and determined as the first-level teaching thinking chain.

4. The scientific education method according to claim 1, characterized in that: The method utilizes the data features of the educational input data to determine a number of first-level candidate thinking chains, including: Inputting the data features of the educational input data into the language macro model to generate a number of first-level candidate thought chains; The scientific education method further comprises: Inputting subject knowledge data, teaching and learning behavior data and training samples into the language macro model, and obtaining teaching suitability evaluation values ​​and educational value evaluation values ​​of the training samples; The teaching suitability evaluation value and the educational value evaluation value are used to train the language model to update parameters of the language model.

5. The scientific education method according to claim 4, characterized in that: After obtaining the teaching suitability evaluation value and the educational value evaluation value of the training sample, the scientific education method further includes: Based on the expert guidance instructions, the teaching appropriateness evaluation value and / or the educational value evaluation value are updated.

6. The scientific education method according to claim 5, characterized in that: The step of training the language model by using the teaching suitability evaluation value and the education value evaluation value to update parameters of the language model includes: Input the teaching suitability evaluation value and the educational value evaluation value into a collaborative reward model to obtain training reward parameters; The training reward parameter is used to feed back the language macro model.

7. The scientific education method according to claim 1, characterized in that: After determining the first-level candidate thinking chain with the qualified evaluation result as the first-level teaching thinking chain, the scientific education method further includes: Generate a seed instruction set using the first-level teaching thinking chain and the data features; The prompt information instructions of the first-level teaching thinking chain are filtered and generated from the seed instruction set.

8. The scientific education method according to claim 1, characterized in that: The educational input data includes: at least two of text modality data, image modality data, and voice modality data; After obtaining the educational input data, the scientific education method includes: extracting multimodal data features of the educational input data; Mapping the multimodal data features to a shared embedding space; Extracting data features of the multimodal data features in the shared embedding space.

9. A scientific education device, characterized in that: The scientific education device includes a memory and a processor coupled to the memory; Wherein, the memory is used to store program data, and the processor is used to execute the program data to implement the scientific education method as described in any one of claims 1 to 8.

10. A computer storage medium, characterized in that: The computer storage medium is used to store program data, and when the program data is executed by a computer, it is used to implement the scientific education method as described in any one of claims 1 to 8.

Images