Answer generation method and related device, equipment and storage medium

Abstract

The present application discloses an answer generation method and related devices, equipment, and storage media. Among them, the answer generation method includes: generating an initial answer to the question to be answered; obtaining a number of knowledge points involved in the question-and-answer data and the explanatory descriptions of each knowledge point; obtaining the feedback result of the target object on the comprehensibility of the explanatory description of the knowledge point; in response to the feedback result indicating incomprehensibility, respectively select each knowledge point for which the feedback result indicates incomprehensibility as the target knowledge point, and re-obtain the explanatory description of the target knowledge point, and return to the step of obtaining the feedback result of the target object on the comprehensibility of the explanatory description of the knowledge point; in response to the feedback results of each knowledge point all indicating comprehensibility, generate a final answer to the question to be answered based on the question to be answered and the explanatory descriptions of each knowledge point. The above solution can improve the comprehensibility of the generated answer.

Description

Technical Field

[0001] This application relates to the technical field of natural language processing, and particularly to a method for generating answers and related devices, equipment, and storage media. Background Art

[0002] In recent years, thanks to the rapid development of generative models such as large language models, intelligent question-and-answer technology has been widely applied in many fields such as education, office work, and medical care.

[0003] However, existing intelligent question-and-answer technology often targets relevant personnel who already have a certain knowledge reserve, such as professionals. However, relevant personnel who have not yet had a certain knowledge reserve, such as non-professionals, are often limited by factors such as cognitive level and are often difficult to understand the output answer content. Taking the education scenario as an example, existing popular science question-and-answer technology is usually applicable to students who are older or have a certain knowledge reserve, while primary school students often have difficulty understanding the output popular science answers. In view of this, how to improve the comprehensibility of the generated answers has become an urgent problem to be solved. Summary of the Invention

[0004] The main technical problem to be solved by this application is to provide a method for generating answers and related devices, equipment, and storage media, which can improve the comprehensibility of the generated answers.

[0005] To solve the above technical problem, a first aspect of this application provides a method for generating answers, including: generating an initial answer to the question to be answered; obtaining a number of knowledge points involved in the question-and-answer data and the explanatory descriptions of each knowledge point; where the question-and-answer data includes at least the question to be answered and the initial answer; obtaining the feedback result of the comprehensibility of the explanatory description of the knowledge point by the target object; in response to the feedback result indicating incomprehensibility, respectively select each knowledge point with the feedback result indicating incomprehensibility as the target knowledge point, and re-obtain the explanatory description of the target knowledge point, and return to the step of obtaining the feedback result of the comprehensibility of the explanatory description of the knowledge point by the target object; in response to the feedback result of each knowledge point indicating comprehensibility, generate a final answer to the question to be answered based on the question to be answered and the explanatory descriptions of each knowledge point.

[0006] To solve the above technical problems, the second aspect of the present application provides an answer generation device, including: an initial answer generation module, a question-and-answer data analysis module, an explanation understanding feedback module, an interactive loop iteration module, and a final answer generation module. The initial answer generation module is used to generate an initial answer to the question to be answered. The question-and-answer data analysis module is used to obtain a number of knowledge points involved in the question-and-answer data and the explanatory descriptions of each knowledge point. Among them, the question-and-answer data at least includes the question to be answered and the initial answer. The explanation understanding feedback module is used to obtain the feedback result of the understandability of the explanatory description of the knowledge point by the target object. The interactive loop iteration module is used to, in response to the feedback result indicating incomprehensibility, respectively select each knowledge point with the feedback result indicating incomprehensibility as the target knowledge point, and re-obtain the explanatory description of the target knowledge point, and return to the step of obtaining the feedback result of the understandability of the explanatory description of the knowledge point by the target object. The final answer generation module is used to, in response to the feedback results of all knowledge points indicating comprehensibility, generate a final answer to the question to be answered based on the question to be answered and the explanatory descriptions of each knowledge point.

[0007] To solve the above technical problems, the third aspect of the present application provides an electronic device, at least including a memory and a processor coupled to each other. The memory stores at least program instructions, and the processor is used to execute the program instructions to implement the answer generation method in the first aspect above.

[0008] To solve the above technical problems, the fourth aspect of the present application provides a computer-readable storage medium, storing program instructions that can be run by a processor, and the program instructions are used to implement the answer generation method in the first aspect above.

[0009] For the above solution, an initial answer to the question to be answered is generated, and several knowledge points involved in the Q&A data and the explanatory descriptions of each knowledge point are obtained. The Q&A data includes at least the question to be answered and the initial answer. Then, the feedback result of the target object on the comprehensibility of the explanatory description of the knowledge point is obtained. Thus, in response to the feedback result indicating incomprehensibility, each knowledge point with a feedback result indicating incomprehensibility is selected as the target knowledge point, and the explanatory description of the target knowledge point is obtained again. The step of obtaining the feedback result of the target object on the comprehensibility of the explanatory description of the knowledge point is repeated. In response to the feedback result of each knowledge point indicating comprehensibility, based on the question to be answered and the explanatory descriptions of each knowledge point, a final answer to the question to be answered is generated. Therefore, on the one hand, when generating the final answer to the question to be answered, by combining the explanatory descriptions of each knowledge point involved in the Q&A data, it is possible to reflect the explanatory descriptions of the knowledge points in the final answer, which helps to improve the comprehensibility of the generated answer. On the other hand, during the Q&A process, the feedback result of the target object on the comprehensibility of the explanatory description of the knowledge point is also referred to. When the feedback result indicates incomprehensibility, the explanatory description of the knowledge point is obtained again and iterated until the feedback result of the target object on its comprehensibility indicates comprehensibility. This can ensure that the target object can understand the explanatory descriptions of each knowledge point when generating the final answer, which helps to improve the comprehensibility of the final answer as much as possible. Therefore, the comprehensibility of the generated answer can be improved. Brief Description of the Drawings

[0010] Figure 1 is a schematic flowchart of an embodiment of the answer generation method of the present application;

[0011] Figure 2 is a schematic process diagram of an embodiment of the answer generation method of the present application;

[0012] Figure 3 is a schematic framework diagram of an embodiment of the answer generation device of the present application;

[0013] Figure 4 is a schematic framework diagram of an embodiment of the electronic device of the present application;

[0014] Figure 5 is a schematic framework diagram of an embodiment of the computer-readable storage medium of the present application. Detailed Description of the Embodiment

[0015] The solution of the embodiment of the present application will be described in detail below with reference to the accompanying drawings of the specification.

[0016] In the following description, specific details such as specific system structures, interfaces, and technologies are presented for the purpose of illustration rather than limitation, so as to thoroughly understand the present application.

[0017] The terms "system" and "network" are often used interchangeably in this document. The term " / or" in this document is merely a description of the relationship between associated objects, indicating that there can be three relationships. For example, A / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the segment " / " in this document generally represents an "or" relationship between the associated objects before and after. Furthermore, "multiple" in this document means two or more than two.

[0018] Please refer to Figure 1 , Figure 1 which is a schematic flowchart of an embodiment of the method for generating answers in this application. Specifically, it may include the following steps:

[0019] Step S11: Generate an initial answer to the question to be answered.

[0020] In an implementation scenario, the question to be answered can be obtained through input such as text, voice, etc., and the acquisition method of the question to be answered is not limited here.

[0021] In an implementation scenario, the specific content of the question to be answered can be set according to the application scenario. For example, taking the education scenario as an example, the question to be answered includes scientific questions (such as, "What is a black hole composed of?", "Is the effect of extinguishing fire with hot water the same as that with cold water?", "Why does ice float on water?", etc.), or, taking the medical scenario as an example, the question to be answered can include consultation questions (such as, "What changes in physical health does hair turning white indicate?", etc.). Of course, the above examples are only several possible examples of the question to be answered taking the education scenario and the medical scenario as examples, and do not limit the specific content of the question to be answered accordingly. The specific content of the question to be answered will not be exemplified one by one here.

[0022] In an implementation scenario, in order to obtain an initial answer to a question to be answered, as a possible example, the question to be answered can be input into a large language model to obtain the output answer of the large language model as the initial answer. Alternatively, as another possible example, a search can be first performed based on the question to be answered (e.g., a search can be performed in external knowledge sources such as scientific literature databases, mainstream academic journals, and authoritative websites) to obtain the reference data required to answer the question to be answered (e.g., the latest data, research results, or theoretical explanations related to the question to be answered). Based on this, a prompt instruction can be constructed in combination with the question to be answered and the reference data, and the prompt instruction is used to instruct the large language model to answer the question to be answered in combination with the reference data, so as to guide the large language model to generate a more accurate initial answer through the reference data, and to avoid the "hallucination" phenomenon of the large language model itself as much as possible. The prompt instruction is then input into the large language model to obtain the output answer of the large language model as the initial answer, so that the large language model can give a more scientific answer based on authoritative knowledge, greatly reducing the risk of scientific errors when the model gives an isolated answer. For ease of description, in this case, the initial answer can be expressed as:

[0023] …… (1)

[0024] In the above formula (1), Indicates the initial answer, represents a large language model, Indicates questions to be answered. In addition, the large language model may specifically include but is not limited to open source models such as LLAMA, Bloom, etc., or the large language model may be obtained by fine-tuning parameters based on a specific corpus and an open source large model, or the large language model may also be a custom large model, and the specific source of the large language model is not limited here.

[0025] Step S12: Obtain several knowledge points involved in the question and answer data and explanations of each knowledge point.

[0026] In the disclosed embodiment, the question and answer data may include at least the question to be answered and the initial answer. For example, still taking the education scenario as an example, the question to be answered is "What is a black hole made of?", and the initial answer may be "A black hole is a celestial body formed by the gravitational collapse of a massive star after a supernova explosion. A black hole consists of an event horizon and a singularity. The event horizon is the boundary of the black hole, and the singularity is the core of the black hole...", then the two can constitute the question and answer data for further analysis of the question and answer data. Of course, the above example is only a possible example of question and answer data when taking the education scenario as an example. The specific content of the question and answer data is not limited here, and examples are not given one by one.

[0027] In an implementation scenario, as a possible example, after obtaining the Q&A data, it is possible to analyze the Q&A data to obtain a number of knowledge points and extract the explanatory descriptions of each knowledge point from the Q&A data. Still taking the aforementioned question to be answered, which is the scientific question "What is a black hole composed of?" as an example, the knowledge points can include but are not limited to at least one of scientific concepts and scientific phenomena. Then, it is possible to analyze the Q&A data to obtain a number of knowledge points: "black hole", "event horizon", "singularity", and extract from the Q&A data the descriptive explanation of the knowledge point "black hole": "A black hole is an object formed by the gravitational collapse of a massive star after a supernova explosion", the descriptive explanation of the knowledge point "event horizon": "The event horizon is the boundary of a black hole", and the descriptive explanation of the knowledge point "singularity": "The singularity is the core of a black hole". Of course, the above example is only a possible example of knowledge points and their descriptive explanations in the actual application process, and other possible situations will not be exemplified one by one here.

[0028] In a specific implementation scenario, the Q&A data can be analyzed through a large language model to obtain a number of knowledge points. Exemplarily, a prompt instruction can be constructed based on the Q&A data, and the prompt instruction is used to instruct the large language model to extract the knowledge points involved in answering the question to be answered from the Q&A data, and then input the prompt instruction into the large language model to obtain the output result of the large language model, thus obtaining a number of knowledge points. It should be noted that for the specific source of the large language model, reference can be made to the aforementioned relevant description of the large language model, which will not be elaborated here.

[0029] In another specific implementation scenario, in addition to the aforementioned implementation method of analyzing a number of knowledge points through a large language model, the Q&A data can also be analyzed through a pre-trained deep learning model to obtain a number of knowledge points. Exemplarily, the deep learning model can include but is not limited to BERT (Bidirectional Encoder Representations from Transformers, bidirectional encoding representation based on Transformer), etc., and the specific architecture of the deep learning model is not limited here. Specifically, sample Q&A data can be obtained in advance, and the sample Q&A data is labeled with a number of sample knowledge points involved, and the sample Q&A data is processed based on the deep learning model to obtain a number of predicted knowledge points. Then, based on the difference between the predicted knowledge points and the sample knowledge points, the network parameters of the deep learning model are adjusted, and the deep learning model can be trained to learn how to analyze the knowledge points involved in answering the question to be answered from the Q&A data.

[0030] In another implementation scenario, as another possible example, after obtaining the question and answer data, it is also possible to analyze based on the question and answer data to obtain several candidate knowledge points. For example, the question and answer data can be analyzed by a large language model, a pre-trained deep learning model, etc., to obtain several knowledge points as candidate knowledge points. For details, please refer to the above-mentioned related descriptions, which will not be repeated here. In addition, during the analysis process, the hierarchical decision-making strategy in reinforcement learning can also be used to determine the core elements in the question and answer data. For details, please refer to the technical details of the hierarchical decision-making strategy in reinforcement learning, which will not be repeated here. On this basis, each candidate knowledge point related to answering the question to be answered can be selected as several knowledge points involved in the question and answer data, that is, the knowledge points involved in answering the question to be answered, and the initial explanation of each knowledge point in the question and answer data is extracted, and the professional explanation of each knowledge point is retrieved separately, and then for each knowledge point, the initial explanation is corrected based on the professional explanation to obtain the explanation description of the corresponding knowledge point. The above method, by selecting candidate knowledge points related to the question to be answered and using professional explanations to correct the extracted initial explanations, can, on the one hand, avoid redundant answers as much as possible and help improve the conciseness of the generated answers; on the other hand, it can avoid inaccurate, imprecise or factually incorrect answers as much as possible.

[0031] In a specific implementation scenario, taking the question to be answered "Why does ice float on the water?" as an example, the initial answer can be "Ice floats on the water because its density is lower than that of water." Then, the question and answer data can be combined for analysis to obtain several candidate knowledge points "density", "density of ice", and "density of water". Since these three are all related to the question to be answered, several knowledge points involved in answering the question to be answered can be obtained: "density", "density of ice", and "density of water". The preceding concepts to be introduced to answer the question to be answered (that is, "density", "density of ice", and "density of water") can be introduced through knowledge point analysis. Alternatively, taking the question to be answered "Why does hair turn white?" as an example, the initial answer can be "Hair turns white because of the reduction of melanin, which is like pigment that can help our hair turn black. In addition, hormonal disorders in the body due to thyroid disease may also cause hair to turn white...", then the question and answer data can be combined for analysis to obtain several candidate knowledge points: "melanin", "thyroid", and "hormones", but the candidate knowledge points "thyroid" and "hormones" are not related to the question to be answered, and only the candidate knowledge point "melanin" is related to the question to be answered. Therefore, the candidate knowledge point "melanin" can be selected as the knowledge point involved in answering the question to be answered, and the knowledge points involved in the redundant answer to the question to be answered (i.e. "thyroid" and "hormones") in the initial answer can be excluded through knowledge point analysis.

[0032] In a specific implementation scenario, after extracting the initial explanations of each knowledge point from the Q&A data (for the specific extraction method and examples, refer to the foregoing relevant descriptions and will not be elaborated here; in addition, it is not excluded that the initial explanation of the knowledge point extracted from the Q&A data is "empty"), the initial explanation of the knowledge point can be corrected by combining the professional explanations retrieved for the knowledge point (for example, the professional explanations of the knowledge point can be retrieved from professional databases such as textbooks and technical dictionaries) to obtain the explanatory description of the knowledge point.

[0033] Step S13: Obtain the feedback result of the understandability of the explanatory description of the knowledge point by the target object.

[0034] In an implementation scenario, as a possible example, the target object can be an agent that simulates a real user. Exemplarily, when the real user is a primary school student, the target object can be an agent that simulates the cognitive level and comprehension ability of a primary school student. Of course, the above example is only a possible example of the agent in the actual application process, and does not limit the other possible configuration methods of the agent in other possible situations. The other possible configuration methods of the agent will not be elaborated one by one here. In addition, for the configuration method of the agent that simulates a real user, refer to the technical details of the agent and will not be elaborated here. On this basis, after obtaining the knowledge point and its explanatory description, an instruction to be executed can be constructed based on the knowledge point and the explanatory description of the knowledge point, and the instruction to be executed is used to execute the agent to provide feedback on the understandability of the knowledge point and its explanatory description, and then the instruction to be executed is input into the agent to obtain the output result of the agent as the feedback result. The above method can improve the feedback efficiency of understandability by implementing interactive feedback on understandability through the agent after obtaining the knowledge point and its explanatory description.

[0035] In a specific implementation scenario, the words or sentences that cannot be understood represent the words or sentences that the agent simulating a real user judges to be unable to be understood when simulating the real user's understanding in the explanatory description; while the expected explanation method represents the method that the agent simulating a real user expects to use to explain the knowledge point when judging that the explanatory description of the knowledge point cannot be understood, such as including but not limited to: giving examples, making analogies, telling stories, etc. The specific possible situations of the expected explanation method will not be elaborated one by one here.

[0036] In a specific implementation scenario, still taking the question to be answered "What is a black hole composed of?" as an example, among the several knowledge points involved, there is the knowledge point "black hole", and its explanatory description can include "A black hole is an astronomical object formed by the gravitational collapse of a massive star after a supernova explosion...". Then, this knowledge point and its explanatory description can be used to construct the instruction to be executed: "Please determine whether the explanatory description of the knowledge point 'black hole', which is 'A black hole is an astronomical object formed by the gravitational collapse of a massive star after a supernova explosion...', can be understood. If it cannot be understood, please point out the sentences that cannot be understood or the expected explanatory method." Then, input this instruction to be executed into the intelligent agent, and the feedback result of the intelligent agent can be obtained, such as "Cannot be understood. The sentences that cannot be understood are'supernova' and 'gravitational collapse', and the expected explanatory method is 'give examples'." Of course, the above example is only one possible example of the feedback result in the actual application process, and other possible situations of the feedback result will not be exemplified one by one here.

[0037] In another implementation scenario, as another possible example, the target object can also be a real user. After obtaining the knowledge point and its explanatory description, a prompt message can be constructed based on the knowledge point and its explanatory description. The prompt message is used to prompt the real user to provide feedback on the understandability of the knowledge point and its explanatory description. Then, the prompt message is output, and the input content of the real user in response to the prompt message is received as the feedback result. It should be noted that the real user can input the question to be answered through terminal devices such as mobile phones, learning machines, and tablet computers. The above terminal devices can be integrated with program instructions for the method flow of the embodiments of the present disclosure. Thus, after obtaining the knowledge point and its explanatory description, the knowledge point and its explanatory description can be encapsulated into a prompt message and output. For example, the prompt message can be output in at least one of the ways of text display and voice playback to prompt the user to provide feedback on the understandability of the explanatory description of the knowledge point. Exemplarily, still taking the aforementioned question to be answered as an example, for the knowledge point "black hole", the prompt message "The explanatory description of the knowledge point 'black hole' is 'A black hole is an object formed by the gravitational collapse of a massive star after a supernova explosion...'. Can you understand it? If not, please point out the sentences you don't understand or the expected way of explanation." can be constructed and output. Thus, the real user can input the feedback result in response to the prompt message, such as "Can't understand. The sentences I don't understand are'supernova' and 'gravitational collapse', and the expected way of explanation is 'giving examples'". Exemplarily, a drop-down menu or relevant options can be displayed on the aforementioned terminal device for the real user to select "Can understand" or "Can't understand". When the real user selects "Can't understand", the real user can be further prompted to input at least one of the sentences they don't understand and the expected way of explanation. Of course, the real user can also provide feedback on the understandability through voice input, and thus the feedback result of the real user on the understandability of the knowledge point and its explanatory description can be obtained through voice recognition. Of course, the above examples are only one possible example of the feedback result in the actual application process, and other possible situations of the feedback result will not be exemplified one by one here. Through the interactive feedback on the understandability with the real user after obtaining the knowledge point and its explanatory description, the above method can enhance the real user's sense of participation before the final answer is provided.

[0038] In yet another implementation scenario, as another possible example, the target object may further include a real user and an agent that simulates a real user. On the one hand, an instruction to be executed can be constructed based on the knowledge point and the explanatory description of the knowledge point, and the instruction to be executed is used to execute the agent's feedback on the understandability of the knowledge point and its explanatory description, and the instruction to be executed is input to the agent to obtain the output result of the agent. On the other hand, a prompt message can be constructed based on the knowledge point and the explanatory description of the knowledge point, and the prompt message is used to prompt the real user to give feedback on the understandability of the knowledge point and its explanatory description, and the prompt message is output. On this basis, it can be detected whether the input content in response to the prompt message from the real user is received within a preset time period (e.g., 10 seconds, etc.). If it is received, the input content can be used as the feedback result; otherwise, the output result of the agent can be used as the feedback result. Or, the output result of the agent and the input content of the real user in response to the prompt message can be mutually supplemented to obtain the feedback result. Or, the real user can also be prompted whether to give feedback on the understandability. For example, if the real user confirms not to give feedback, the output result of the agent can be directly used as the feedback result. Otherwise, if the real user confirms to give feedback, the input content of the real user in response to the prompt message can be waited for and, after being received, the input content of the real user can be used as the feedback result. Of course, the above three examples are only several possible ways to obtain the feedback result when the target object includes a real user and an agent that simulates a real user, and other possible ways in this case are not limited thereby, nor will they be exemplified one by one.

[0039] Step S14: In response to the feedback result indicating non-understandability, each knowledge point with a feedback result indicating non-understandability is separately selected as the target knowledge point, and the explanatory description of the target knowledge point is re-obtained, and the step of obtaining the feedback result of the understandability of the explanatory description of the knowledge point by the target object is returned.

[0040] In an implementation scenario, as a possible example, as mentioned above, when the feedback result is characterized as incomprehensible, the feedback result may further include at least one of the incomprehensible words and sentences and the expected explanation method. Then, a prompt instruction can be constructed based on the explanation description of the target knowledge point and the feedback result, and the prompt instruction is used to instruct the large language model to adjust the explanation description of the target knowledge point with reference to at least one of the incomprehensible words and sentences and the expected explanation method in the feedback result. Then, input the prompt instruction into the large language model to obtain the output content of the large language model as the new explanation description of the target knowledge point. For the sake of easy understanding, still taking the aforementioned question to be answered "What is a black hole made of?" as an example, as mentioned above, for the knowledge point "black hole" and its description and explanation, there is the following feedback result: "Incomprehensible, the incomprehensible words and sentences are'supernova' and 'gravitational collapse', and the expected explanation method is 'giving examples'". Then, the knowledge point "black hole" can be used as the target knowledge point, and combined with its explanation description and the above feedback result, a prompt instruction including but not limited to can be constructed:

[0041] The explanatory description of the target knowledge point "black hole" is that "a black hole is an object formed by the gravitational collapse after a massive star explodes in a supernova explosion...". In this explanatory description, there are two incomprehensible terms, "supernova" and "gravitational collapse". Please use the "giving examples" explanatory method to explain these two incomprehensible terms in order to output a new explanatory description of the target knowledge point "black hole". Imagine that the stars in the night sky are like the light bulbs in our lives. Most of them burn their nuclear fuel stably and emit a peaceful light. However, some stars will experience a spectacular and

[0042] It should be noted that the above example is only a possible example of the prompt instruction taking the question to be answered "What is a black hole made of?" as an example, and other possible situations will not be exemplified one by one here. Still taking the above example as an example, after the above prompt instruction is input into the large language model, the output content of the large language model can be obtained, such as including but not limited to:

[0043] disastrous "explosion" in the final stage of their lives. The brightness produced by this "explosion" is many times higher than the normal brightness of the star and can even be seen with the naked eye on distant Earth. This is a supernova. A black hole is an object formed by the gravitational collapse after a massive star explodes in a supernova explosion. So what is gravitational collapse? Gravitational collapse is like a game of piling up sand that we play. When we pile up the sand higher and higher, the gravitational force between the sand grains becomes greater and greater until a critical point is reached, and the sand pile can no longer remain stable and begins to collapse inward. Similarly, stars or interstellar matter will also undergo such an inward collapse process under the gravitational force of their own matter. This is gravitational collapse. Figure 2 Figure 2 Figure 2 Figure 2

[0044] It should be noted that the above example is only a possible example of the output content of the large language model taking the question to be answered "What is a black hole made of?" as an example, and other possible situations will not be exemplified one by one here. By constructing a prompt instruction based on the explanation description of the target knowledge point and the feedback result to instruct the large language model to adjust the explanation description of the target knowledge point with reference to at least one of the incomprehensible words and sentences and the expected explanation method in the feedback result, the general understanding ability of the large language model can be utilized to optimize the explanation description of the target knowledge point with reference to the feedback result, which helps to improve the comprehensibility of the explanation description of the target knowledge point and improve the optimization efficiency of the explanation description.

[0045] In another implementation scenario, as another possible example, as mentioned above, in the case where the feedback result is characterized as being incomprehensible, the feedback result may further include at least one of the incomprehensible words and sentences and the expected explanation method. Then, a preset knowledge base can be obtained, and the preset knowledge base may contain a number of preset knowledge points and their respective explanatory descriptions. The number of preset knowledge points includes the target knowledge point. Then, the explanatory description of the target knowledge point in the preset knowledge base is selected as the target description. Thus, based on at least one of the incomprehensible words and sentences and the expected explanation method in the feedback result, at least a part is extracted as the reference description from the target description. Furthermore, based on the reference description, the explanatory description of the target knowledge point can be adjusted to obtain a new explanatory description of the target knowledge point. Still taking the aforementioned question to be answered "What is a black hole composed of?" as an example, as mentioned above, for the knowledge point "black hole" and its descriptive explanation, there is the following feedback result: "Incomprehensible, the incomprehensible words and sentences are'supernova' and 'gravitational collapse', and the expected explanation method is 'giving examples'". Then, the knowledge point "black hole" can be used as the target knowledge point, and the explanatory description of the target knowledge point "black hole" in the preset knowledge base (for example, at least including the aforementioned example "Imagine that the stars in the night sky are like...") is used as the target description. Thus, based on the incomprehensible word "supernova" in the feedback result and the expected explanation method of "giving examples", "Imagine that the stars in the night sky are like... and can even be seen with the naked eye on distant Earth. This is a supernova" is extracted as the reference description from the target description. And based on the incomprehensible word "gravitational collapse" in the feedback result and the expected explanation method of "giving examples", "Gravitational collapse is like us playing a game of piling up sand... This is gravitational collapse" is extracted as the reference description from the target description. Furthermore, based on the above two paragraphs of reference descriptions, the explanatory description of the target knowledge point "black hole" can be adjusted to obtain a new explanatory description of the target knowledge point "black hole". It should be noted that the above example is only a possible example of selecting the target description and extracting the reference description by taking the question to be answered "What is a black hole composed of?" as an example, and other possible situations will not be exemplified one by one here. Through a series of operations such as obtaining the target description, extracting the reference description from the target description, and using the reference description to adjust the explanatory description, the above method obtains a new explanatory description of the target knowledge point, which can make the optimization of the explanatory description of the target knowledge point more targeted.

[0046] It should be noted that after obtaining the new explanatory description of the target knowledge point, for the new explanatory description of the target knowledge point, the steps of obtaining the feedback result of the understandability of the explanatory description of the knowledge point by the target object can be returned to obtain the feedback result of the understandability again for the new explanatory description of the target knowledge point. Such a cycle of iteration continues until the latest explanatory description of the target knowledge point is characterized as understandable. In addition, during this process, in order to improve the efficiency of answering questions, the explanatory descriptions of the knowledge points with feedback results of understandable do not need to obtain the feedback results of understandability again.

[0047] Step S15: In response to the feedback results of all knowledge points being characterized as understandable, generate the final answer to the question to be answered based on the question to be answered and the explanatory descriptions of all knowledge points.

[0048] Specifically, when the latest feedback results of all knowledge points are characterized as understandable, the final answer to the question to be answered can be generated based on the question to be answered and the latest explanatory descriptions of all knowledge points. Exemplarily, the question to be answered and the latest explanatory descriptions of all knowledge points can be combined to construct a prompt instruction, and the prompt instruction is used to instruct the large language model to answer the question to be answered with reference to the latest explanatory descriptions of all knowledge points. In this way, not only can it be ensured as much as possible that the final answer can answer the question to be answered, but also the final answer can be made to conform to the cognitive level and understanding ability of real users as much as possible, and it can also effectively guide the large language model to avoid generating "hallucinations" (that is, incorrect or fictional content) as much as possible. Then, input the prompt instruction into the large language model to obtain the output content of the large language model as the final answer to the question to be answered. Exemplarily, taking the question to be answered as a scientific question, the final answer can be expressed as:

[0049] ……(2)

[0050] In the above formula (2), represents the final answer, represents the large language model, represents the prompt applicable to scientific questions, represents the knowledge point, represents the explanatory description of the knowledge point, Represents a popular explanation of knowledge points. It should be noted that in the embodiments of the present disclosure, the popular explanation of knowledge points is obtained by regenerating a new explanatory description when the feedback result of the understandability of the knowledge points is characterized as not understandable. That is, the popular explanation of knowledge points can be considered as a part of the new explanatory description. For specific details, reference can be made to the foregoing related descriptions and their examples, which will not be elaborated here. It should be noted that in the embodiments of the present disclosure, the large language model can be replaced by an artificial intelligence model (the number of parameters of the artificial intelligence model is less than that of the large language model) obtained after knowledge distillation of the large language model, so as to improve the model response speed while still having high accuracy and professionalism. For specific details, reference can be made to the technical details of knowledge distillation, which will not be elaborated here.

[0051] As a possible example, please refer to Figure 2 , Figure 2 which is a schematic diagram of the process of an embodiment of the answer generation method of the present application. As Figure 3 shown, still taking the education scenario as an example, for the question to be answered "What is a black hole composed of?" which is a scientific question, the initial answer "A black hole is composed of an event horizon and a singularity..." of the question to be answered can be obtained by combining the retrieval enhancement technology of the large language model, and several knowledge points and their explanatory descriptions can be analyzed from the Q&A data through a knowledge extractor (such as, it can be a large language model, or a deep learning model such as BERT, etc. For specific details, reference can be made to the foregoing related descriptions), and the extracted explanatory descriptions can be corrected using the professional knowledge of the knowledge points in the professional knowledge base to improve the scientific nature of the explanatory descriptions of the knowledge points as much as possible. On this basis, for each knowledge point and its explanatory description (such as Figure 3 knowledge point 1 "black hole", knowledge point 2 "event horizon" and knowledge point 3 "singularity" and their explanatory descriptions of the three in ), the feedback result of the understandability of the target object can be obtained, and when the feedback result is characterized as not understandable, a new explanatory description of the knowledge point can be regenerated by the knowledge parser (such as a large language model, a preset knowledge base, etc. For specific details, reference can be made to the foregoing related descriptions) in combination with the feedback result until the feedback result of the understandability of the explanatory description of the knowledge point by the target object is characterized as understandable. Then, a prompt instruction can be generated by combining the question to be answered, the knowledge point, its explanatory description and the popular explanation. In addition, when generating the prompt instruction, the task description, output specification and examples, etc. can be further combined, which are not limited here. Finally, the prompt instruction can be input into the answer generator (such as, a large language model, for specific details, reference can be made to the foregoing related descriptions) to obtain the final answer to the question to be answered. Of course, Figure 4 shown is only a possible example of the generation process of the final answer when the question to be answered is the scientific question "What is a black hole composed of?", and other possible situations will not be exemplified one by one here.

[0052] In the above solution, an initial answer to the question to be answered is generated, and several knowledge points involved in the Q&A data and the explanatory descriptions of each knowledge point are obtained. The Q&A data includes at least the question to be answered and the initial answer. Then, the feedback result of the understandability of the explanatory description of the knowledge point by the target object is obtained. Thus, in response to the feedback result indicating incomprehensibility, each knowledge point with the feedback result indicating incomprehensibility is selected as the target knowledge point, and the explanatory description of the target knowledge point is obtained again, and the step of obtaining the feedback result of the understandability of the explanatory description of the knowledge point by the target object is returned. In response to the feedback result of each knowledge point indicating comprehensibility, based on the question to be answered and the explanatory descriptions of each knowledge point, a final answer to the question to be answered is generated. Therefore, on the one hand, when generating the final answer to the question to be answered, combining the explanatory descriptions of each knowledge point involved in the Q&A data can reflect the explanatory descriptions of the knowledge points in the final answer, which helps to improve the understandability of the generated answer. On the other hand, in the Q&A process, the feedback result of the understandability of the explanatory description of the knowledge point by the target object is also referred to. When it is indicated as incomprehensible, the explanatory description of the knowledge point is obtained again and iterated until the feedback result of the understandability of the target object to it indicates comprehensibility, which can ensure that the target object can understand the explanatory descriptions of each knowledge point when generating the final answer, and helps to improve the understandability of the final answer as much as possible. Therefore, the understandability of the generated answer can be improved.

[0053] Please refer to Figure 4 , Figure 5 which is a schematic framework diagram of an embodiment of the answer generation device of the present application. The answer generation device 30 includes: an initial answer generation module 31, a Q&A data analysis module 32, an explanatory understanding feedback module 33, an interactive loop iteration module 34, and a final answer generation module 35. The initial answer generation module 31 is used to generate an initial answer to the question to be answered. The Q&A data analysis module 32 is used to obtain several knowledge points involved in the Q&A data and the explanatory descriptions of each knowledge point. Among them, the Q&A data includes at least the question to be answered and the initial answer. The explanatory understanding feedback module 33 is used to obtain the feedback result of the understandability of the explanatory description of the knowledge point by the target object. The interactive loop iteration module 34 is used to, in response to the feedback result indicating incomprehensibility, select each knowledge point with the feedback result indicating incomprehensibility as the target knowledge point, and obtain the explanatory description of the target knowledge point again, and return to the step of obtaining the feedback result of the understandability of the explanatory description of the knowledge point by the target object. The final answer generation module 35 is used to, in response to the feedback result of each knowledge point indicating comprehensibility, generate a final answer to the question to be answered based on the question to be answered and the explanatory descriptions of each knowledge point.

[0054] In the above solution, the answer generation device 30 generates an initial answer to the question to be answered, obtains several knowledge points involved in the Q&A data and the explanatory descriptions of each knowledge point, and the Q&A data includes at least the question to be answered and the initial answer. Then, it obtains the feedback result of the target object on the comprehensibility of the explanatory description of the knowledge point. Thus, in response to the feedback result indicating incomprehensibility, it respectively selects each knowledge point with the feedback result indicating incomprehensibility as the target knowledge point, and re-obtains the explanatory description of the target knowledge point, and returns to the step of obtaining the feedback result of the target object on the comprehensibility of the explanatory description of the knowledge point. In response to the feedback result of each knowledge point indicating comprehensibility, based on the question to be answered and the explanatory descriptions of each knowledge point, it generates a final answer to the question to be answered. Therefore, on the one hand, when generating the final answer to the question to be answered, combining the explanatory descriptions of each knowledge point involved in the Q&A data can reflect the explanatory descriptions of the knowledge points in the final answer, which helps to improve the comprehensibility of the generated answer. On the other hand, in the Q&A process, it also refers to the feedback result of the target object on the comprehensibility of the explanatory description of the knowledge point, and when it indicates incomprehensibility, it re-obtains the explanatory description of the knowledge point and conducts iterative loops until the feedback result of the target object on its comprehensibility indicates comprehensibility, which can ensure that the target object can understand the explanatory descriptions of each knowledge point when generating the final answer, and helps to improve the comprehensibility of the final answer as much as possible. Therefore, it can improve the comprehensibility of the generated answer.

[0055] In some disclosed embodiments, the target object is an intelligent agent that simulates a real user. The interpretation and understanding feedback module 33 includes a first construction sub-module for constructing an instruction to be executed based on the knowledge point and the explanatory description of the knowledge point; wherein, the instruction to be executed is used to execute the feedback of the intelligent agent on the comprehensibility of the knowledge point and its explanatory description. The interpretation and understanding feedback module 33 includes a first acquisition sub-module for inputting the instruction to be executed to the intelligent agent to obtain the output result of the intelligent agent as the feedback result.

[0056] In some disclosed embodiments, the target object is a real user. The interpretation and understanding feedback module 33 includes a second construction sub-module for constructing a prompt message based on the knowledge point and the explanatory description of the knowledge point; wherein, the prompt message is used to prompt the real user to give feedback on the comprehensibility of the knowledge point and its explanatory description. The interpretation and understanding feedback module 33 includes a second acquisition sub-module for outputting the prompt message and receiving the input content of the real user in response to the prompt message as the feedback result.

[0057] In some disclosed embodiments, when the feedback result is characterized as being incomprehensible, the feedback result further includes at least one of an incomprehensible sentence and an expected explanation method. The interaction loop iteration module 34 includes a third construction sub-module for constructing a prompt instruction based on the explanation description of the target knowledge point and the feedback result. The prompt instruction is used to instruct the large language model to adjust the explanation description of the target knowledge point with reference to at least one of the incomprehensible sentence and the expected explanation method in the feedback result. The interaction loop iteration module 34 includes a third acquisition sub-module for inputting the prompt instruction into the large language model to obtain the output content of the large language model as the new explanation description of the target knowledge point.

[0058] In some disclosed embodiments, when the feedback result is characterized as being incomprehensible, the feedback result further includes at least one of an incomprehensible sentence and an expected explanation method. The interaction loop iteration module 34 includes a knowledge base acquisition sub-module for acquiring a preset knowledge base. The preset knowledge base contains a number of preset knowledge points and their respective explanation descriptions, and the number of preset knowledge points includes the target knowledge point. The interaction loop iteration module 34 includes an explanation description selection sub-module for selecting the explanation description of the target knowledge point in the preset knowledge base as the target description. The interaction loop iteration module 34 includes an explanation description extraction sub-module for extracting at least part of the target description as a reference description based on at least one of the incomprehensible sentence and the expected explanation method in the feedback result. The interaction loop iteration module 34 includes an explanation description correction sub-module for adjusting the explanation description of the target knowledge point based on the reference description to obtain the new explanation description of the target knowledge point.

[0059] In some disclosed embodiments, the Q&A data analysis module 32 includes an analysis sub-module for analyzing the Q&A data to obtain a number of candidate knowledge points. The Q&A data analysis module 32 includes a selection sub-module for selecting each candidate knowledge point related to answering the question to be answered as a number of knowledge points. The Q&A data analysis module 32 includes an extraction sub-module for extracting the initial explanation of each knowledge point in the Q&A data. The Q&A data analysis module 32 includes a retrieval sub-module for respectively retrieving the professional explanations of each knowledge point. The Q&A data analysis module 32 includes a correction sub-module for correcting the initial explanation of each knowledge point based on the professional explanation to obtain the explanation description of the corresponding knowledge point.

[0060] In some disclosed embodiments, the question to be answered includes a scientific question, and the knowledge points include at least one of a scientific concept and a scientific phenomenon.

[0061] Please refer to Figure 5 , ​It is a schematic diagram of the framework of an embodiment of the electronic device of the present application. The electronic device 40 at least includes a memory 41 and a processor 42 that are coupled to each other. At least program instructions are stored in the memory 41, and the processor 42 is configured to execute the program instructions to implement the steps in any of the above-mentioned method embodiments for generating answers. For details, reference can be made to the foregoing disclosed embodiments, which will not be elaborated here. As a possible example, the electronic device 40 may include, but is not limited to, an office notebook, an e-book reader, a smart phone, a tablet computer, a learning machine, etc. The specific type of the electronic device 40 is not limited herein.

[0062] Specifically, the processor 42 is configured to control itself and the memory 41 to implement the steps in any of the above-mentioned method embodiments for generating answers. The processor 42 may also be referred to as a CPU (Central Processing Unit). The processor 42 may be an integrated circuit chip with signal processing capabilities. The processor 42 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 gate or transistor logic devices, discrete hardware components. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. Additionally, the processor 42 may be implemented jointly by integrated circuit chips.

[0063] In the above solution, the electronic device 40 generates an initial answer to the question to be answered, obtains several knowledge points involved in the Q&A data and the explanatory descriptions of each knowledge point, and the Q&A data includes at least the question to be answered and the initial answer. Then, it obtains the feedback result of the understandability of the explanatory description of the knowledge point by the target object. Thus, in response to the feedback result indicating incomprehensibility, it respectively selects each knowledge point with the feedback result indicating incomprehensibility as the target knowledge point, and re-obtains the explanatory description of the target knowledge point, and returns to the step of obtaining the feedback result of the understandability of the explanatory description of the knowledge point by the target object. In response to the feedback result of each knowledge point indicating comprehensibility, based on the question to be answered and the explanatory descriptions of each knowledge point, it generates a final answer to the question to be answered. Therefore, on the one hand, when generating the final answer to the question to be answered, combining the explanatory descriptions of each knowledge point involved in the Q&A data can reflect the explanatory descriptions of the knowledge points in the final answer, which helps to improve the understandability of the generated answer. On the other hand, during the Q&A process, it also refers to the feedback result of the understandability of the explanatory description of the knowledge point by the target object, and when it indicates incomprehensibility, it re-obtains the explanatory description of the knowledge point and performs iterative loops until the feedback result of the understandability by the target object indicates comprehensibility, which can ensure that the target object can understand the explanatory descriptions of each knowledge point when generating the final answer, and helps to improve the understandability of the final answer as much as possible. Therefore, it can improve the understandability of the generated answer.

[0064] Please refer to ​ , ​ is a schematic framework diagram of an embodiment of the computer-readable storage medium 50 of the present application. The computer-readable storage medium 50 stores program instructions 51 that can be run by a processor, and the program instructions 51 are used to implement the steps in any of the above method embodiments for generating answers.

[0065] In the above solution, the computer-readable storage medium 50 generates an initial answer to the question to be answered, obtains a number of knowledge points involved in the Q&A data and the explanatory descriptions of each knowledge point, and the Q&A data includes at least the question to be answered and the initial answer. Then, it obtains the feedback result of the understandability of the explanatory description of the knowledge point by the target object. Thus, in response to the feedback result indicating incomprehensibility, it respectively selects each knowledge point with the feedback result indicating incomprehensibility as the target knowledge point, and re-obtains the explanatory description of the target knowledge point, and returns to the step of obtaining the feedback result of the understandability of the explanatory description of the knowledge point by the target object. And in response to the feedback result of each knowledge point indicating comprehensibility, based on the question to be answered and the explanatory descriptions of each knowledge point, it generates a final answer to the question to be answered. Therefore, on the one hand, when generating the final answer to the question to be answered, combining the explanatory descriptions of each knowledge point involved in the Q&A data can reflect the explanatory descriptions of the knowledge points in the final answer, which helps to improve the understandability of the generated answer. On the other hand, in the Q&A process, it also refers to the feedback result of the understandability of the explanatory description of the knowledge point by the target object, and when it indicates incomprehensibility, it re-obtains the explanatory description of the knowledge point and performs iterative loops until the feedback result of the understandability of the target object to it indicates comprehensibility, which can enable the target object to understand the explanatory descriptions of each knowledge point when generating the final answer, and helps to improve the understandability of the final answer as much as possible. Therefore, it can improve the understandability of the generated answer.

[0066] In some embodiments, the functions or modules included in the device provided by the embodiments of the present disclosure can be used to execute the methods described in the above method embodiments. The specific implementation can refer to the description of the above method embodiments. For the sake of brevity, it will not be repeated here.

[0067] The above descriptions of the respective embodiments tend to emphasize the differences between the respective embodiments. Their similarities or similarities can be referred to each other. For the sake of brevity, they will not be repeated in this article.

[0068] In several embodiments provided by the present application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of modules or units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection to each other can be through some interfaces. The indirect coupling or communication connection of the device or unit can be in an electrical, mechanical or other form.

[0069] The unit described as a separation component may or may not be physically separated, and the component displayed as a unit may or may not be a physical unit, that is, it may be located in one place or may be distributed across multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0070] In addition, in each embodiment of the present application, each functional unit may be integrated into a processing unit, may exist separately as individual physical units, or two or more units may be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0071] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods of each embodiment of the present application. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0072] If the technical solution of the present application involves personal information, before the product applying the technical solution of the present application processes personal information, it has clearly informed the personal information processing rules and obtained the personal's independent consent. If the technical solution of the present application involves sensitive personal information, before the product applying the technical solution of the present application processes sensitive personal information, it has obtained the personal's separate consent and at the same time meets the requirement of "express consent". For example, at a personal information collection device such as a camera, a clear and prominent label is set to inform that the personal information collection range has been entered and personal information will be collected. If an individual voluntarily enters the collection range, it is regarded as consenting to the collection of their personal information; or on the personal information processing device, when the personal information processing rules are informed by obvious labels / information, personal authorization is obtained through pop-up messages or by asking the individual to upload their personal information by themselves; among them, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the types of personal information processed.

Claims

1. A method for generating an answer, characterized in that: include: Generate initial answers to the questions to be answered; Acquire several knowledge points involved in the question and answer data and explanations of each of the knowledge points; wherein the question and answer data at least includes the question to be answered and the initial answer; Obtaining feedback results on the understandability of the target object's explanation and description of the knowledge point; In response to the feedback result being characterized as incomprehensible, each of the knowledge points characterized as incomprehensible by the feedback result is selected as a target knowledge point, and the explanation description of the target knowledge point is adjusted based on the expected explanation method contained in the feedback result to re-obtain the explanation description of the target knowledge point, and the step of obtaining the feedback result of the comprehensibility of the explanation description of the knowledge point by the target object is returned; In response to the feedback results of each of the knowledge points being characterized as understandable, generating a final answer to the question to be answered based on the question to be answered and the explanation description of each of the knowledge points; The target object is an intelligent agent simulating a real user, and the step of obtaining the feedback result of the understandability of the target object's explanation and description of the knowledge point includes: Based on the knowledge point and the explanation description of the knowledge point, construct an instruction to be executed; wherein the instruction to be executed is used to execute the agent to provide feedback on the understandability of the knowledge point and its explanation description; The to-be-executed instruction is input to the agent to obtain an output result of the agent as the feedback result.

2. The method according to claim 1, characterized in that: The target object is a real user, and the step of obtaining the feedback result of the understandability of the target object's explanation and description of the knowledge point includes: Based on the knowledge point and the explanation description of the knowledge point, construct a prompt message; wherein the prompt message is used to prompt the real user to provide feedback on the comprehensibility of the knowledge point and its explanation description; The prompt message is output, and input content of the real user in response to the prompt message is received as the feedback result.

3. The method according to claim 1, characterized in that The reacquiring the explanation description of the target knowledge point includes: Based on the explanation description of the target knowledge point and the feedback result, construct a prompt instruction; wherein the prompt instruction is used to instruct the large language model to adjust the explanation description of the target knowledge point with reference to the expected explanation mode; The prompt instruction is input into the large language model to obtain the output content of the large language model as a new explanation description of the target knowledge point.

4. The method according to claim 3, characterized in that: In the case where the feedback result is characterized as being incomprehensible, the feedback result also includes incomprehensible words and sentences, and constructing a prompt instruction based on the explanation description of the target knowledge point and the feedback result includes: Based on the explanation description of the target knowledge point, the expected explanation method in the feedback result and the incomprehensible words and sentences, a prompt instruction is constructed; wherein the prompt instruction is used to instruct the large language model to adjust the explanation description of the target knowledge point with reference to the expected explanation method and the incomprehensible words and sentences.

5. The method according to claim 1, characterized in that: The reacquiring the explanation description of the target knowledge point includes: Acquire a preset knowledge base; wherein the preset knowledge base includes a number of preset knowledge points and their respective explanation descriptions, and the number of preset knowledge points includes the target knowledge point; Selecting the explanation description of the target knowledge point in the preset knowledge base as the target description; Based on the desired interpretation mode, extract at least a portion of the target description as a reference description; The explanation description of the target knowledge point is adjusted based on the reference description to obtain a new explanation description of the target knowledge point.

6. The method according to claim 5, characterized in that In the case where the feedback result is characterized as being incomprehensible, the feedback result also includes incomprehensible words and sentences, and extracting at least part of the target description as a reference description based on the expected interpretation method includes: Based on the expected interpretation method and the incomprehensible words and sentences in the feedback result, at least a part of the target description is extracted as a reference description.

7. The method according to claim 1, characterized in that The acquisition of question and answer data involves several knowledge points and explanations of each of the knowledge points, including: Analyze the question and answer data to obtain several candidate knowledge points; Selecting each of the candidate knowledge points related to answering the question to be answered as the plurality of knowledge points; Extracting the initial explanation of each of the knowledge points in the question and answer data, and respectively retrieving the professional explanation of each of the knowledge points; For each of the knowledge points, the initial interpretation is corrected based on the professional interpretation to obtain an interpretation description corresponding to the knowledge point.

8. The method according to any one of claims 1 to 7, characterized in that: The questions to be answered include scientific questions, and the knowledge points include at least one of scientific concepts and scientific phenomena.

9. An answer generation device, characterized in that: include: An initial answer generation module, used to generate an initial answer to the question to be answered; A question and answer data analysis module, used to obtain a number of knowledge points involved in the question and answer data and an explanation description of each of the knowledge points; wherein the question and answer data at least includes the question to be answered and the initial answer; An explanation and understanding feedback module is used to obtain feedback results on the understandability of the explanation and description of the knowledge point by the target object; An interactive loop iteration module, for, in response to the feedback result being characterized as incomprehensible, selecting each of the knowledge points characterized as incomprehensible by the feedback result as a target knowledge point, adjusting the explanation description of the target knowledge point based on the expected explanation method contained in the feedback result to re-acquire the explanation description of the target knowledge point, and returning to the step of obtaining the feedback result of the comprehensibility of the explanation description of the knowledge point by the target object; A final answer generating module, configured to generate a final answer to the question to be answered based on the question to be answered and the explanation description of each of the knowledge points in response to the feedback results of each of the knowledge points being characterized as being understandable; Wherein, the target object is an intelligent agent simulating a real user, and the answer generation device further comprises: A first construction submodule is used to construct instructions to be executed based on the knowledge point and the explanation description of the knowledge point; wherein the instructions to be executed are used to execute the agent to provide feedback on the understandability of the knowledge point and its explanation description; The first acquisition submodule is used to input the to-be-executed instruction to the agent to obtain the output result of the agent as the feedback result.

10. An electronic device, characterized in that: The system at least comprises a memory and a processor coupled to each other, wherein the memory at least stores program instructions, and the processor is used to execute the program instructions to implement the answer generation method according to any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that: Program instructions that can be executed by a processor are stored, and the program instructions are used to implement the answer generation method according to any one of claims 1 to 8.

Images