Personalized science popularization content generation method and system based on user portrait and real-time feedback

Abstract

The invention is suitable for the technical field of science popularization content interaction, and particularly relates to a personalized science popularization content generation method and system based on a user portrait and real-time feedback, and the method comprises the steps: obtaining personnel information data, and constructing a user portrait database and a science popularization database; during interaction, identifying the personnel, determining the identity of the interaction personnel, performing content screening, generating science popularization data framework display content for a user to select, and obtaining a selection result; calling corresponding data contents, generating prompt words, generating corresponding science popularization information, and displaying the science popularization information; feedback information is obtained in real time, and follow-up science popularization information is dynamically adjusted. According to the method, the content framework of the cognitive level is matched for the user based on the accurate portrait, the personalized narration is dynamically generated by utilizing the large language model, the interest focus is locked through sight tracking, and the understanding degree is judged through expression recognition, so that the real-time dynamic adjustment of the content is realized, and the pertinence, interactivity and user experience of science popularization are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of interactive science popularization content technology, and in particular relates to a method and system for generating personalized science popularization content based on user profiles and real-time feedback. Background Technology

[0002] Personalized science content generation refers to dynamically customizing and pushing science popularization content that highly matches the needs of users based on their individual characteristics such as interests, knowledge level, learning goals, and behavioral habits, using artificial intelligence and big data technologies. This approach not only improves the readability and acceptance of science information but also stimulates public interest in science, achieving "personalized education" and enabling scientific knowledge to reach different audience groups more accurately and efficiently.

[0003] Traditional methods of presenting science information are mostly one-way, static broadcasts, which cannot adapt to the personalized needs of users of different ages, knowledge backgrounds, and interests. While some existing personalized recommendation systems can filter content based on users' historical behavior, they lack the ability to perceive the user's real-time state during interaction, resulting in delayed and inaccurate content adjustments. Therefore, there is an urgent need for a personalized content generation solution that can deeply understand users and respond in real-time during interaction. Summary of the Invention

[0004] The purpose of this invention is to provide a method for generating personalized science popularization content based on user profiles and real-time feedback. This method aims to solve the problem that some existing personalized recommendation systems, although able to filter content based on users' historical behavior, lack the ability to perceive users' real-time status during interaction, resulting in delayed and inaccurate content adjustments.

[0005] This invention is implemented as follows: a method for generating personalized science popularization content based on user profiles and real-time feedback, the method comprising: Acquire personnel information data to construct a user profile database and a science popularization database; the user profile database is constructed based on the personnel information data, and the science popularization database contains data related to all science popularization content included in this science popularization. During interaction, the system identifies personnel through an image acquisition device to determine their identity, filters the science database based on the user profile, generates a science data framework to display content for the user to choose from, and obtains the selection result. Based on the selection results, relevant data content is retrieved from the science popularization database, prompt words are generated for the large language model according to the user profile, corresponding science popularization information is generated, and the science popularization information is displayed. During the display process, user feedback is obtained in real time, and subsequent popular science information is dynamically adjusted based on the feedback. The dynamic adjustment includes expanding content, shortening content, and changing the accessibility of the text description.

[0006] Preferably, the steps of identifying personnel through an image acquisition device during interaction, determining the identity of the interacting personnel, filtering content from the science popularization database based on the user profile, generating a science popularization data framework to display content for the user to choose from, and obtaining the selection result include: The facial image of the person interacting is captured by an image acquisition device and matched with the pre-stored user identity information to determine the unique identity of the person interacting. The corresponding user profile is retrieved from the user profile database based on the identity identifier. The course content learned by the user at the current age is inferred based on the education stage information in order to determine the knowledge boundary. The user profile includes at least the user's age and education stage information. Based on knowledge boundaries, science popularization topics containing knowledge already mastered by the user and some extended exploration knowledge are selected from the science popularization database. A structured science popularization data framework is generated and displayed in a visual form for the user to choose from. Finally, the user's selection results are received.

[0007] Preferably, the steps of retrieving relevant data content from the science popularization database based on the selection result, generating prompt words for the large language model according to the user profile, generating relevant science popularization information, and displaying the science popularization information include: Analyze the user's selection results, use them as target science popularization topics, and retrieve relevant knowledge data, images, and video materials from the science popularization database; Knowledge data is fused with user profile features extracted from a user profile database, and prompt words are constructed based on the fused information. The user profile features include knowledge level, interest preferences, and historical interaction records. The prompt words are input into a large language model to generate popular science narrative text that meets the user's personalized needs. The generated text is then integrated with corresponding materials and rendered and displayed on a display device.

[0008] Preferably, the step of acquiring user feedback information in real time during display and dynamically adjusting subsequent science popularization information based on the feedback information specifically includes: When displaying science information, the screen is divided into at least four logical display areas, and science content of different dimensions is displayed in different areas in a staggered manner. By tracking the user's gaze in real time through an image acquisition device, the system can determine the specific display area that the user is currently focusing on and identify the sub-topics of the content displayed in that area as the part that the user is most interested in. During the explanation, the system simultaneously captures the user's facial expressions through an image acquisition device, and determines the user's real-time understanding of the current explanation content based on a pre-trained expression recognition model, thereby dynamically adjusting the content based on the real-time understanding.

[0009] Preferably, when tracking the user's gaze focus, binocular images are acquired using an image acquisition device. Based on this, the user's height and distance from the display area are calculated. Information is displayed in different areas of the screen in turn. At this time, images of the user's eye area are acquired using a near-infrared camera to obtain eye images. Perspective correction is performed on the eye images to correct them to face the display area, thus determining the basic eye images when the user is looking at different positions on the display. In subsequent use, real-time eye images of the user are acquired, and after perspective transformation, the similarity between the real-time eye images and all basic eye images is calculated to construct a similarity matrix. The user's gaze focus is determined based on the values ​​of each element in the similarity matrix.

[0010] Another objective of this invention is to provide a personalized science popularization content generation system based on user profiles and real-time feedback, the system comprising: The database construction module is used to acquire personnel information data and build a user profile database and a science popularization database; the user profile database is built based on the personnel information data, and the science popularization database contains data related to all science popularization content included in this science popularization. The content filtering module is used to identify people through an image acquisition device during interaction, determine the identity of the interacting person, filter the content of the popular science database based on the user profile, generate a popular science data framework to display content for the user to choose from, and obtain the selection result. The information display module is used to retrieve relevant data content from the science popularization database based on the selection results, generate prompt words for the large language model according to the user profile, generate corresponding science popularization information, and display the science popularization information; The information dynamic adjustment module is used to obtain user feedback information in real time during display, and to dynamically adjust subsequent popular science information based on the feedback information. The dynamic adjustment includes expanding content, shortening content, and changing the accessibility of text descriptions.

[0011] Preferably, the content filtering module includes: The identity recognition unit is used to capture the facial image of the interacting person through the image acquisition device, match it with the pre-stored user identity information, and determine the unique identity of the current interacting person; The user information calculation unit is used to retrieve the corresponding user profile from the user profile database based on the identity identifier, and infer the course content that the user is learning at the current age based on the education stage information in order to determine the knowledge boundary. The user profile includes at least the user's age and education stage information. The knowledge filtering unit is used to filter science topics from the science database based on knowledge boundaries. These topics include knowledge already mastered by the user and some extended exploration knowledge. The system generates a structured science data framework, which is then displayed in a visual format for the user to choose from. Finally, the system receives the user's selection results.

[0012] Preferably, the information display module includes: The data retrieval unit is used to analyze the user's selection results, take them as the target science popularization topic, and retrieve knowledge data, images, and video materials related to the topic from the science popularization database. The prompt word generation unit is used to fuse knowledge data with user profile features extracted from the user profile database, and construct prompt words based on the fused information. The user profile features include knowledge level, interest preferences and historical interaction records. The content generation unit is used to input prompt words into the large language model, generate popular science narrative text that meets the user's personalized needs, integrate the generated text with corresponding materials, and render and display it on the display device.

[0013] Preferably, the information dynamic adjustment module includes: The display area division unit is used to divide the screen into at least four logical display areas when displaying popular science information, and to display popular science content of different dimensions in different areas in a staggered manner; The gaze-tracking unit is used to track the user's gaze focus in real time through an image acquisition device, determine the specific display area that the user is currently focusing on, and identify the sub-topic of the content displayed in that area as the part that the user is most interested in. The interactive adjustment unit is used to capture the user's facial expressions simultaneously through an image acquisition device during the explanation, and to determine the user's real-time understanding of the current explanation content based on a pre-trained expression recognition model, and to dynamically adjust the content based on the real-time understanding.

[0014] Preferably, when tracking the user's gaze focus, binocular images are acquired using an image acquisition device. Based on this, the user's height and distance from the display area are calculated. Information is displayed in different areas of the screen in turn. At this time, images of the user's eye area are acquired using a near-infrared camera to obtain eye images. Perspective correction is performed on the eye images to correct them to face the display area, thus determining the basic eye images when the user is looking at different positions on the display. In subsequent use, real-time eye images of the user are acquired, and after perspective transformation, the similarity between the real-time eye images and all basic eye images is calculated to construct a similarity matrix. The user's gaze focus is determined based on the values ​​of each element in the similarity matrix.

[0015] This invention achieves a leap from one-size-fits-all science popularization content to personalized content by constructing a closed loop of user profiles and real-time feedback. Based on accurate user profiles, it matches content frameworks to users' cognitive levels and uses a large language model to dynamically generate personalized narratives. It locks in the focus of interest through eye tracking and judges the degree of understanding through facial expression recognition, thereby realizing real-time dynamic adjustment of content and significantly improving the relevance, interactivity and user experience of science popularization. Attached Figure Description

[0016] Figure 1 A flowchart illustrating the personalized science popularization content generation method based on user profiles and real-time feedback provided in this embodiment of the invention; Figure 2 A flowchart illustrating the steps for constructing a user profile database and a science popularization database provided in this embodiment of the invention; Figure 3 The flowchart illustrates the steps of retrieving relevant data from a science popularization database based on the selection result, generating prompt words for a large language model based on the user profile, generating relevant science popularization information, and displaying the science popularization information, as provided in this embodiment of the invention. Figure 4 A flowchart illustrating the steps of dynamically adjusting subsequent science popularization information based on real-time user feedback during display, as provided in this embodiment of the invention. Figure 5 An architecture diagram of a personalized science popularization content generation system based on user profiles and real-time feedback provided in an embodiment of the present invention; Figure 6 An architecture diagram of the content filtering module provided in an embodiment of the present invention; Figure 7 An architecture diagram of the information display module provided in an embodiment of the present invention; Figure 8 This is an architecture diagram of the information dynamic adjustment module provided in an embodiment of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0018] like Figure 1 The diagram shows a flowchart of a personalized science popularization content generation method based on user profiles and real-time feedback provided in an embodiment of the present invention. The method includes: S100, acquire personnel information data, construct a user profile database and a science popularization database. The user profile database is constructed based on the personnel information data, and the science popularization database contains data related to all science popularization content included in this science popularization.

[0019] In this step, personnel information data is acquired. When users enter their exhibition information, user information is entered, including the user's age, grade, occupation, and self-described areas of interest. A structured user profile database is constructed. Based on the theme of this science popularization project, a structured science popularization database is constructed, which includes science popularization text, pictures, videos, audio, knowledge tags, and difficulty levels.

[0020] S200: During interaction, the image acquisition device identifies the person and determines their identity. Based on the user profile, the science popularization database is filtered to generate a science popularization data framework for the user to select from, and the selection result is obtained.

[0021] In this step, during interaction, it is determined that the user has entered the interactive environment. Non-contact identification of the person is performed using image acquisition devices, including binocular cameras and near-infrared cameras. The user's identity is matched by facial recognition to determine the matching relationship between the current interactive user and the user profile in the user profile database. Information in the profile, such as age and knowledge level, is used as a filtering condition to initially filter the science popularization data in the current science popularization database, generating a science popularization theme framework that matches the user's current cognitive ability. This framework may be presented in the form of a graphic menu for the user to actively select and obtain the selection result.

[0022] S300 retrieves relevant data from the science popularization database based on the selection results, generates prompt words for the large language model according to the user profile, generates corresponding science popularization information, and displays the science popularization information.

[0023] In this step, based on the selection results, relevant data content is retrieved from the science popularization database. The core knowledge points and data materials most relevant to the topic are extracted from the database. The characteristics of the user profile, such as a third-grade elementary school student, interest in dinosaurs, and a beginner level of knowledge, are combined with the instructions required by the large language model to construct a highly customized prompt word. The prompt word is used to guide the large language model to generate the corresponding science popularization narrative. After being imported into the large language model, the large language model generates the corresponding science popularization narrative text through role-playing, and plans and classifies the science popularization information to determine the display order of the science popularization information. The generated text is integrated with the multimedia materials in the database and displayed on the screen.

[0024] S400, during display, obtains user feedback information in real time, and dynamically adjusts subsequent popular science information based on the feedback information. The dynamic adjustment includes expanding content, shortening content, and changing the accessibility of text descriptions.

[0025] In this step, while presenting science content to users, non-contact feedback information is continuously collected from users through image acquisition devices. This non-contact feedback information includes facial images, which are divided into visible light binocular images and near-infrared images, acquired by different image acquisition devices. Visible light binocular images can determine the distance between the user and the display area and identify spatial positional relationships. Near-infrared images can help determine the current focus of the user's gaze. Image recognition can further identify the user's current level of interest in the content. Based on user feedback, the prepared science information is dynamically adjusted, including expanding, shortening, and changing the accessibility of the text descriptions. For example, content can be expanded on parts that the user is interested in, and parts that the user finds confusing can be re-explained in more accessible language, or content that the user is showing signs of fatigue can be skipped, achieving real-time matching between content and user state.

[0026] like Figure 2 As shown, in a preferred embodiment of the present invention, the steps of identifying personnel through an image acquisition device to determine the identity of the interacting personnel, filtering the content of the science popularization database based on the user profile, generating a science popularization data framework to display content for the user to choose from, and obtaining the selection result include: S201: The face image of the person interacting is captured by the image acquisition device and matched with the pre-stored user identity information to determine the unique identity of the person interacting.

[0027] In this step, a high-resolution camera captures the face image of the person interacting in real time, extracts facial features, compares and matches the user's facial features with the pre-entered user identity information database, and determines whether there is a match based on the matching results in order to obtain the unique identity of the person interacting.

[0028] S202, retrieve the corresponding user profile from the user profile database based on the identity identifier, infer the course content learned by the user at the current age based on the education stage information, so as to determine the knowledge boundary. The user profile includes at least the user's age and education stage information.

[0029] In this step, based on the successfully identified identity, the user profile of the user is retrieved from the user profile database. Key fields in the user profile, such as age and education level, are analyzed. With the help of the education knowledge graph, the knowledge scope that the user has learned or been exposed to in the current age group is inferred. For example, a fifth-grade student usually has basic mechanics concepts. Based on this, the knowledge boundaries of the current user are determined.

[0030] S203, based on knowledge boundaries, selects popular science topics from the popular science database that include knowledge already mastered by the user and some extended exploration knowledge, generates a structured popular science data framework, and displays it in a visual form for the user to choose from, and finally receives the user's selection results.

[0031] In this step, the science database is queried using defined knowledge boundaries as filtering criteria. Knowledge content that matches the user's existing knowledge is selected. Based on this, some extended content that the user needs to master next is added to generate a structured science data framework, such as a selection menu with icons and brief text descriptions. This framework is then presented to the user through a display screen, allowing the user to actively select and clarify the science content that interests them.

[0032] like Figure 3 As shown, in a preferred embodiment of the present invention, the steps of retrieving relevant data content from a science popularization database based on the selection result, generating prompt words for a large language model based on user profiles, generating relevant science popularization information, and displaying the science popularization information include: S301 analyzes the user's selection results, uses them as the target science popularization topic, and retrieves relevant knowledge data, images, and video materials from the science popularization database.

[0033] In this step, the user's selection results are analyzed, semantic analysis is performed on the selection made by the user on the initial interface, and all relevant material data, including science popularization texts, science popularization videos, and science popularization audio, are retrieved from the science popularization database.

[0034] S302, the knowledge data is fused with the user profile features extracted from the user profile database, and prompt words are constructed based on the fused information. The user profile features include knowledge level, interest preferences and historical interaction records.

[0035] In this step, knowledge data is fused with user profile features extracted from the user profile database. These user profile features include the user's historical knowledge level, recorded interests and preferences, and past interaction records. The fused information is presented as cue words, which are used to indicate the role the large language model needs to play, the topic to be explained, the characteristics of the target audience, the depth of knowledge, and the expected language style, such as being vivid, interesting, and using analogies, in order to generate highly customized content.

[0036] S303 inputs prompt words into a large language model to generate popular science narrative text that meets the user's personalized needs, and integrates the generated text with corresponding materials for rendering and display on a display device.

[0037] In this step, prompt words are input into a large language model to generate popular science narrative text that meets the user's personalized needs. For large text information in the popular science data, content conversion prompt words are generated based on the large text information. Based on the corresponding video generation model, a video or explanatory image matching the large text information is generated to facilitate user understanding. The generated text is then integrated with the corresponding text and rendered and displayed on the display device.

[0038] like Figure 4 As shown, in a preferred embodiment of the present invention, the step of acquiring user feedback information in real time during display and dynamically adjusting subsequent science popularization information based on the feedback information specifically includes: S401, when displaying popular science information, divides the screen into at least four logical display areas and displays popular science content of different dimensions in different areas in a staggered manner.

[0039] In this step, when displaying science information, the screen is divided into at least four logical display areas. The number of logical display areas is determined by the size of the display screen. The larger the display screen, the more logical display areas can be set, such as four areas a, b, c, and d. Different sub-themes or different dimensions of content under a science theme are placed in these areas, such as area a showing the size of dinosaurs, area b showing dinosaur food, area c showing the dinosaur extinction hypothesis, and area d showing fossil discoveries.

[0040] S402 uses an image acquisition device to track the user's gaze focus in real time, determines the specific display area that the user is currently focusing on, and identifies the sub-topic of the content displayed in that area as the part that the user is most interested in.

[0041] In this step, the user's gaze focus is tracked in real time using an image acquisition device. Binocular images are acquired, and the user's height and distance to the display area are calculated. Information is displayed alternately in different areas of the screen. Simultaneously, an infrared camera captures images of the user's eye area, obtaining eye images. Perspective correction is performed on these eye images to align them towards the display area, determining the baseline eye images for different viewing positions on the monitor. In subsequent use, real-time eye images are acquired, perspective transformed, and the similarity between the real-time eye images and all baseline eye images is calculated to construct a similarity matrix. The user's gaze focus is determined based on the values ​​of each element in the similarity matrix. Specifically, the precise distance between the user's pupil and the camera is calculated using the principle of binocular parallax. Combined with the known relative positions of the camera and the display screen, the user's height and the 3D coordinates (X, Y, Z) of the user's eyes relative to the center of the display screen are calculated. The screen is controlled at multiple predefined points, such as the center marker of each display area, and their screen coordinates are recorded. , When a marker appears, a near-infrared camera simultaneously captures the user's eye image. Based on the user's spatial position determined by binocular recognition, perspective correction is performed on the captured raw eye image to eliminate image distortion caused by head rotation when the user views different positions on the screen. This is converted into an equivalent eye image when the user is facing the camera directly. The corrected image is used as the base eye image corresponding to that screen position and stored in the benchmark library along with its coordinates. For each frame of real-time eye image, perspective transformation is performed using the same parameters. The transformed real-time eye image is then compared with the base eye images corresponding to all predefined points in the benchmark library to calculate similarity. The similarity algorithm is characterized by a structural similarity index, yielding a similarity score for each point. Construct a similarity matrix containing the coordinates of all predefined points. , ) and the corresponding similarity The coordinates of the viewpoint are calculated using a similarity-weighted method: similarity values ​​are used as weights to calculate the weighted average coordinates of all predefined points, thus determining the viewpoint. The formula is as follows: ; ; in, It is the number of predefined points, ( , ) are the calculated coordinates of the line-of-sight focus; The following example uses four pre-determined points: Suppose there are 4 predefined points, corresponding to the centers of display areas a, b, c, and d, and their coordinates and real-time similarity are: point a: ( , ), = 0.15, point b: ( , ), = 0.85, point c: ( , ), = 0.25, point d: ( , ), = 0.10; Calculate weighted coordinates = (0.15 + 0.85 + 0.25 + 0.10 ) / (0.15 + 0.85 + 0.25 + 0.10) = (0.15) + 0.85 + 0.25 + 0.10 / 1.35; = (0.15 + 0.85 + 0.25 + 0.10 ) / 1.35; Since point b has the highest similarity, the focus coordinates will be closer to point b. However, other points will also shift according to their similarity contribution. Based on the display area corresponding to the focus coordinates, the display area currently being viewed by the user is determined, and the sub-topics of the content displayed in that area are identified as the part that the user is currently most interested in.

[0042] S403, during the explanation, simultaneously captures the user's facial expressions through the image acquisition device, and determines the user's real-time understanding of the current explanation content based on the pre-trained expression recognition model, and realizes dynamic adjustment of the content based on the real-time understanding.

[0043] In this step, as the user identifies a sub-topic of interest and begins a detailed explanation, images captured by the image acquisition device are imported into a pre-trained facial expression recognition model. This model analyzes the user's facial expression features in real time, such as whether there are signs of frowning, downturned corners of the mouth, or confused or bewildered expressions. When such signals indicating a lack of understanding are detected, a content adjustment mechanism is immediately triggered. For example, more detailed explanations of the steps can be automatically inserted, more accessible analogies can be provided, or the explanation can be delivered at a slower pace. When making analogies, the model incorporates the user's existing knowledge. For instance, if the user has already learned the function of the hands on a clock, the hour hand, minute hand, and second hand can be used to represent different objects to facilitate understanding.

[0044] like Figure 5 As shown, in a preferred embodiment of the present invention, a personalized science popularization content generation system based on user profiles and real-time feedback includes: The database construction module 100 is used to acquire personnel information data and construct a user profile database and a science popularization database; the user profile database is constructed based on the personnel information data, and the science popularization database contains data related to all science popularization content included in this science popularization.

[0045] In this system, the database construction module 100 acquires personnel information data. When users enter their exhibition information, user information is entered, including the user's age, grade, occupation, and self-described areas of interest, etc., to build a structured user profile database. Based on the theme of this science popularization project, a structured science popularization database is built, which includes science popularization text, pictures, videos, audio, knowledge tags, and difficulty levels.

[0046] The content filtering module 200 is used to identify people through an image acquisition device during interaction, determine the identity of the interacting person, filter the content of the popular science database based on the user profile, generate a popular science data framework to display content for the user to choose from, and obtain the selection result.

[0047] In this system, when the content filtering module 200 interacts, it determines that the user has entered the interactive environment. Through image acquisition devices, including binocular cameras and near-infrared cameras, it performs non-contact identification of the person and matches their identity through facial recognition to determine the matching relationship between the current interactive user and the user profile in the user profile database. Using information in the profile, such as age and knowledge level, as filtering conditions, it performs initial filtering of the popular science data in the current popular science database and generates a popular science theme framework that matches the user's current cognitive ability. This framework may be presented in the form of a graphic menu for the user to actively select and obtain the selection result.

[0048] The information display module 300 is used to retrieve relevant data content from the science popularization database based on the selection results, generate prompt words for the large language model according to the user profile, generate corresponding science popularization information, and display the science popularization information.

[0049] In this system, the information display module 300 retrieves relevant data content from the science popularization database based on the selection results. It extracts the core knowledge points and data materials most relevant to the topic from the database and combines the characteristics of the user profile, such as a third-grade elementary school student, interest in dinosaurs, and a beginner level of knowledge, with the instructions required by the large language model to construct a highly customized prompt word. The prompt word guides the large language model to generate the corresponding science popularization narrative. After being imported into the large language model, the model generates the corresponding science popularization narrative text through role-playing and plans and classifies the science popularization information to determine the display order. The generated text is then integrated with multimedia materials in the database and displayed on the screen.

[0050] The information dynamic adjustment module 400 is used to obtain user feedback information in real time during display, and to dynamically adjust subsequent popular science information based on the feedback information. The dynamic adjustment includes expanding content, shortening content, and changing the accessibility of text descriptions.

[0051] In this system, the information dynamic adjustment module 400 continuously collects non-contact feedback information from users during the presentation of science popularization content. This non-contact feedback information includes facial images of the users, which are divided into visible light binocular images and near-infrared images, acquired by different image acquisition devices. The visible light binocular images can determine the distance between the user and the display area and identify spatial positional relationships. The near-infrared images can help determine the current focus of the user's gaze. Image recognition can further identify the user's current level of interest in the content. Based on the user's feedback, the prepared science popularization information is dynamically adjusted, including expanding, shortening, and changing the accessibility of the text descriptions. For example, the content of parts that the user is interested in is expanded, parts that the user finds confusing are re-explained in more accessible language, or content that the user is showing signs of fatigue is skipped, achieving real-time matching between content and user state.

[0052] like Figure 6 As shown, in a preferred embodiment of the present invention, the content filtering module 200 includes: The identity recognition unit 201 is used to capture the facial image of the interacting person through the image acquisition device, match it with the pre-stored user identity information, and determine the unique identity of the current interacting person.

[0053] In this module, the identity recognition unit 201 captures the face image of the interacting person in real time through a high-resolution camera, extracts facial features, compares and matches the user's facial features with the pre-entered user identity information database, and determines whether there is a matching item based on the matching result in order to obtain the unique identity identifier of the current interacting person.

[0054] User information calculation unit 202 is used to retrieve the corresponding user profile from the user profile database based on the identity identifier, and infer the course content that the user is learning at the current age based on the education stage information in order to determine the knowledge boundary. The user profile includes at least the user's age and education stage information.

[0055] In this module, the user information calculation unit 202 retrieves the user's profile from the user profile database based on the successfully identified identity, analyzes key fields in the user profile such as age and education stage, and uses an educational knowledge graph to infer the knowledge scope that the user has learned or been exposed to at the current age, such as the basic mechanics concepts that a fifth-grade elementary school student usually has, thereby determining the current user's knowledge boundaries.

[0056] The knowledge filtering unit 203 is used to filter out popular science topics from the popular science database based on knowledge boundaries. These topics include knowledge that the user has already mastered and some extended knowledge for exploration. The unit generates a structured popular science data framework and displays it in a visual form for the user to choose from. Finally, the unit receives the user's selection results.

[0057] In this module, the knowledge filtering unit 203 uses defined knowledge boundaries as filtering conditions to query the science popularization database, filtering out knowledge content that matches the user's existing knowledge. On this basis, it adds some extended content that the user needs to master next, generating a structured science popularization data framework, such as a selection menu with icons and brief text descriptions, and presents it to the user through the display screen. Through active selection, the user can clarify the science popularization content that interests them.

[0058] like Figure 7 As shown, in a preferred embodiment of the present invention, the information display module 300 includes: The data retrieval unit 301 is used to parse the user's selection results, take them as the target popular science topic, and retrieve knowledge data, pictures and video materials related to the topic from the popular science database.

[0059] In this module, the data retrieval unit 301 parses the user's selection results, performs semantic analysis on the selection made by the user on the initial interface, and retrieves all relevant material data from the science popularization database, including science popularization texts, science popularization videos, and science popularization audio.

[0060] The prompt word generation unit 302 is used to fuse knowledge data with user profile features extracted from the user profile database, and construct prompt words based on the fused information. The user profile features include knowledge level, interest preferences and historical interaction records.

[0061] In this module, the prompt word generation unit 302 integrates knowledge data with user profile features extracted from the user profile database. The user profile features include the user's historical knowledge level, recorded interests and preferences, and past interaction records. The integrated information is represented by prompt words, which are used to indicate the role that the large language model needs to play, the topic to be explained, the characteristics of the target audience, the depth of knowledge, and the expected language style, such as being vivid and interesting and using analogies frequently, in order to generate highly customized content.

[0062] The content generation unit 303 is used to input prompt words into the large language model, generate popular science narrative text that meets the user's personalized needs, and integrate the generated text with corresponding materials for rendering and display on the display device.

[0063] In this module, the content generation unit 303 inputs prompt words into the large language model to generate popular science narrative text that meets the user's personalized needs. For large text information in popular science data, it generates content conversion prompt words based on the large text information. Based on the corresponding video generation model, it generates a video or explanatory image that matches the large text information to facilitate user understanding. The generated text is then integrated with the corresponding text and rendered and displayed on the display device.

[0064] like Figure 8 As shown, in a preferred embodiment of the present invention, the information dynamic adjustment module 400 includes: The display area division unit 401 is used to divide the screen into at least four logical display areas when displaying popular science information, and to display popular science content of different dimensions in different areas in a staggered manner.

[0065] In this module, the display area division unit 401 divides the screen into at least four logical display areas when displaying popular science information. The number of logical display areas is determined by the size of the display screen. The larger the display screen, the more logical display areas can be set, such as four areas a, b, c, and d. Different sub-topics or different dimensions of content under a popular science theme are placed in these areas, such as area a showing the size of dinosaurs, area b showing dinosaur food, area c showing the dinosaur extinction hypothesis, and area d showing fossil discoveries.

[0066] The gaze-tracking unit 402 is used to track the user's gaze focus in real time through the image acquisition device, determine the specific display area that the user is currently focusing on, and identify the sub-topic of the content displayed in that area as the part that the user is most interested in.

[0067] In this module, the gaze-tracking unit 402 tracks the user's gaze focus in real time through an image acquisition device. It acquires binocular images to calculate the user's height and distance from the display area. Information is displayed alternately in different areas of the screen. Simultaneously, a near-infrared camera captures images of the user's eye area, obtaining eye images. Perspective correction is applied to these eye images to align them towards the display area, determining the base eye images for different viewing positions on the screen. In subsequent use, real-time eye images are acquired, and after perspective transformation, the similarity between the real-time eye images and all base eye images is calculated to construct a similarity matrix. The user's gaze focus is determined based on the values ​​of each element in the similarity matrix. Specifically, the precise distance between the user's pupil and the camera is calculated using the binocular parallax principle. Combined with the known relative positions of the camera and the display screen, the user's height and the 3D coordinates (X, Y, Z) of the user's eyes relative to the center of the display screen are calculated. The screen is controlled to record the screen coordinates at multiple predefined points, such as the center marker of each display area. , When a marker appears, a near-infrared camera simultaneously captures the user's eye image. Based on the user's spatial position determined by binocular recognition, perspective correction is performed on the captured raw eye image to eliminate image distortion caused by head rotation when the user views different positions on the screen. This is converted into an equivalent eye image when the user is facing the camera directly. The corrected image is used as the base eye image corresponding to that screen position and stored in the benchmark library along with its coordinates. For each frame of real-time eye image, perspective transformation is performed using the same parameters. The transformed real-time eye image is then compared with the base eye images corresponding to all predefined points in the benchmark library to calculate similarity. The similarity algorithm is characterized by a structural similarity index, yielding a similarity score for each point. Construct a similarity matrix containing the coordinates of all predefined points. , ) and the corresponding similarity The coordinates of the viewpoint are calculated using a similarity-weighted method: similarity values ​​are used as weights to calculate the weighted average coordinates of all predefined points, thus determining the viewpoint. The formula is as follows: ; ; in, It is the number of predefined points, ( , ) are the calculated coordinates of the line-of-sight focus; The following example uses four pre-determined points: Suppose there are 4 predefined points, corresponding to the centers of display areas a, b, c, and d, and their coordinates and real-time similarity are: point a: ( , ), = 0.15, point b: ( , ), = 0.85, point c: ( , ), = 0.25, point d: ( , ), = 0.10; Calculate weighted coordinates = (0.15 + 0.85 + 0.25 + 0.10 ) / (0.15 + 0.85 + 0.25 + 0.10) = (0.15) + 0.85 + 0.25 + 0.10 / 1.35; = (0.15 + 0.85 + 0.25 + 0.10 ) / 1.35; Since point b has the highest similarity, the focus coordinates will be closer to point b. However, other points will also shift according to their similarity contribution. Based on the display area corresponding to the focus coordinates, the display area currently being viewed by the user is determined, and the sub-topics of the content displayed in that area are identified as the part that the user is currently most interested in.

[0068] The interactive adjustment unit 403 is used to capture the user's facial expressions simultaneously through an image acquisition device during the explanation, and to determine the user's real-time understanding of the current explanation content based on a pre-trained expression recognition model, and to dynamically adjust the content based on the real-time understanding.

[0069] In this module, when the interactive adjustment unit 403 locks onto a sub-topic of interest to the user and begins a detailed explanation, the image acquired by the image acquisition device is imported into the pre-trained facial expression recognition model. The model analyzes the user's facial expression features in real time, such as whether there are frowning, downturned corners of the mouth, or confused or bewildered expressions. When such signals indicating a lack of understanding are detected, the content adjustment mechanism is immediately triggered. For example, more detailed step-by-step explanations can be automatically inserted, more colloquial analogies can be provided, or the explanation can be delivered at a slower pace. When making analogies, the analogies are combined with the user's existing knowledge. For example, if the user has already learned the function of the various hands on a clock, the hour hand, minute hand, and second hand can be used to represent different objects to facilitate the user's understanding.

[0070] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for generating personalized science popularization content based on user profiles and real-time feedback, characterized in that: The method includes: Acquire personnel information data to construct a user profile database and a science popularization database; the user profile database is constructed based on the personnel information data, and the science popularization database contains data related to all science popularization content included in this science popularization. During interaction, the system identifies personnel through an image acquisition device to determine their identity, filters the science database based on the user profile, generates a science data framework to display content for the user to choose from, and obtains the selection result. Based on the selection results, relevant data content is retrieved from the science popularization database, prompt words are generated for the large language model according to the user profile, corresponding science popularization information is generated, and the science popularization information is displayed. During the display process, user feedback is obtained in real time, and subsequent popular science information is dynamically adjusted based on the feedback. The dynamic adjustment includes expanding content, shortening content, and changing the accessibility of the text description.

2. The method for generating personalized science popularization content based on user profiles and real-time feedback according to claim 1, characterized in that, During the interaction, the steps include: identifying personnel through an image acquisition device to determine their identity; filtering content from the science database based on the user profile; generating a science data framework to display content for the user to choose from; and obtaining the selection result. The facial image of the person interacting is captured by an image acquisition device and matched with the pre-stored user identity information to determine the unique identity of the person interacting. The corresponding user profile is retrieved from the user profile database based on the identity identifier. The course content learned by the user at the current age is inferred based on the education stage information in order to determine the knowledge boundary. The user profile includes at least the user's age and education stage information. Based on knowledge boundaries, science popularization topics containing knowledge already mastered by the user and some extended exploration knowledge are selected from the science popularization database. A structured science popularization data framework is generated and displayed in a visual form for the user to choose from. Finally, the user's selection results are received.

3. The method for generating personalized science popularization content based on user profiles and real-time feedback according to claim 1, characterized in that, The steps of retrieving relevant data from the science popularization database based on the selection result, generating prompt words for the large language model according to the user profile, generating corresponding science popularization information, and displaying the science popularization information include: Analyze the user's selection results, use them as target science popularization topics, and retrieve relevant knowledge data, images, and video materials from the science popularization database; Knowledge data is fused with user profile features extracted from a user profile database, and prompt words are constructed based on the fused information. The user profile features include knowledge level, interest preferences, and historical interaction records. The prompt words are input into a large language model to generate popular science narrative text that meets the user's personalized needs. The generated text is then integrated with corresponding materials and rendered and displayed on a display device.

4. The method for generating personalized science popularization content based on user profiles and real-time feedback according to claim 1, characterized in that, The step of acquiring user feedback in real time during display and dynamically adjusting subsequent science popularization information based on the feedback information specifically includes: When displaying science information, the screen is divided into at least four logical display areas, and science content of different dimensions is displayed in different areas in a staggered manner. By tracking the user's gaze in real time through an image acquisition device, the system can determine the specific display area that the user is currently focusing on and identify the sub-topics of the content displayed in that area as the part that the user is most interested in. During the explanation, the system simultaneously captures the user's facial expressions through an image acquisition device, and determines the user's real-time understanding of the current explanation content based on a pre-trained expression recognition model, thereby dynamically adjusting the content based on the real-time understanding.

5. The method for generating personalized science popularization content based on user profiles and real-time feedback according to claim 4, characterized in that, When tracking the user's gaze focus, binocular images are acquired using an image acquisition device. Based on this, the user's height and distance from the display area are calculated. Information is displayed in different areas of the screen in turn. At this time, images of the user's eye area are acquired through a near-infrared camera to obtain eye images. Perspective correction is performed on the eye images to adjust their orientation towards the display area, thus determining the base eye images when the user is looking at different positions on the display. In subsequent use, real-time eye images of the user are acquired, and after perspective transformation, the similarity between the real-time eye images and all base eye images is calculated to construct a similarity matrix. The user's gaze focus is determined based on the values ​​of each element in the similarity matrix.

6. A personalized science popularization content generation system based on user profiles and real-time feedback, characterized in that: The system includes: The database construction module is used to acquire personnel information data and build a user profile database and a science popularization database; the user profile database is built based on the personnel information data, and the science popularization database contains data related to all science popularization content included in this science popularization. The content filtering module is used to identify people through an image acquisition device during interaction, determine the identity of the interacting person, filter the content of the popular science database based on the user profile, generate a popular science data framework to display content for the user to choose from, and obtain the selection result. The information display module is used to retrieve relevant data content from the science popularization database based on the selection results, generate prompt words for the large language model according to the user profile, generate corresponding science popularization information, and display the science popularization information; The information dynamic adjustment module is used to obtain user feedback information in real time during display, and to dynamically adjust subsequent popular science information based on the feedback information. The dynamic adjustment includes expanding content, shortening content, and changing the accessibility of text descriptions.

7. The personalized science popularization content generation system based on user profiles and real-time feedback according to claim 6, characterized in that, The content filtering module includes: The identity recognition unit is used to capture the facial image of the interacting person through the image acquisition device, match it with the pre-stored user identity information, and determine the unique identity of the current interacting person; The user information calculation unit is used to retrieve the corresponding user profile from the user profile database based on the identity identifier, and infer the course content that the user is learning at the current age based on the education stage information in order to determine the knowledge boundary. The user profile includes at least the user's age and education stage information. The knowledge filtering unit is used to filter science topics from the science database based on knowledge boundaries. These topics include knowledge already mastered by the user and some extended exploration knowledge. The system generates a structured science data framework, which is then displayed in a visual format for the user to choose from. Finally, the system receives the user's selection results.

8. The personalized science popularization content generation system based on user profiles and real-time feedback according to claim 6, characterized in that, The information display module includes: The data retrieval unit is used to analyze the user's selection results, take them as the target science popularization topic, and retrieve knowledge data, images, and video materials related to the topic from the science popularization database. The prompt word generation unit is used to fuse knowledge data with user profile features extracted from the user profile database, and construct prompt words based on the fused information. The user profile features include knowledge level, interest preferences and historical interaction records. The content generation unit is used to input prompt words into the large language model, generate popular science narrative text that meets the user's personalized needs, integrate the generated text with corresponding materials, and render and display it on the display device.

9. The personalized science popularization content generation system based on user profiles and real-time feedback according to claim 6, characterized in that, The information dynamic adjustment module includes: The display area division unit is used to divide the screen into at least four logical display areas when displaying popular science information, and to display popular science content of different dimensions in different areas in a staggered manner; The gaze-tracking unit is used to track the user's gaze focus in real time through an image acquisition device, determine the specific display area that the user is currently focusing on, and identify the sub-topic of the content displayed in that area as the part that the user is most interested in. The interactive adjustment unit is used to capture the user's facial expressions simultaneously through an image acquisition device during the explanation, and to determine the user's real-time understanding of the current explanation content based on a pre-trained expression recognition model, and to dynamically adjust the content based on the real-time understanding.

10. The personalized science popularization content generation system based on user profiles and real-time feedback according to claim 9, characterized in that, When tracking the user's gaze focus, binocular images are acquired using an image acquisition device. Based on this, the user's height and distance from the display area are calculated. Information is displayed in different areas of the screen in turn. At this time, images of the user's eye area are acquired through a near-infrared camera to obtain eye images. Perspective correction is performed on the eye images to adjust their orientation towards the display area, thus determining the base eye images when the user is looking at different positions on the display. In subsequent use, real-time eye images of the user are acquired, and after perspective transformation, the similarity between the real-time eye images and all base eye images is calculated to construct a similarity matrix. The user's gaze focus is determined based on the values ​​of each element in the similarity matrix.

Images