An intelligent question-answering method and system based on a large model

Abstract

The present application provides an intelligent question-answering method and system based on a large model. Among them, the eye movement trajectory of the user and the interaction response delay are monitored to generate a dynamic cognitive load level, and the shallow and deep thinking modes are divided; according to the cognitive load level, two-way linkage control is performed on the screen display parameters and the question-answering content generation strategy to activate the eye protection or focused display interface respectively; through the information adaptation engine, the language structure and knowledge granularity of the question-answering content are reorganized to generate a simplified concept topology expression in the eye protection mode or a nested logical reasoning path in the focused mode; according to the matching degree between the user interaction behavior and the content browsing trajectory, the reorganization strategy is dynamically corrected, and the concept topology node density and the logical path branch depth are reversely adjusted; finally, the reorganized content and the adjustment parameters are synchronously rendered and dynamically displayed to form a cognitive load balance closed loop. The technical solution provided by the present application effectively reduces the cognitive load and improves the information absorption efficiency.

Description

Technical Field

[0001] The present application relates to the field of intelligent question answering technology, and in particular to an intelligent question answering method and system based on a large model. Background Art

[0002] In highly interactive scenarios such as online education and intelligent customer service, the efficiency of users' understanding of question and answer content is affected by multiple factors such as real-time cognitive status and ambient light conditions. It is necessary to dynamically adjust the content presentation format to adapt to different cognitive rhythms. At the same time, it is necessary to avoid visual fatigue or information overload caused by the mismatch between screen optical parameters and content complexity.

[0003] The current mainstream solution uses a static adaptation mechanism based on user historical behavior data. After analyzing the user's query intent through a pre-trained language model, it combines the user's profile (such as learning stage, interaction frequency) to generate a fixed-level content structure (such as a folded outline or an expanded detail page), and adjusts the screen brightness based on the ambient light sensor. For example, some systems automatically switch between bright / dark modes by monitoring the ambient light intensity of the device, and adjust the text size and paragraph spacing of the Q&A content in a linked manner.

[0004] Existing solutions have the following defects: they rely on the one-way correlation between historical behavior and ambient light, and cannot capture the user's dynamic cognitive state in real time (such as attention drift and cognitive load changes during deep thinking), resulting in the content complexity and optical parameter adjustment lagging behind actual needs. For example, the dark mode only reduces the brightness according to the ambient light intensity, but does not take into account the cognitive fatigue caused by users reading high-density content for a long time, resulting in a split between visual comfort and information comprehension efficiency. Summary of the invention

[0005] The present application provides an intelligent question-answering method and system based on a large model to solve the problem of low efficiency of information absorption in the prior art.

[0006] In a first aspect, the present application provides an intelligent question answering method based on a large model, comprising:

[0007] When receiving a question-and-answer request sent by a user, the collected user eye movement trajectory data and interactive response delay data are correlated and analyzed to generate a cognitive load level, and the shallow understanding mode and the deep thinking mode are divided according to the cognitive load level;

[0008] According to the cognitive load level, the generation strategy corresponding to the screen display parameters and the question and answer content parameters is bidirectionally linked and controlled, and when in the shallow understanding mode, the eye protection display interface is activated according to the bidirectional linkage control result; when in the deep thinking mode, the focus display interface is activated according to the bidirectional linkage control result;

[0009] Based on the cognitive load level and the optical characteristics of the current display interface, the information adaptation engine reorganizes the language structure and knowledge granularity of the Q&A content parameters, generates a conceptual topology expression according to the reorganization result under the eye protection display interface, and generates a logical reasoning path according to the reorganization result under the focused display interface;

[0010] According to the matching degree between the user's real-time interaction behavior and the content browsing trajectory, the reorganization strategy of the information adaptation engine is corrected, and the corrected reorganization strategy reversely adjusts the node density of the conceptual topology expression and the branch depth of the logical reasoning path through the change of the cognitive load level;

[0011] Synchronously render the reorganized Q&A content parameters, the adjusted node density or the adjusted branch depth to generate a Q&A rendering result, and display the Q&A rendering result to the user.

[0012] Optionally, the generation strategies corresponding to the screen display parameters and the Q&A content parameters are controlled in a two-way linkage according to the cognitive load level. When in the shallow understanding mode, the eye protection display interface is activated according to the two-way linkage control result; when in the deep thinking mode, the focused display interface is activated according to the two-way linkage control result, including:

[0013] Construct generation strategies for screen display parameters and Q&A content parameters. The screen display parameters include the blue light suppression intensity and the contrast gain value, and the Q&A content parameters include the information density attenuation gradient and the logic path expansion intensity;

[0014] According to the numerical range of the cognitive load level, extract the blue light suppression intensity and the information density attenuation gradient corresponding to the shallow understanding mode, and the contrast gain value and the logic path expansion intensity corresponding to the deep thinking mode from the generation strategies;

[0015] When in the shallow understanding mode, input the blue light suppression intensity into the screen backlight wavelength controller to generate an eye protection optical signal, and at the same time input the information density attenuation gradient into the compression engine corresponding to the Q&A content parameters to generate a simplified semantic unit sequence, and synchronously load the eye protection optical signal and the simplified semantic unit sequence to activate the eye protection display interface;

[0016] When in the deep thinking mode, input the contrast gain value into the screen pixel driving module to generate a focused optical signal, and at the same time input the logic path expansion intensity into the reasoning engine corresponding to the Q&A content parameters to generate a nested logic chain sequence, and synchronously load the focused optical signal and the nested logic chain sequence to activate the focused display interface.

[0017] Optionally, based on the cognitive load level and the optical characteristics of the current display interface, the information adaptation engine reorganizes the language structure and knowledge granularity of the Q&A content parameters, generates a conceptual topology expression according to the reorganization result under the eye protection display interface, and generates a logical reasoning path according to the reorganization result under the focused display interface, including:

[0018] The information adaptation engine loads a set of content structure mapping rules bound to the optical characteristics of the current display interface, where the eye protection display interface corresponds to the first mapping rule group in the set of content structure mapping rules, and the focused display interface corresponds to the second mapping rule group in the set of content structure mapping rules;

[0019] According to the currently activated display interface type, extract the corresponding mapping rule group from the set of content structure mapping rules, and use the weight calculation unit of the information adaptation engine to quantify the cognitive load level into a structure adjustment weight factor, and establish a proportional relationship between the structure adjustment weight factor and the compression coefficient or expansion coefficient in the mapping rule group;

[0020] Based on the proportional relationship, the language structure parsing unit of the information adaptation engine performs hierarchical marking on the core semantic units of the Q&A content parameters. Under the eye protection display interface, fold the semantic units with a hierarchical marking result higher than the preset level according to the compression coefficient of the first mapping rule group, and generate a conceptual topology expression according to the folded semantic units;

[0021] Under the focused display interface, the knowledge granularity control unit of the information adaptation engine performs branch extension processing on the semantic units with a marked level lower than the preset level according to the expansion coefficient of the second mapping rule group, and generates a logical reasoning path according to the branch extension processed semantic units.

[0022] Optionally, synchronously rendering the reorganized Q&A content parameters, the adjusted node density, or the adjusted branch depth to generate a Q&A rendering result, and presenting the Q&A rendering result to the user, including:

[0023] Synchronously render the reorganized Q&A content parameters and the adjusted node density to generate content distribution data with density markings; or synchronously render the reorganized Q&A content parameters and the adjusted branch depth to generate content distribution data with depth markings;

[0024] Select the corresponding density mark or depth mark in the content distribution data according to the current display interface type, calculate the basic spacing of visual elements according to the density mark or calculate the associated line strength of visual elements according to the depth mark. The basic spacing has an inverse fluctuation relationship with the adjusted node density, and the associated line strength has a positive fluctuation relationship with the adjusted branch depth;

[0025] Adjust the arrangement distribution of Q&A content parameters based on the basic spacing, or generate logical connection identifiers based on the associated line strength, and form a switchable dynamic visual template according to the adjusted arrangement distribution or the logical connection identifiers;

[0026] Convert the dynamic visual template into a screen pixel drive signal and a content rendering instruction. After inputting the screen pixel drive signal and the content rendering instruction into the display controller, output the Q&A rendering result and display the Q&A rendering result to the user.

[0027] Optionally, when in the shallow understanding mode, input the blue light suppression intensity into the screen backlight wavelength controller to generate an eye protection optical signal, and at the same time input the information density attenuation gradient into the Q&A content parameter compression engine to generate a simplified semantic unit sequence, and synchronously load the eye protection optical signal and the simplified semantic unit sequence to activate the eye protection display interface, including:

[0028] Convert the blue light suppression intensity into an adjustment pulse for the backlight wavelength. The duty cycle of the adjustment pulse has a positive linear relationship with the blue light suppression intensity, and send the adjustment pulse to the screen backlight wavelength controller through the pulse width modulation interface;

[0029] Convert the information density attenuation gradient into a hierarchical compression step size. The hierarchical compression step size has an exponential decreasing relationship with the information density attenuation gradient, and input the hierarchical compression step size into the Q&A content parameter compression engine through a step size distributor;

[0030] In the Q&A content parameter compression engine, perform cross-hierarchical folding on the original semantic units according to the hierarchical compression step size. When the semantic association strength between adjacent levels during the folding process is lower than the folding threshold, perform redundant level merging, and generate a simplified semantic unit sequence according to the merging result;

[0031] Synchronously capture the effective timestamp of the adjustment pulse and the rendering ready signal of the simplified semantic unit sequence through the hardware rendering interface. When the difference between the effective timestamp and the rendering ready signal is less than the synchronization tolerance, load the superimposed output of the eye protection optical signal and the simplified semantic unit sequence, and activate the eye protection display interface based on the superimposed output.

[0032] Optionally, according to the currently activated display interface type, extract the corresponding mapping rule group from the set of content structure mapping rules, and quantify the cognitive load level into a structure adjustment weight factor through the weight calculation unit of the information adaptation engine, and establish a proportional relationship between the structure adjustment weight factor and the compression coefficient or expansion coefficient in the mapping rule group, including:

[0033] Read the characteristic value of the optical wavelength of the current display interface through the interface type identifier. When the characteristic value is lower than the blue light suppression threshold, it is marked as an eye protection interface identifier, and when it is higher than the contrast enhancement threshold, it is marked as a focus interface identifier;

[0034] According to the eye protection interface identifier or the focus interface identifier, extract the corresponding mapping rule group from the set of content structure mapping rules respectively. The mapping rule group stores the association configuration table of the compression coefficient and the expansion coefficient in a three-dimensional tree structure;

[0035] Quantify the cognitive load level into a structure adjustment weight factor through the weight calculation unit of the information adaptation engine, and input the structure adjustment weight factor into a numerical converter to output the corresponding scalar value;

[0036] Create a parameter binding area in the association configuration table, establish a cross-section linear proportional relationship among the scalar value, the compression coefficient, and the expansion coefficient in the parameter binding area, and generate a real-time adjustment parameter group that can drive the semantic unit level marker based on the linear proportional relationship.

[0037] Optionally, according to the matching degree between the user's real-time interaction behavior and the content browsing trajectory, correct the reorganization strategy of the information adaptation engine. The corrected reorganization strategy reversely adjusts the node density of the concept topology expression and the branch depth of the logical reasoning path through the change of the cognitive load level, including:

[0038] Extract the operation event sequence from the user's real-time interaction behavior, parse the path node sequence from the content browsing trajectory, and calculate the matching degree index through the spatial coincidence degree between the event trigger position in the operation event sequence and the path node sequence;

[0039] Generate a correction coefficient for the reorganization strategy according to the fluctuation amplitude of the matching degree index within a preset time window. The correction coefficient establishes an inverse association gradient with the rising and falling trend of the cognitive load level;

[0040] Input the correction coefficient into the strategy converter to output a node density adjustment factor and a branch depth adjustment factor;

[0041] A parameter injection channel is established in the information adaptation engine, and the node density adjustment factor and the branch depth adjustment factor are respectively injected into the hierarchical folding module expressed by the concept topology and the branch generation module of the logical reasoning path through the parameter injection channel, forming a two-way feedback of the recombination strategy and the cognitive load.

[0042] In a second aspect, the present application provides an intelligent question-answering system based on a large model, including:

[0043] A receiving module, configured to, when receiving a question-answering request sent by a user, generate a cognitive load level by performing correlation analysis on the collected user eye movement trajectory data and interaction response delay data, and divide a shallow understanding mode and a deep thinking mode according to the cognitive load level;

[0044] A control module, configured to perform two-way linkage control on the generation strategies corresponding to the screen display parameters and the question-answering content parameters according to the cognitive load level. When in the shallow understanding mode, activate an eye protection display interface according to the two-way linkage control result; when in the deep thinking mode, activate a focused display interface according to the two-way linkage control result;

[0045] A recombination module, configured to, based on the cognitive load level and the optical characteristics of the current display interface, recombine the language structure and knowledge granularity of the question-answering content parameters through an information adaptation engine, generate a concept topology expression according to the recombination result under the eye protection display interface, and generate a logical reasoning path according to the recombination result under the focused display interface;

[0046] A correction module, configured to correct the recombination strategy of the information adaptation engine according to the matching degree between the user's real-time interaction behavior and the content browsing trajectory, and the corrected recombination strategy reversely adjusts the node density of the concept topology expression and the branch depth of the logical reasoning path through the change of the cognitive load level;

[0047] A generation module, configured to synchronously render the recombined question-answering content parameters, the adjusted node density or the adjusted branch depth, generate a question-answering rendering result, and display the question-answering rendering result to the user.

[0048] In a third aspect, an embodiment of the present application provides a computing device, including a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a large model-based intelligent question-answering method as described in the first aspect above.

[0049] In a fourth aspect, an embodiment of the present application provides a computer storage medium, storing a computer program, and when the computer program is executed by a computer, it implements a large model-based intelligent question-answering method as described in the first aspect.

[0050] In the embodiments of the present application, when a Q&A request sent by a user is received, through the correlation analysis of the collected user eye movement trajectory data and interaction response delay data, a cognitive load level is generated, and a shallow understanding mode and a deep thinking mode are divided according to the cognitive load level; according to the cognitive load level, a two-way linkage control is performed on the generation strategies corresponding to the screen display parameters and the Q&A content parameters. When in the shallow understanding mode, a eye protection display interface is activated according to the two-way linkage control result; when in the deep thinking mode, a focus display interface is activated according to the two-way linkage control result; based on the cognitive load level and the optical characteristics of the current display interface, the language structure and knowledge granularity of the Q&A content parameters are reorganized by an information adaptation engine. Under the eye protection display interface, a concept topology expression is generated according to the reorganization result, and under the focus display interface, a logical reasoning path is generated according to the reorganization result; according to the matching degree between the user's real-time interaction behavior and the content browsing trajectory, the reorganization strategy of the information adaptation engine is corrected, and the corrected reorganization strategy reversely adjusts the node density of the concept topology expression and the branch depth of the logical reasoning path through the change of the cognitive load level; the reorganized Q&A content parameters, the adjusted node density or the adjusted branch depth are synchronously rendered to generate a Q&A rendering result, and the Q&A rendering result is displayed to the user. The technical solution of the present application has the following beneficial effects:

[0051] By performing real-time correlation analysis on the user's eye movement trajectory and interaction delay, dynamically quantifying the user's cognitive load level, and accurately identifying two cognitive states of shallow understanding / deep thinking; synchronously regulating the screen optical characteristics (blue light suppression / contrast gain) and the Q&A content complexity (information density / logical path) based on the cognitive load level to achieve the collaborative adaptation of hardware display and software content generation; combining the current interface optical parameters, folding the semantic level or extending the logical branch of the Q&A content to generate a cognitive-friendly content structure matching the eye protection / focus mode; according to the matching degree between the user's real-time interaction behavior and the content browsing trajectory, reversely adjusting the node density of the concept topology and the branch depth of the logical path to achieve the continuous balance of content complexity and cognitive load; fusing and rendering the reorganized content structure and the adjustment parameters to ensure that the optical display effect and the information presentation rhythm match the user's cognitive absorption rate in real time.

[0052] Furthermore, a two-way generation strategy including screen display parameters (blue light suppression intensity, contrast gain value) and Q&A content parameters (information density attenuation gradient, logical path expansion intensity) is constructed; the blue light suppression intensity and information density attenuation gradient corresponding to the shallow mode, or the contrast gain value and logical path expansion intensity corresponding to the deep mode are extracted according to the cognitive load level; in the shallow mode, an eye protection optical signal and a simplified semantic sequence are synchronously generated to activate the eye protection interface, and in the deep mode, a focusing optical signal and a nested logical chain are synchronously generated to activate the focusing interface. Realize the real-time collaborative optimization of the screen optical characteristics (blue light intensity / contrast) and the Q&A content complexity (information density / logical depth). In the eye protection mode, the visual cognitive load is reduced through blue light suppression and content simplification, and in the focusing mode, the information processing depth is enhanced through contrast enhancement and logical deepening, forming a two-way gain of optical comfort and content comprehensibility.

[0053] These aspects or other aspects of the present application will be more clearly understood in the following description of the embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0055] Figure 1 The flowchart of an intelligent Q&A method based on a large model provided by the present application is shown;

[0056] Figure 2 The structural schematic diagram of an intelligent Q&A system based on a large model provided by the present application is shown;

[0057] Figure 3 The structural schematic diagram of a computing device provided by the present application is shown. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0058] In order to enable those skilled in the art to better understand the solutions of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application.

[0059] In some processes described in the specification, claims, and the above-mentioned drawings of the present application, a plurality of operations appear in a specific order. However, it should be clearly understood that these operations may not be executed in the order in which they appear herein or may be executed in parallel. The operation numbers, such as 101, 102, etc., are only used to distinguish different operations, and the numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions such as "first" and "second" in this article are used to distinguish different messages, devices, modules, etc., do not represent a sequence, and do not limit that "first" and "second" are of different types.

[0060] Researchers have found that existing intelligent question-and-answer systems generally have the problem of disconnection between the content presentation form and the user's real-time cognitive state when presenting complex information: on the one hand, the static settings of screen optical parameters (such as blue light intensity, contrast) are difficult to adapt to the dynamically changing cognitive load, which is likely to cause visual fatigue or distraction; on the other hand, the content generation strategy lacks the ability to dynamically reorganize language structure and knowledge granularity, resulting in a mismatch between information density and the user's understanding rhythm, affecting the interaction efficiency. Based on this, an intelligent question-and-answer method based on a large model is provided. Specifically, a cognitive state quantification index is generated by real-time fusion of user eye movement behavior and interaction delay data, driving the two-way collaborative control of screen display parameters and content generation strategies, and reconstructing the semantic topology and logical path of the question-and-answer content based on a dynamic feedback mechanism. This method can achieve real-time balance between optical comfort and content comprehensibility, make the information presentation rate accurately match the user's cognitive absorption ability, and significantly reduce the cognitive load in high-density knowledge interaction scenarios.

[0061] The technical solution of the present application is applicable to the scenario of dynamically adapting the personalized question-and-answer content presentation method.

[0062] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present application.

[0063] Figure 1 For the embodiments of the present application, a flowchart of an intelligent question-and-answer method based on a large model is provided, as Figure 1 shown, the method includes:

[0064] 101. When receiving a question-and-answer request sent by a user, generate a cognitive load level by correlating and analyzing the collected user eye movement trajectory data and interactive response delay data, and divide the shallow understanding mode and the deep thinking mode according to the cognitive load level;

[0065] In this step, the user's eye movement trajectory data is a sequence of gaze point coordinates captured by an infrared sensor, reflecting the movement path of the visual focus. The interactive response delay data is the time interval between the user's click / slide operation and the system feedback, which represents the cognitive processing speed.

[0066] In the embodiment of the present application, firstly, an eye tracker is used to capture the user's gaze point coordinate sequence at a sampling rate of 120 Hz, and the response delay of each interactive operation is synchronously recorded; secondly, a sliding window mechanism (500 ms window) is used to calculate the gaze dwell variance (reflecting the concentration of attention) and the standard deviation of the response delay (reflecting the fluctuation of cognitive pressure) of the user's gaze point coordinate sequence in continuous windows, and the two are weightedly fused to generate a dynamic cognitive load index; finally, the cognitive state is divided according to a preset threshold (shallow mode <0.35, deep mode >0.65), and the mode switching is triggered when the load index of three consecutive windows is stable in the threshold range.

[0067] Suppose that on a programming teaching platform, a user initiates a question-and-answer request for "Python recursive algorithm optimization". The system detects that his eye movement trajectory repeatedly scans back at the recursive termination condition of the code example (residence variance 0.31), and the response time of the debugging operation is extended to 14 seconds (standard deviation 0.48). After weighted calculation, the cognitive load index is generated to be 0.72, indicating that the user has entered deep thinking mode.

[0068] 102. According to the cognitive load level, the generation strategy corresponding to the screen display parameters and the question and answer content parameters is bidirectionally linked and controlled, and when in the shallow understanding mode, the eye protection display interface is activated according to the bidirectional linkage control result; when in the deep thinking mode, the focus display interface is activated according to the bidirectional linkage control result;

[0069] In this step, the screen display parameters include blue light suppression intensity (400-450nm wavelength attenuation ratio) and contrast gain value (the increase relative to the baseline value). Question and answer content parameters include information density attenuation gradient (semantic unit folding rate) and logic path expansion intensity (the number of branch level increases).

[0070] In the embodiments of the present application, first, the semantic theme of the user's current Q&A request is parsed (such as algorithm optimization, historical event analysis), and the preset parameter mapping rules are matched according to the theme type: for example, for technical questions, in the shallow mode, the blue light suppression intensity (70% wavelength attenuation) and the information density attenuation gradient (60% folding rate) are called, and in the deep mode, the contrast gain value (150%) and the logical path expansion intensity (+3 layers) are enabled; secondly, the blue light suppression intensity is converted into a backlight drive pulse sequence (PWM duty cycle 70%), and the information density attenuation gradient is synchronously input into the content compression engine to fold non-core cases (such as secondary code comments) according to the theme relevance; thirdly, for logical questions in the deep mode, the contrast gain value is injected into the pixel drive module to generate a high-sharpness display waveform, and at the same time, the time complexity comparison and the memory optimization scheme branch are added to the key argument nodes (such as recursive stack analysis) through the inference engine; finally, when the optical parameter and the content parameter ready flags are both valid, the activation instruction of the eye protection display interface or the focused display interface is triggered through the hardware signal synchronizer.

[0071] For example, continuing the above example, after the user enters the deep thinking mode, the system increases the screen contrast to 150% to enhance the code readability, and adds two extension branches, namely, the "stack frame comparison diagram" and the "Python decorator implementation case", to the "tail recursion optimization" node in the "recursive algorithm optimization" analysis, and at the same time folds the detailed code in the "basic recursive example" part.

[0072] 103. Based on the cognitive load level and the optical characteristics of the current display interface, the information adaptation engine reorganizes the language structure and knowledge granularity of the Q&A content parameters, generates a concept topology expression according to the reorganization result under the eye protection display interface, and generates a logical reasoning path according to the reorganization result under the focused display interface;

[0073] In this step, the concept topology expression is a core concept tree formed by folding secondary semantic units, and strong association edges are retained between nodes. The logical reasoning path is a multi-layer logical chain formed by extending auxiliary arguments, and the main path nodes carry verification cases.

[0074] In the embodiments of the present application, first, a restructuring strategy is selected according to the topic type (technical / theoretical) of the Q&A request: for technical questions, the code structure is preferentially retained, and for theoretical questions, the logical chain is strengthened; second, in the eye protection mode, a syntactic analyzer is used to extract core semantic units (such as function definitions, key parameters), and auxiliary explanations with a level ≥ 3 (such as exception handling code blocks) are folded according to the compression coefficient (0.6) to generate a conceptual topology expression; third, in the focus mode, the logical backbone is identified through a dependency parser, and branches are added to nodes with a level ≤ 2 according to the expansion coefficient (1.5), and an interactive code comparison window is embedded in each branch to generate a logical reasoning path; finally, the restructured structure is converted into a visual component with an indentation level.

[0075] For example, continuing with the above example, in the eye protection mode, the system folds the underlying implementation code of "tail recursion optimization" into a three-level topology of "decorator syntax → stack frame optimization → performance comparison"; after switching to the focus mode, two practical branches of "Dis module bytecode analysis" and "PyCharm debugger memory monitoring" are expanded under the "stack frame optimization" node, and each branch is accompanied by a runnable code snippet.

[0076] 104. According to the matching degree between the user's real-time interaction behavior and the content browsing track, the restructuring strategy of the information adaptation engine is corrected, and the corrected restructuring strategy reversely adjusts the node density of the conceptual topology expression and the branch depth of the logical reasoning path through the change of the cognitive load level;

[0077] In this step, the node density is the number of effective nodes per unit area in the conceptual topology. The branch depth is the maximum number of levels from the root node to the leaf node in the logical reasoning path.

[0078] In the embodiments of the present application, first, the matching degree between the actual click sequence of the user and the preset content browsing path is calculated through the dynamic time warping algorithm, and correction is triggered when the matching degree of technical questions < 0.6 or theoretical questions < 0.7; second, the parameters are adjusted reversely according to the change direction of the cognitive load: when the load rises, the node density is reduced by 12% per 10% increase (merging adjacent technical detail nodes) and the branch depth is increased by 1 level (adding a floating window of reference documents for theoretical nodes); when the load drops, the folded code comments are unfolded per 10% decrease; finally, the interface is locally refreshed through an incremental update engine, and only the changed area (such as the canvas partition where the extended branch is located) is redrawn.

[0079] For example, continuing with the above example, when the user returns to the parent node multiple times while studying "Dis module bytecode analysis", the system detects that the matching degree drops to 0.58, immediately merges "opcode parsing" and "instruction stream visualization" into a composite node, adds a link to Chapter 7 of "Python Source Code Analysis" under the "performance comparison" branch, and at the same time reduces the code folding rate to display more implementation details.

[0080] 105. Synchronously render the reorganized Q&A content parameters, adjusted node density, or adjusted branch depth to generate a Q&A rendering result, and display the Q&A rendering result to the user.

[0081] In this step, the Q&A rendering result is a composite output including optical parameters (backlight wavelength, contrast value) and content layout (node coordinates, connection line style).

[0082] In the embodiment of the present application, first, input the reorganized Q&A content parameters (concept topology / logical path) and real-time adjustment parameters (node density / branch depth) into the rendering engine; secondly, calculate the basic spacing of the topological layout according to the node density (the higher the density value, the smaller the spacing), and at the same time set the extension length of the logical connection line according to the branch depth (the greater the depth value, the longer the extension); thirdly, based on the basic spacing and extension length, synchronously adjust the screen backlight wavelength (enable a 450nm filter in the eye protection mode) or the contrast curve (increase to 150% in the focus mode) through the optical parameter driving module; finally, convert the semantic structure data into visual elements (interactive nodes, logical connection lines with arrows), generate a complete Q&A rendering result in combination with real-time optical parameters, and refresh the user interface through the display controller.

[0083] For example, continuing the above example, after the user's Q&A request for "recursive algorithm memory optimization" enters the deep thinking mode, the system automatically arranges key concepts such as "tail recursion → stack frame optimization → decorator implementation" according to the adjusted node density (display 5 core nodes per screen), and compresses the spacing to a compact layout; at the same time, based on the branch depth (3-layer expansion), extend the logical connection line of the "stack frame optimization" node to the sub-nodes of "memory allocation comparison diagram" and "PyCharm debugging case"; synchronously enable a 150% contrast to enhance the border of the core nodes; finally, render a highly readable code topology diagram, where the core nodes are suspended and displayed with high contrast, and the sub-nodes are folded into expandable suspended labels.

[0084] Steps 101-105 dynamically sense the user's cognitive load and coordinate the screen optical characteristics and content presentation form in real time: reduce blue light interference and simplify the information hierarchy in the shallow understanding mode to relieve visual fatigue; enhance the contrast and expand the logical chain in the deep thinking mode to improve the analysis depth; continuously optimize the content density and structural complexity in combination with behavior feedback to form a closed-loop system of coordinated adaptation among optical parameters, information organization, and cognitive state, and significantly optimize the information absorption efficiency and interaction comfort in high-density knowledge interaction scenarios.

[0085] In order to solve the problem of the disconnection between visual comfort and information complexity and the lack of dynamic collaborative adaptation in the content presentation of existing intelligent question - answering systems, and to further improve the cognitive efficiency of human - machine interaction, in some embodiments, the two - way linkage control is performed on the generation strategies corresponding to the screen display parameters and the question - answering content parameters according to the cognitive load level. When in the shallow understanding mode, the eye - protection display interface is activated according to the two - way linkage control result; when in the deep thinking mode, the focused display interface is activated according to the two - way linkage control result, including:

[0086] 201. Construct the generation strategies for the screen display parameters and the question - answering content parameters. The screen display parameters include the blue - light suppression intensity and the contrast gain value, and the question - answering content parameters include the information density attenuation gradient and the logical path expansion intensity;

[0087] In step 201, the screen display parameters include the blue - light suppression intensity (the attenuation ratio of the light intensity in the 400 - 450nm wavelength band in the screen backlight) and the contrast gain value (the increase amplitude of the brightness ratio between the bright and dark areas of the display screen). The question - answering content parameters include the information density attenuation gradient (the proportion of the core semantic units retained in the unit text area) and the logical path expansion intensity (the number of levels of the newly added auxiliary argument branches in the reasoning chain).

[0088] In the embodiments of the present application, first, a mapping rule library for the screen display parameters and the question - answering content parameters is established: for technical analysis questions, the blue - light suppression intensity and the information density attenuation gradient are set to be negatively correlated (the stronger the blue - light suppression, the higher the information folding rate); for theoretical derivation questions, the contrast gain value and the logical path expansion intensity are set to be positively correlated (the higher the contrast, the more levels of branches). Secondly, the parameter combinations of typical question - answering scenarios are marked by expert experience to form a decision matrix containing 12 preset strategies, and each strategy defines the corresponding relationship between the optical parameter threshold range and the content parameter adjustment amplitude. Finally, the strategy library is encoded into a dynamically loadable configuration file for the real - time decision - making engine to call.

[0089] 202. According to the numerical range of the cognitive load level, extract the blue - light suppression intensity and the information density attenuation gradient corresponding to the shallow understanding mode, and the contrast gain value and the logical path expansion intensity corresponding to the deep thinking mode from the generation strategies;

[0090] In step 202, the numerical range of the cognitive load level is a comprehensive scoring interval calculated by weighting the eye - movement trajectory stability index (0 - 1) and the delay fluctuation coefficient (0 - 1).

[0091] In the embodiments of the present application, first, a shallow understanding mode threshold (0 - 0.4) and a deep thinking mode threshold (0.6 - 1.0) are set, and the middle interval (0.4 - 0.6) is the transition state; second, when the load score enters the shallow mode interval, a parameter group marked as "low load adaptation" is retrieved from the policy library, and the blue light suppression intensity (such as 70% wavelength attenuation) and the information density attenuation gradient (such as 60% folding rate) are obtained; when the score enters the deep mode interval, a "high load adaptation" parameter group is retrieved, and the contrast gain value (such as 150%) and the logical path expansion intensity (such as +3 layers) are obtained; finally, the validity of the extracted values is verified by a parameter verification module, and if it exceeds the hardware support range, it is automatically scaled proportionally.

[0092] 203. When in the shallow understanding mode, the blue light suppression intensity is input into the screen backlight wavelength controller to generate an eye - protecting optical signal, and at the same time, the information density attenuation gradient is input into the compression engine corresponding to the Q&A content parameters to generate a simplified semantic unit sequence, and the eye - protecting optical signal and the simplified semantic unit sequence are synchronously loaded to activate the eye - protecting display interface;

[0093] In step 203, the eye - protecting optical signal is a specific band light intensity suppression waveform output by the backlight wavelength controller. The simplified semantic unit sequence is a set of core concepts retained after being processed by the content compression engine.

[0094] In the embodiments of the present application, first, the blue light suppression intensity is converted into a pulse width modulation (PWM) signal to control the light intensity attenuation of the band below 450nm in the backlight LED array to the target value; synchronously, the information density attenuation gradient is input into the natural language processing engine, and non - core semantic units such as attributive clauses and illustrative examples are identified and deleted through dependency syntax analysis to generate a simplified sequence containing only the subject - predicate - object structure; second, a timing synchronizer is used to ensure that the backlight adjustment and content compression are completed within a 200ms time window; finally, the graphics rendering interface is called to convert the simplified sequence into a tree - shaped topology map, and it is combined with the adjusted backlight parameters and loaded into the display buffer to activate the eye - protecting display interface.

[0095] 204. When in the deep thinking mode, the contrast gain value is input into the screen pixel driving module to generate a focusing optical signal, and at the same time, the logical path expansion intensity is input into the inference engine corresponding to the Q&A content parameters to generate a nested logical chain sequence, and the focusing optical signal and the nested logical chain sequence are synchronously loaded to activate the focusing display interface.

[0096] In step 204, the focusing optical signal is a high - contrast display waveform generated by the pixel driving module. The nested logical chain sequence is a multi - layer argument structure formed after being expanded by the inference engine.

[0097] In an embodiment of the present application, the contrast gain value is first converted into a pixel voltage adjustment curve to dynamically increase the difference between light and dark areas to a set ratio; the logic path expansion strength is simultaneously input into the knowledge graph reasoning engine to add three types of branches, namely counterexample verification, historical background, and cross-disciplinary association, to the original argumentation node; secondly, the expanded logic chain is converted into a visual structure with an indented hierarchy through a hardware accelerated rendering pipeline, in which the main path node has a bold border and the auxiliary branches are presented in a translucent floating frame; finally, the interface refresh is triggered by a vertical synchronization signal to activate the focus display interface.

[0098] Here is a specific example:

[0099] In the online question-and-answer scenario of machine learning model tuning, the user initiates a request for the "overfitting solution", and it is detected that his eye movement trajectory frequently stays at the regularization formula (load score 0.68), triggering the deep thinking mode: Step 201 calls the technical problem strategy library, selects the contrast gain value (160%) and the logic path expansion strength (+4 layers); Step 202 extracts the parameter group, verifies that the contrast value is within the screen hardware support range (maximum value 200%), and adjusts it to 144% in proportion; Step 204 injects the adjusted contrast gain into the pixel driver module to generate a high-sharpness display waveform, and simultaneously adds three branches of "weight decay coefficient optimization", "cross-validation curve comparison", and "TensorBoard visualization case" to the "L2 regularization" node; in the final activated focus display interface, the core formula is highlighted with a contrast of 144%, and the newly added branch is suspended on the right side of the node in the form of a folded label. The user clicks to expand the detailed derivation process and code snippet.

[0100] Steps 201-202 achieve real-time collaborative adaptation of screen display characteristics and information organization forms by constructing a dynamic mapping strategy between optical parameters and content parameters: reduce blue light interference and simplify content structure in shallow understanding scenarios to reduce visual cognitive load; enhance contrast and expand logical chains in deep thinking scenarios to improve efficiency in complex problem analysis; combine hardware-level signal synchronization mechanisms to ensure seamless connection between optical adjustment and content reorganization, form an adaptive question-and-answer interface that conforms to ergonomic cognitive laws, and significantly optimize information absorption rate and operational comfort in high-density knowledge interaction scenarios.

[0101] To solve the problems of the disconnection between optical parameters and semantic structures and the lack of hierarchical cognitive adaptation in the existing Q&A systems when dynamically adapting content presentation, and to further improve the information comprehensibility in complex knowledge scenarios, in some embodiments, based on the cognitive load level and the optical characteristics of the current display interface, the information adaptation engine reorganizes the language structure and knowledge granularity of the Q&A content parameters, generates a conceptual topology expression according to the reorganization result under the eye protection display interface, and generates a logical reasoning path according to the reorganization result under the focused display interface, including:

[0102] 301. Load, through the information adaptation engine, a set of content structure mapping rules bound to the optical characteristics of the current display interface, where the eye protection display interface corresponds to the first mapping rule group in the set of content structure mapping rules, and the focused display interface corresponds to the second mapping rule group in the set of content structure mapping rules;

[0103] In step 301, the set of content structure mapping rules includes a strategy library of semantic unit processing rules in different display modes, and the rule group defines the intensity threshold for content folding / expansion.

[0104] In the embodiments of the present application, first, a set of rules is preset in the information adaptation engine. The eye protection display interface corresponds to the first rule group (defining the semantic level compression coefficient range of 0.3 - 0.7), and the focused display interface corresponds to the second rule group (defining the logical branch expansion coefficient range of 1.2 - 2.0); second, a hardware-level binding relationship between the optical parameters and the rule group is established. When it is detected that the backlight wavelength is lower than 450 nm, the first rule group is automatically associated, and when the contrast ratio is higher than 140%, the second rule group is associated; finally, the rule group corresponding to the activated interface type is loaded into the real-time processing memory through a dynamic loader.

[0105] 302. According to the currently activated display interface type, extract the corresponding mapping rule group from the set of content structure mapping rules, and quantify the cognitive load level into a structure adjustment weight factor through the weight calculation unit of the information adaptation engine, and establish a proportional relationship between the structure adjustment weight factor and the compression coefficient or expansion coefficient in the mapping rule group;

[0106] In step 302, the structure adjustment weight factor is a scalar value reflecting the adjustment amplitude of the cognitive load on the content structure deformation intensity. The proportional relationship is a linear or non-linear mathematical relationship between the weight factor and the rule group parameters (compression / expansion coefficient).

[0107] In the embodiment of the present application, first, the weight calculation unit receives the cognitive load level data and normalizes it into a weight factor in the range of 0-1 (for example, a load level of 0.8 corresponds to a weight of 0.95); secondly, a proportional function is selected according to the rule group type to establish a proportional relationship: in the eye protection mode, the weight factor is positively correlated with the compression coefficient (a weight of 0.95 maps to a compression coefficient of 0.65), and in the focus mode, the weight factor is piecewise linearly correlated with the expansion coefficient (a weight of 0.8-1.0 maps to an expansion coefficient of 1.8-2.0); finally, the calculated coefficient is written into the rule executor to overwrite the default parameter values in the rule group.

[0108] 303. Based on the proportional relationship, the language structure parsing unit of the information adaptation engine hierarchically marks the core semantic units of the Q&A content parameters. Under the eye protection display interface, the semantic units with a hierarchical mark higher than the preset level are folded according to the compression coefficient of the first mapping rule group, and a concept topology expression is generated based on the folded semantic units.

[0109] In step 303, the hierarchical marking is a 1-N level classification marking of text units based on semantic importance. The folding process is a process of deleting or aggregating semantic units below the importance threshold.

[0110] In the embodiment of the present application, first, the language structure parsing unit performs dependency analysis on the original content, identifies the subject-predicate-object core structure and marks it as level 1, and marks attributive clauses and illustrative examples as ≥ level 3; secondly, the folding threshold is calculated according to the compression coefficient (such as 0.65) (level ≥ 3×0.65≈2), and the units with a marked level > 2 are deleted; thirdly, the topological connection strength of the remaining units is calculated, and only the edges with an association strength > 0.6 are retained; finally, a concept topology expression containing only the core nodes and their strong association edges is generated.

[0111] 304. Under the focus display interface, the knowledge granularity control unit of the information adaptation engine performs branch extension processing on the semantic units with a marked level lower than the preset level according to the expansion coefficient of the second mapping rule group, and generates a logical reasoning path based on the branch extension processed semantic units.

[0112] In step 304, the branch extension adds auxiliary arguments or verification cases to the low-level semantic units. The logical reasoning path is a multi-level node connection structure with a clear derivation direction.

[0113] In the embodiments of the present application, first, a semantic unit with a label level ≤ 2 is identified by a knowledge granularity control unit; second, the extendable number of layers (basic layer number 1 + 1.8 ≈ 3 layers) is calculated according to an expansion coefficient (such as 1.8), and three types of branches, namely historical background, counterexample verification, and cross-reference, are added to each target node; third, redundant connections are eliminated through a path optimization algorithm to ensure the priority of the main path visibility; finally, a logical reasoning path with collapsible labels is generated, where the main nodes are displayed in bold and the auxiliary branches are folded by default.

[0114] The following is a specific example:

[0115] Suppose the user queries "the pathogenesis of coronary atherosclerosis" in a medical knowledge Q&A. In step 301, it is detected that the current is the eye protection mode (backlight wavelength 430nm), and the first rule group (compression coefficient 0.6) is loaded; in step 302, the weight factor 0.85 is calculated according to the cognitive load level 0.75, and the actual compression coefficient 0.6 × 0.85 = 0.51 is obtained by mapping; in step 303, the original content is parsed, and the main chain of "abnormal lipid metabolism → endothelial injury → inflammatory response → plaque formation" is marked as level 1-2, and the detailed biochemical reaction process is marked as level 3; the content with a level > 1.53 (≈ 2 levels) is folded according to the compression coefficient 0.51 to generate a two-level concept topology expression of "abnormal blood lipid → plaque formation"; after the user switches to the focus mode, in step 301, the second rule group (expansion coefficient 1.8) is loaded; in step 304, three layers of branches, namely "LDL receptor gene mutation", "quantitative analysis of dietary factors", and "action path of statins", are extended for the node of "abnormal lipid metabolism", and a dynamic demonstration diagram of the molecular structure that can be expanded is embedded in each layer of the branch to generate a logical reasoning path.

[0116] Steps 301-304 implement an accurate matching of the content reorganization strategy and the display environment through a dynamic rule loading mechanism driven by optical characteristics: in the eye protection mode, secondary semantic details are compressed to construct a concise cognitive framework to reduce the understanding burden; in the focus mode, multi-level argument branches are extended to deepen the logical integrity of key knowledge points; the reorganization intensity is dynamically adjusted in combination with the weight factor to make the content complexity adapt to the change of the user's cognitive load in real time, forming an adaptive knowledge presentation system with triple coordination of "optical environment - content structure - cognitive state", and significantly improving the information transmission efficiency and cognitive comfort in high-professional Q&A scenarios.

[0117] To solve the problems of the disconnection between the layout of visual elements and the display of logical relationships and the lack of dynamic environment adaptation ability in the rendering process of existing Q&A content, and to further improve the visual cognitive efficiency in complex information scenarios, in some embodiments, synchronously rendering the reorganized Q&A content parameters, adjusted node density, or adjusted branch depth to generate a Q&A rendering result, and presenting the Q&A rendering result to the user includes:

[0118] 401. Synchronously render the reorganized Q&A content parameters with the adjusted node density to generate content distribution data with density markers; or synchronously render the reorganized Q&A content parameters with the adjusted branch depth to generate content distribution data with depth markers.

[0119] In step 401, the density marker is a quantitative label reflecting the number of nodes in a unit area in the concept topology. The depth marker is a visual identifier characterizing the branch levels of the logical reasoning path.

[0120] In the embodiment of the present application, first select the rendering dimension according to the type of the reorganized Q&A content parameters: for the concept topology structure, extract the adjusted node density value and encode it as a density marker (such as 5 nodes per square centimeter marked as D5); for the logical reasoning path, extract the adjusted branch depth value and encode it as a depth marker (such as 3 levels marked as L3); then bind the semantic unit coordinates with different types of marker values through a spatial mapping algorithm to generate content distribution data with optical parameter adaptation identifiers.

[0121] 402. Select the corresponding density marker or depth marker in the content distribution data according to the current display interface type, calculate the basic spacing of visual elements according to the density marker or calculate the connection line strength of visual elements according to the depth marker. The basic spacing has an inverse fluctuation relationship with the adjusted node density, and the connection line strength has a positive fluctuation relationship with the adjusted branch depth.

[0122] In step 402, the basic spacing is the minimum interval distance between the center points of adjacent visual elements. The connection line strength is the visual salience of the logical connection line (including thickness, transparency, animation frequency).

[0123] In the embodiment of the present application, first analyze the marker type in the content distribution data: read the density marker in the eye protection mode and calculate the basic spacing through an inverse proportional function (such as D5 corresponding to a spacing of 12px); read the depth marker in the focus mode and calculate the connection line strength through a direct proportional function (such as L3 corresponding to a line width of 3px and a transparency of 70%); secondly, adaptively scale the calculation results according to the current screen resolution to ensure visual consistency under different device sizes; finally, write the adaptively scaled parameter set into the rendering configuration file.

[0124] 403. Adjust the arrangement distribution of the Q&A content parameters based on the basic spacing, or generate logical connection identifiers based on the connection line strength, and form a switchable dynamic visual template according to the adjusted arrangement distribution or the logical connection identifiers.

[0125] In step 403, the logical connection identifier is an interactive graphical element used to indicate the direction of the inference path. The switchable template is an interface layout solution that supports instant switching between the eye protection / focus modes.

[0126] In the embodiment of the present application, first, the basic spacing parameter is called in the eye protection mode, and the force-directed layout algorithm is used to automatically arrange the concept nodes based on the basic spacing parameter, ensuring that the spacing between adjacent nodes meets the calculated value and there is no overlap; second, in the focus mode, a gradient connection line with an arrow is generated according to the connection line strength parameter, and an expandable identifier (such as a triangle icon) is added to the end branch node to form the logical connection identifier; third, the arranged concept nodes and the logical connection identifier are encapsulated into a dynamically loadable dynamic visual template through a template combiner, supporting millisecond-level mode switching.

[0127] 404. Convert the dynamic visual template into a screen pixel drive signal and a content rendering instruction, input the screen pixel drive signal and the content rendering instruction into the display controller, and then output a Q&A rendering result, and display the Q&A rendering result to the user.

[0128] In step 404, the pixel drive signal is a hardware instruction for controlling optical characteristics such as the backlight wavelength and contrast of the screen. The content rendering instruction is a set of graphic drawing commands for describing the positions and styles of visual elements.

[0129] In the embodiment of the present application, first, the optical parameters (such as the 450nm backlight in the eye protection mode) in the dynamic visual template are converted into a PWM pulse sequence through a signal converter; second, the node layout data and the connection line style are compiled into OpenGL drawing instructions through a graphics engine; third, the PWM pulse sequence and the OpenGL drawing instructions are synchronously sent through the vertical synchronization interface of the display controller to ensure the timing alignment of the backlight adjustment and the content refresh; finally, the full interface rendering update is completed within the screen refresh cycle, and the Q&A rendering result is output and displayed to the user.

[0130] The following is a specific example:

[0131] Suppose the user queries "the criteria for determining justifiable defense" in the Q&A scenario of legal provision analysis: Step 401 generates content distribution data with density marker D6 (6 core nodes per screen) in the eye protection mode, including nodes such as "defense intention → urgency of infringement → limit conditions"; Step 402 calculates the basic spacing as 10px and adopts a compact layout to avoid frequent scrolling; Step 403 automatically arranges the nodes to form a dynamic visual template, and the key provisions are displayed in bold font. After the user switches to the focus mode, Step 401 generates depth marker L4 content data, extending three layers of branches for the "limit conditions" node: "judicial interpretation cases → sentencing guidelines → international law comparison"; Step 402 calculates the connection line strength as a line width of 4px and a pulse animation frequency of 2Hz to enhance the path guidance; Step 404 synchronously increases the contrast to 160% to render a Q&A rendering result with dynamic arrow flow, and the summary of the judgment document is expanded after clicking on the case branch.

[0132] Steps 401-404 achieve precise adaptation of the information presentation form to the display environment through the collaborative calculation of tokenized content data and dynamic visual parameters: optimize the space utilization rate in the eye protection mode and reduce the visual search cost through a compact layout; strengthen the logical guidance in the focus mode and improve the path tracking efficiency through animated connection lines; combine with the hardware-level synchronous rendering technology to ensure the real-time balance of optical comfort and content readability, forming an adaptive visual Q&A interface across devices and scenarios, and significantly improving the interactive analysis efficiency of complex knowledge points in professional fields.

[0133] To solve the problems of visual stuttering and cognitive interference caused by the asynchronous adjustment of optical parameters and content simplification processing during the activation of the existing eye protection display interface, and further improve the information acquisition fluency in low-load scenarios, in some embodiments, when in the shallow understanding mode, the blue light suppression intensity is input into the screen backlight wavelength controller to generate an eye protection optical signal, and at the same time, the information density attenuation gradient is input into the Q&A content parameter compression engine to generate a simplified semantic unit sequence, and the eye protection optical signal and the simplified semantic unit sequence are synchronously loaded to activate the eye protection display interface, including:

[0134] 501. Convert the blue light suppression intensity into an adjustment pulse for the backlight wavelength, where the duty cycle of the adjustment pulse has a positive linear relationship with the blue light suppression intensity, and send the adjustment pulse to the screen backlight wavelength controller through a pulse width modulation interface;

[0135] In Step 501, the adjustment pulse is a periodic square wave signal generated by pulse width modulation, which is used to precisely control the backlight wavelength.

[0136] In the embodiments of the present application, first, the target wavelength attenuation ratio is calculated according to the blue light suppression intensity (such as 70%), and it is mapped to a pulse duty cycle (such as 70% high level); secondly, the pulse duty cycle parameter is converted into a PWM square wave signal through a digital-to-analog converter, where the pulse frequency is synchronized with the screen refresh rate; finally, the square wave signal is transmitted to the backlight wavelength controller through a hardware interface to drive the light emission intensity attenuation of the short-wave blue light chips in the LED array.

[0137] 502. Convert the information density attenuation gradient into a hierarchical compression step size, where the hierarchical compression step size has an exponential decreasing relationship with the information density attenuation gradient, and input the hierarchical compression step size into the Q&A content parameter compression engine through a step size distributor;

[0138] In step 502, the information density attenuation gradient is the proportion of core semantic units to be retained in a unit text area. The hierarchical compression step size is the number of levels skipped each time during the semantic unit folding process.

[0139] In the embodiments of the present application, first, an exponential mapping relationship between the information density attenuation gradient and the hierarchical compression step size is established (such as gradient 0.6 corresponding to step size 2^0.6≈1.5 levels); secondly, the information density attenuation gradient value is converted into a floating-point step size parameter through a non-linear interpolation algorithm; finally, the floating-point step size parameter is distributed to each processing thread of the Q&A content compression engine through a step size distributor to ensure parameter consistency during multi-module parallel operations.

[0140] 503. In the Q&A content parameter compression engine, perform cross-level folding on the original semantic units according to the hierarchical compression step size, and when the semantic association strength between adjacent levels during the folding process is lower than the folding threshold, perform redundant level merging, and generate a simplified semantic unit sequence according to the merging result;

[0141] In step 503, the semantic association strength is an inter-unit association degree index calculated based on the co-occurrence frequency and the dependency distance. Redundant level merging is the process of deleting intermediate levels with an association strength lower than the threshold and directly connecting the upper and lower nodes.

[0142] In the embodiments of the present application, first, traverse the original semantic hierarchical structure according to the hierarchical compression step size, and mark the hierarchical intervals to be folded; secondly, calculate the association strength between adjacent levels, and when the strength value is lower than the folding threshold (such as 0.4), delete the intermediate level and directly connect the upper node to the nearest valid lower node; thirdly, perform loop detection and conflict resolution on the structure after redundant level merging to ensure topological logic integrity; finally, output a simplified semantic unit sequence that retains the core path.

[0143] 504. Synchronously capture the effective timestamp of the adjustment pulse and the rendering ready signal of the simplified semantic unit sequence through the hardware rendering interface. When the difference between the effective timestamp and the rendering ready signal is less than the synchronization tolerance, load the superimposed output of the eye protection optical signal and the simplified semantic unit sequence, and activate the eye protection display interface based on the superimposed output.

[0144] In step 504, the effective timestamp is the hardware feedback time when the backlight controller completes wavelength adjustment. The rendering ready signal is the completion identifier for the content compression engine to output the simplified sequence.

[0145] In the embodiment of the present application, first, capture the interrupt signal of the completion of backlight adjustment through the hardware interrupt mechanism, and record the timestamp accurate to the microsecond level; second, monitor the status register of the content compression engine, and generate a synchronization trigger signal when detecting the jump of the "rendering ready" flag bit; third, calculate the difference between the timestamp and the ready signal. If it is less than the preset tolerance (such as 16 ms), then call the graphics synthesizer to superimpose and output the eye protection optical signal and the simplified semantic unit sequence; finally, refresh the interface through the vertical synchronization signal of the display controller to activate the eye protection display interface.

[0146] The following is a specific example:

[0147] Suppose in the Q&A scenario of philosophical concept analysis, the user queries "Kant's transcendental philosophical system". Step 501 converts the blue light suppression intensity of 70% into a PWM pulse with a duty cycle of 70% to drive the backlight wavelength to drop below 450 nm; step 502 maps the information density attenuation gradient of 0.6 to a hierarchical compression step size of 1.5 and inputs it into the content compression engine; step 503 folds the original content hierarchy: compresses the three-layer structure of "Critique of Pure Reason → Transcendental Aesthetic → Forms of Intuition of Space and Time" into a direct connection path of "Critique of Pure Reason → Forms of Intuition of Space and Time", deletes the intermediate transition layer with an association intensity of 0.38, and generates a simplified semantic unit sequence; step 504 captures the content rendering ready signal within 15 ms after the backlight adjustment is completed, synchronously loads the eye protection optical signal and the simplified "Topological Map of the Relationship between the Three Critiques", and activates the eye protection display interface after the interface switches without flickering.

[0148] Steps 501 - 504 achieve instantaneous coordination of optical comfort and information simplicity through the hardware-level synchronization mechanism of backlight wavelength adjustment and content hierarchical compression: precisely control the blue light radiation intensity while dynamically optimizing the content structure complexity, eliminating the visual jump caused by the lag of parameter adjustment in the traditional scheme; combine the cross-layer folding algorithm to retain the core cognitive framework, maintain logical coherence while reducing the information density, form an eye protection display interface that conforms to the low-load cognitive law, and significantly improve the knowledge acquisition efficiency and visual comfort experience in the shallow understanding scenario.

[0149] To solve the problem of insufficient cross-scenario adaptability caused by the static binding of optical parameters and semantic recombination rules in existing content adaptation strategies, and to further improve the cognitive load matching accuracy in a dynamic environment, in some embodiments, according to the currently activated display interface type, extract the corresponding mapping rule group from the content structure mapping rule set, and quantify the cognitive load level into a structure adjustment weight factor through the weight calculation unit of the information adaptation engine, and establish a proportional relationship between the structure adjustment weight factor and the compression coefficient or expansion coefficient in the mapping rule group, including:

[0150] 601. Read the characteristic value of the optical wavelength of the current display interface through an interface type identifier. When the characteristic value is lower than the blue light suppression threshold, it is marked as an eye protection interface identifier, and when it is higher than the contrast enhancement threshold, it is marked as a focus interface identifier;

[0151] In step 601, the eye protection interface identifier is a mode activation label that triggers blue light suppression and content simplification. The focus interface identifier is a mode activation label that enables high contrast and logical deepening.

[0152] In the embodiments of the present application, first, continuously sample the backlight wavelength data through a spectral sensor, and extract the intensity in the 400-450nm band as the characteristic value; secondly, compare the characteristic value with a preset threshold: when the data of 5 consecutive frames is lower than the blue light suppression threshold (such as 50 lumens), generate an eye protection interface identifier, and when it is higher than the contrast enhancement threshold (such as 120 lumens), generate a focus interface identifier; finally, write the identifier into the mode status register through an interrupt service program.

[0153] 602. According to the eye protection interface identifier or the focus interface identifier, respectively extract the corresponding mapping rule group from the content structure mapping rule set, and the mapping rule group stores an association configuration table of the compression coefficient and the expansion coefficient in a three-dimensional tree structure;

[0154] In step 602, the three-dimensional tree association configuration table is a rule library that stores parameter relationships with the interface type, semantic category, and load level as three-dimensional coordinates. The compression coefficient is an intensity control parameter for folding processing of semantic units. The expansion coefficient is an intensity control parameter for extending processing of logical branches.

[0155] In the embodiments of the present application, first, determine the first dimension index of the three-dimensional tree according to the interface identifier (0 on the X-axis for the eye protection mode, 1 on the X-axis for the focus); secondly, parse the semantic category (technical / theoretical) of the current Q&A content as the Y-axis index; thirdly, determine the Z-axis index based on the interval where the cognitive load level is located; finally, store the association configuration table of the compression coefficient and the expansion coefficient in the three-dimensional tree node.

[0156] 603. Quantify the cognitive load level into a structure adjustment weight factor through the weight calculation unit of the information adaptation engine, and input the structure adjustment weight factor into a numerical converter to output a corresponding scalar value;

[0157] In step 603, the structure adjustment weight factor is the adjustment weight reflecting the deformation intensity of the content structure by the current cognitive load. The scalar value is the normalized operable parameter value.

[0158] In the embodiment of the present application, first, the weight calculation unit receives the cognitive load level data and uses the S-shaped function to map it to the 0-1 interval to generate a structure adjustment weight factor; second, select the numerical conversion rule according to the interface type: linear scaling is adopted in the eye protection mode (weight 0.8 → scalar value 0.8), and square amplification is adopted in the focus mode (weight 0.8 → scalar value 0.64); finally, write the output scalar value into the parameter buffer queue for downstream modules to call.

[0159] 604. Create a parameter binding area in the association configuration table, and establish a cross-segment linear proportional relationship among the scalar value, the compression coefficient, and the expansion coefficient in the parameter binding area, and generate a real-time adjustment parameter group that can drive the semantic unit level marking based on the linear proportional relationship.

[0160] In step 604, the cross-segment linear proportion is a dynamic binding mechanism that establishes a differential proportional relationship in different parameter intervals. The real-time adjustment parameter group is an operation instruction set that drives the semantic processing engine to perform level marking.

[0161] In the embodiment of the present application, first, create a temporary storage area in the association configuration table, and store the scalar value and the compression / expansion coefficient of the current rule group according to the interface type; second, establish a negative proportional relationship for the eye protection mode (for every 0.1 increase in the scalar value, the compression coefficient increases by 0.08), and establish a positive proportional relationship for the focus mode (for every 0.1 increase in the scalar value, the expansion coefficient increases by 0.12); finally, generate a real-time adjustment parameter group that can drive the semantic unit level marking through the parameter binding engine and push it to the semantic processing pipeline.

[0162] The following is a specific example:

[0163] Suppose in a financial investment knowledge Q&A scenario, when a user studies "quantitative hedging strategy": In step 601, the current backlight wavelength 430nm (below the threshold) is detected and marked as the eye protection mode identifier; in step 602, the compression coefficient configuration group of technical Q&A is extracted from the three-dimensional tree (basic value 0.6, maximum 0.8); in step 603, a weight factor 0.75 is generated according to the cognitive load level 0.7 and converted into a scalar value 0.75; in step 604, a negative proportional relationship is established, and the actual compression coefficient 0.6+(0.75×0.08)=0.66 is calculated. After the user switches to the focus mode, in step 601, the wavelength 510nm (above the threshold) is detected and marked as the focus identifier; in step 602, the theoretical expansion coefficient configuration group (basic value 1.2, maximum 2.0) is extracted; in step 603, the load level 0.8 is converted into a scalar value 0.64 (square processing); in step 604, the actual expansion coefficient 1.2+(0.64×0.12)=1.28 is calculated, and a real-time adjustment parameter group is generated to drive content extension.

[0164] To solve the problem of lag in cognitive load matching caused by the lack of a real-time feedback mechanism in the existing content reorganization strategy and further improve the adaptive presentation accuracy in a dynamic interaction scenario, in some embodiments, the reorganization strategy of the information adaptation engine is corrected according to the matching degree between the user's real-time interaction behavior and the content browsing trajectory, and the corrected reorganization strategy reversely adjusts the node density of the conceptual topology expression and the branch depth of the logical reasoning path through the change of the cognitive load level, including:

[0165] 701. Extract the operation event sequence from the user's real-time interaction behavior, parse the path node sequence from the content browsing trajectory, and calculate the matching degree index through the spatial coincidence degree between the event trigger position in the operation event sequence and the path node sequence;

[0166] In step 701, the operation event sequence is a set of timestamps and coordinates of the user's interaction actions such as clicks, swipes, and long presses on the interface. The path node sequence is a set of coordinates of key semantic units in the preset Q&A content in the screen space. The spatial coincidence degree is the proportion of the user's operation position and the coverage area of the preset content node on the two-dimensional plane.

[0167] In the embodiments of the present application, first, the coordinates and time of the user's interaction actions are recorded by the event capture module to generate an operation event sequence with timestamps; second, the preset path node coordinates are extracted from the content rendering engine and arranged in chronological order as a node sequence; third, the operation event coordinates and the node sequence are mapped to the same screen coordinate system, and the proportion of the operation points falling into the node buffer area in each time window is calculated (such as a circular area with a radius of 50px); finally, the sliding average value of the coincidence ratio of consecutive windows is taken to generate a matching degree index in the 0-1 interval.

[0168] 702. Generate a correction coefficient for the recombination strategy according to the fluctuation range of the matching degree index within a preset time window, and establish an inverse correlation gradient between the correction coefficient and the rising and falling trend of the cognitive load level;

[0169] In step 702, the fluctuation range is the difference between the maximum value and the minimum value of the matching degree index within a set time period. The inverse correlation gradient is an adjustment proportional relationship in which the correction coefficient is opposite to the change trend of the cognitive load.

[0170] In the embodiment of the present application, first, set the time window length (such as 10 seconds), and calculate the standard deviation of the matching degree index within the window as the correction coefficient of the recombination strategy; second, when the fluctuation range exceeds a threshold (such as 0.25), generate a correction coefficient according to the change direction of the cognitive load level: if the load level rises, the correction coefficient decreases according to the gradient formula (1 - load change rate); if the load decreases, the correction coefficient increases according to (1 + load change rate), and limit the correction coefficient within a preset interval (such as 0.3 - 1.5).

[0171] 703. Input the correction coefficient into the strategy converter to output a node density adjustment factor and a branch depth adjustment factor;

[0172] In step 703, the strategy converter is a computing device that maps the correction coefficient to content structure adjustment parameters.

[0173] In the embodiment of the present application, first, establish a negative linear mapping rule for the node density adjustment factor in the strategy converter (for every 0.1 decrease in the correction coefficient, the density factor decreases by 0.08); second, establish a positive exponential mapping rule for the branch depth adjustment factor (for every 0.1 increase in the correction coefficient, the depth factor increases by 0.12); finally, write the two types of adjusted factors into the real-time control register through the dual-channel output module.

[0174] 704. Establish a parameter injection channel in the information adaptation engine, and the node density adjustment factor and the branch depth adjustment factor are respectively injected into the hierarchical folding module of the conceptual topology expression and the branch generation module of the logical reasoning path through the parameter injection channel to form a two-way feedback between the recombination strategy and the cognitive load.

[0175] In step 704, the parameter injection channel is a low-latency data path connecting the strategy engine and the content processing module. The hierarchical folding module is a processor device that performs semantic unit merging and deletion. The branch generation module is a controller that manages the extension and contraction of the logical path.

[0176] In the embodiments of the present application, first, an independently addressable injection channel is created in the information adaptation engine, and different memory blocks are allocated for the node density and branch depth adjustment factors. Secondly, the node density adjustment factor is written into the configuration register of the hierarchical folding module in real time through the direct memory access (DMA) technology to control the dynamic adjustment of the folding threshold. At the same time, the branch depth adjustment factor is injected into the priority queue of the branch generation module to adjust the rate and depth of branch extension. Finally, the content structure change event is transmitted back to the cognitive load assessment module through the interrupt feedback mechanism to form a two-way feedback between the reorganization strategy and the cognitive load.

[0177] The following is a specific example:

[0178] Suppose in the Q&A scenario of machine learning model tuning, when the user studies "solutions to convolutional neural network overfitting": In step 701, it is detected that the user repeatedly clicks on the "L2 penalty term" sub-node under the "weight regularization" node, but completely skips the "data augmentation strategy" and "Dropout layer optimization" branches in the preset path. The spatial coincidence degree between the calculation operation event and the path node is only 0.39. In step 702, it is analyzed that the matching degree fluctuates by 0.41 within 10 seconds (standard deviation threshold 0.3), and the cognitive load level increases by 18% due to the user's frequent backward operations, generating a correction coefficient of 0.72 for the reverse association. In step 703, the correction coefficient is mapped to a node density adjustment factor of 0.85 (reducing the density by 15%) and a branch depth adjustment factor of 1.15 (increasing the depth by 15%) through the policy converter. In step 704, the density adjustment factor is injected into the hierarchical folding module to merge "L1 / L2 regularization comparison" and "elastic net implementation" into a composite node; at the same time, the depth adjustment factor is injected into the branch generation module to extend two practical branches of "image rotation parameter optimization" and "adversarial sample generation" under the "data augmentation" node. After adjustment, the matching degree of the user operation path is increased to 0.67, and the cognitive load index is decreased by 12%, forming a two-way feedback between the reorganization strategy and the cognitive load.

[0179] Steps 701 - 704 dynamically adjust the complexity distribution of the content structure by capturing the degree of deviation of the user's behavior from the preset path in real time: compressing redundant technical details and strengthening the visual expression of the core solution when the cognitive load is overloaded, and extending the practical case and tool integration module when the load is relieved; combining the two-way feedback mechanism to achieve the precise resonance between the content presentation strategy and the user's cognitive rhythm, significantly improving the understanding efficiency and practical conversion rate of complex technical solutions.

[0180] Figure 2 The following is a schematic structural diagram of an intelligent Q&A system based on a large model provided by the embodiments of the present application. As Figure 2 shown, the system includes:

[0181] A receiving module 21, configured to, when receiving a question-and-answer request sent by a user, generate a cognitive load level through correlation analysis of the collected user eye movement trajectory data and interaction response latency data, and divide a shallow understanding mode and a deep thinking mode according to the cognitive load level;

[0182] A control module 22, configured to perform two-way linkage control on the generation strategies corresponding to the screen display parameters and the question-and-answer content parameters according to the cognitive load level. When in the shallow understanding mode, activate an eye protection display interface according to the two-way linkage control result; when in the deep thinking mode, activate a focused display interface according to the two-way linkage control result;

[0183] A recombination module 23, configured to, based on the cognitive load level and the optical characteristics of the current display interface, recombine the language structure and knowledge granularity of the question-and-answer content parameters through an information adaptation engine, generate a concept topology expression according to the recombination result under the eye protection display interface, and generate a logical reasoning path according to the recombination result under the focused display interface;

[0184] A correction module 24, configured to correct the recombination strategy of the information adaptation engine according to the matching degree between the user's real-time interaction behavior and the content browsing trajectory, and the corrected recombination strategy reversely adjusts the node density of the concept topology expression and the branch depth of the logical reasoning path through the change of the cognitive load level;

[0185] A generation module 25, configured to synchronously render the recombined question-and-answer content parameters, the adjusted node density or the adjusted branch depth, generate a question-and-answer rendering result, and display the question-and-answer rendering result to the user.

[0186] Figure 2 The described intelligent question-and-answer system based on a large model can execute Figure 1 The described intelligent question-and-answer method based on a large model in the illustrated embodiment, and its implementation principle and technical effects will not be elaborated further. For the intelligent question-and-answer system based on a large model in the above embodiment, the specific manners in which each module and unit perform operations have been described in detail in the embodiment related to the method, and will not be elaborated here.

[0187] In a possible design, Figure 2 The intelligent question-and-answer system based on a large model in the illustrated embodiment can be implemented as a computing device, such as Figure 3 shown, and this computing device can include a storage component 31 and a processing component 32;

[0188] The storage component 31 stores one or more computer instructions, and the one or more computer instructions are called and executed by the processing component 32.

[0189] The processing component 32 is used for the above Figure 1 An intelligent question-answering method based on a large model in the above-mentioned embodiment.

[0190] Among them, the processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components for executing the above method.

[0191] The storage component 31 is configured to store various types of data to support the operation of the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0192] Of course, the computing device may also necessarily include other components, such as input / output interfaces, display components, communication components, etc.

[0193] The input / output interface provides an interface between the processing component and the peripheral interface module, and the above peripheral interface module may be an output device, an input device, etc.

[0194] The communication component is configured to facilitate communication between the computing device and other devices in a wired or wireless manner, etc.

[0195] Among them, the computing device may be a physical device or an elastic computing host provided by a cloud computing platform, etc. At this time, the computing device may refer to a cloud server, and the above processing component, storage component, etc. may be basic server resources leased or purchased from a cloud computing platform.

[0196] The embodiment of the present application also provides a computer storage medium storing a computer program, and when the computer program is executed by a computer, it can implement the above Figure 1 An intelligent question-answering method based on a large model in the above-mentioned embodiment.

[0197] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the above-described systems, devices, and units can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0198] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0199] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0200] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of each embodiment of the present application.

Claims

1. An intelligent question-answering method based on a large model, characterized in that Including: When receiving a Q&A request sent by a user, through the correlation analysis of the collected user eye movement trajectory data and interaction response delay data, to generate a cognitive load level, and divide the shallow understanding mode and the deep thinking mode according to the cognitive load level; According to the cognitive load level, perform two-way linkage control on the generation strategies corresponding to the screen display parameters and the Q&A content parameters. When in the shallow understanding mode, activate the eye protection display interface according to the two-way linkage control result; when in the deep thinking mode, activate the focus display interface according to the two-way linkage control result. The screen display parameters include the blue light suppression intensity and the contrast gain value, and the Q&A content parameters include the information density attenuation gradient and the logical path expansion intensity; Based on the cognitive load level and the optical characteristics of the current display interface, the information adaptation engine reorganizes the language structure and knowledge granularity of the Q&A content parameters. Under the eye protection display interface, generate a concept topology expression according to the reorganization result, and under the focus display interface, generate a logical reasoning path according to the reorganization result. The logical reasoning path is generated by, under the focus display interface, the knowledge granularity control unit of the information adaptation engine performing branch extension processing on semantic units with a marked level lower than the preset level according to the expansion coefficient of the second mapping rule group, and according to the semantic units after the branch extension processing; According to the matching degree between the user's real-time interaction behavior and the content browsing trajectory, correct the reorganization strategy of the information adaptation engine. The corrected reorganization strategy reversely adjusts the node density of the concept topology expression and the branch depth of the logical reasoning path through the change of the cognitive load level; Synchronously render the reorganized Q&A content parameters, the adjusted node density or the adjusted branch depth to generate a Q&A rendering result, and display the Q&A rendering result to the user.

2. The method according to claim 1, characterized in that, The two-way linkage control of the generation strategies corresponding to the screen display parameters and the Q&A content parameters according to the cognitive load level. When in the shallow understanding mode, activate the eye protection display interface according to the two-way linkage control result; When in the deep thinking mode, activate the focus display interface according to the two-way linkage control result, including: Construct the generation strategies of the screen display parameters and the Q&A content parameters; According to the numerical range of the cognitive load level, extract the blue light suppression intensity and the information density attenuation gradient corresponding to the shallow understanding mode, and the contrast gain value and the logical path expansion intensity corresponding to the deep thinking mode from the generation strategies; When in the shallow understanding mode, input the blue light suppression intensity into the screen backlight wavelength controller to generate an eye protection optical signal, and at the same time input the information density attenuation gradient into the compression engine corresponding to the Q&A content parameters to generate a simplified semantic unit sequence, and synchronously load the eye protection optical signal and the simplified semantic unit sequence to activate the eye protection display interface; When in the deep thinking mode, the contrast gain value is input into the screen pixel driving module to generate a focusing optical signal. At the same time, the logical path expansion intensity is input into the inference engine corresponding to the Q&A content parameters to generate a nested logical chain sequence. The focusing optical signal and the nested logical chain sequence are synchronously loaded to activate the focusing display interface.

3. The method according to claim 1, characterized in that Based on the cognitive load level and the optical characteristics of the current display interface, the information adaptation engine reorganizes the language structure and knowledge granularity of the Q&A content parameters. Under the eye protection display interface, a concept topology expression is generated according to the reorganization result. Under the focusing display interface, a logical reasoning path is generated according to the reorganization result, including: The information adaptation engine loads a set of content structure mapping rules bound to the optical characteristics of the current display interface, where the eye protection display interface corresponds to the first mapping rule group in the content structure mapping rule set, and the focusing display interface corresponds to the second mapping rule group in the content structure mapping rule set; According to the currently activated display interface type, the corresponding mapping rule group is extracted from the content structure mapping rule set, and the cognitive load level is quantified into a structure adjustment weight factor by the weight calculation unit of the information adaptation engine, and a proportional relationship is established between the structure adjustment weight factor and the compression coefficient or expansion coefficient in the mapping rule group; Based on the proportional relationship, the core semantic units of the Q&A content parameters are hierarchically marked by the language structure parsing unit of the information adaptation engine. Under the eye protection display interface, semantic units with a hierarchical marking result higher than the preset level are folded according to the compression coefficient of the first mapping rule group, and a concept topology expression is generated according to the folded semantic units; Under the focusing display interface, the knowledge granularity control unit of the information adaptation engine performs branch extension processing on semantic units with a marked level lower than the preset level according to the expansion coefficient of the second mapping rule group, and a logical reasoning path is generated according to the semantic units after the branch extension processing.

4. The method according to claim 1, characterized in that, Synchronously rendering the reorganized Q&A content parameters, the adjusted node density, or the adjusted branch depth to generate a Q&A rendering result, and presenting the Q&A rendering result to the user, including: Synchronously rendering the reorganized Q&A content parameters and the adjusted node density to generate content distribution data with density marks; or synchronously rendering the reorganized Q&A content parameters and the adjusted branch depth to generate content distribution data with depth marks; Select the corresponding density mark or depth mark in the content distribution data according to the current display interface type, calculate the basic spacing of visual elements according to the density mark, or calculate the associated line strength of visual elements according to the depth mark. The basic spacing has an inverse fluctuation relationship with the adjusted node density, and the associated line strength has a positive fluctuation relationship with the adjusted branch depth; Adjust the arrangement distribution of the Q&A content parameters based on the base spacing, or generate logical connection identifiers based on the associated line strength, and form a switchable dynamic visual template according to the adjusted arrangement distribution or the logical connection identifiers; Convert the dynamic visual template into a screen pixel drive signal and a content rendering instruction, input the screen pixel drive signal and the content rendering instruction in the display controller, and then output the Q&A rendering result, and display the Q&A rendering result to the user.

5. The method according to claim 2, wherein When in the shallow understanding mode, input the blue light suppression intensity into the screen backlight wavelength controller to generate an eye protection optical signal, and at the same time input the information density attenuation gradient into the Q&A content parameter compression engine to generate a simplified semantic unit sequence, and synchronously load the eye protection optical signal and the simplified semantic unit sequence to activate the eye protection display interface, including: Convert the blue light suppression intensity into an adjustment pulse for the backlight wavelength, and the duty cycle of the adjustment pulse has a positive linear relationship with the blue light suppression intensity, and send the adjustment pulse to the screen backlight wavelength controller through the pulse width modulation interface; Convert the information density attenuation gradient into a hierarchical compression step size, and the hierarchical compression step size has an exponential decay relationship with the information density attenuation gradient, and input the hierarchical compression step size into the Q&A content parameter compression engine through the step size distributor; In the Q&A content parameter compression engine, perform cross-level folding on the original semantic units according to the hierarchical compression step size, and perform redundant level merging when the semantic association strength between adjacent levels is lower than the folding threshold during the folding process, and generate a simplified semantic unit sequence according to the merging result; Synchronously capture the effective timestamp of the adjustment pulse and the rendering ready signal of the simplified semantic unit sequence through the hardware rendering interface. When the difference between the effective timestamp and the rendering ready signal is less than the synchronization tolerance, load the superimposed output of the eye protection optical signal and the simplified semantic unit sequence, and activate the eye protection display interface based on the superimposed output.

6. The method according to claim 3, wherein According to the currently activated display interface type, extract the corresponding mapping rule group from the content structure mapping rule set, and quantify the cognitive load level into a structure adjustment weight factor through the weight calculation unit of the information adaptation engine, and establish a proportional relationship between the structure adjustment weight factor and the compression coefficient or expansion coefficient in the mapping rule group, including: Read the characteristic value of the optical wavelength of the current display interface through the interface type identifier, and mark it as an eye protection interface identifier when the characteristic value is lower than the blue light suppression threshold, and mark it as a focus interface identifier when it is higher than the contrast enhancement threshold; According to the eye protection interface identifier or the focus interface identifier, respectively extract the corresponding mapping rule group from the content structure mapping rule set, and the mapping rule group stores the association configuration table of the compression coefficient and the expansion coefficient in a three-dimensional tree structure; Quantify the cognitive load level into a structure adjustment weight factor through the weight calculation unit of the information adaptation engine, and input the structure adjustment weight factor into the numerical converter to output the corresponding scalar value; Create a parameter binding area in the associated configuration table, and establish a cross-section linear proportional relationship between the scalar value, the compression coefficient, and the expansion coefficient in the parameter binding area, and generate a real-time adjustment parameter group that can drive the semantic unit level mark based on the linear proportional relationship.

7. The method according to claim 1, wherein Modify the reorganization strategy of the information adaptation engine according to the matching degree between the user's real-time interaction behavior and the content browsing track. The modified reorganization strategy reversely adjusts the node density of the conceptual topology expression and the branch depth of the logical reasoning path through the change of the cognitive load level, including: Extract the operation event sequence from the user's real-time interaction behavior, parse the path node sequence from the content browsing track, and calculate the matching degree index through the spatial coincidence degree between the event trigger position in the operation event sequence and the path node sequence; Generate a correction coefficient for the reorganization strategy according to the fluctuation range of the matching degree index within a preset time window, and establish an inverse correlation gradient between the correction coefficient and the rising and falling trend of the cognitive load level; Input the correction coefficient into the strategy converter, and output the node density adjustment factor and the branch depth adjustment factor; Establish a parameter injection channel in the information adaptation engine, and the node density adjustment factor and the branch depth adjustment factor are respectively injected into the hierarchical folding module of the conceptual topology expression and the branch generation module of the logical reasoning path through the parameter injection channel, forming a two-way feedback between the reorganization strategy and the cognitive load.

8. An intelligent question-answering system based on a large model, which is used to execute the intelligent question-answering method based on a large model according to any one of claims 1-7, and is characterized in that, Including: A receiving module, used for when receiving a question-and-answer request sent by the user, generating a cognitive load level through an association analysis of the collected user eye movement trajectory data and interaction response delay data, and dividing the shallow understanding mode and the deep thinking mode according to the cognitive load level; A control module, used for performing two-way linkage control on the generation strategies corresponding to the screen display parameters and the question-and-answer content parameters according to the cognitive load level. When in the shallow understanding mode, activate the eye protection display interface according to the two-way linkage control result; when in the deep thinking mode, activate the focus display interface according to the two-way linkage control result; A reorganization module, used for based on the cognitive load level and the optical characteristics of the current display interface, reorganize the language structure and knowledge granularity of the question-and-answer content parameters through the information adaptation engine, generate a conceptual topology expression according to the reorganization result under the eye protection display interface, and generate a logical reasoning path according to the reorganization result under the focus display interface; A correction module, used for correcting the reorganization strategy of the information adaptation engine according to the matching degree between the user's real-time interaction behavior and the content browsing track, and the modified reorganization strategy reversely adjusts the node density of the conceptual topology expression and the branch depth of the logical reasoning path through the change of the cognitive load level; A generation module, used for synchronously rendering the reorganized question-and-answer content parameters, the adjusted node density, or the adjusted branch depth to generate a question-and-answer rendering result, and display the question-and-answer rendering result to the user.

9. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement an intelligent question-answering method based on a large model as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that, A computer program is stored, and when the computer program is executed by a computer, it implements an intelligent question-answering method based on a large model as described in any one of claims 1 to 7.

Images