Exhibition dynamic optimization method based on generative AI and visitor emotional feedback

Abstract

This invention belongs to the field of intelligent exhibition optimization technology. It discloses a dynamic exhibition optimization method based on generative AI and visitor emotional feedback. By non-intrusively collecting visitor facial image sequences and spatial behavior trajectories, it refines the analysis of facial micro-movements and path features to construct an emotional response index and spatial attention evaluation system. An emotional behavior consistency index is introduced to achieve a multi-dimensional quantitative evaluation of the overall attractiveness of exhibits. The attributes of the exhibits to be optimized and visitor emotional characteristics are input into the generative AI, which automatically generates targeted optimization schemes including display order, explanatory text, and interactive elements. This invention establishes a complete "collection-analysis-optimization" closed loop, ensuring optimization effectiveness through an incremental threshold mechanism. This invention enables dynamic adjustment and personalized optimization of exhibition content, providing cultural venues with a scientific, precise, and efficient solution for enhancing the exhibition experience.

Description

Technical Field

[0001] This invention relates to the field of intelligent exhibition optimization technology, and more specifically, to a method for dynamic optimization of exhibitions based on generative AI and visitor emotional feedback. Background Technology

[0002] In the current field of museums, art galleries, and cultural exhibitions, the evaluation and optimization of exhibition effectiveness faces multiple technical bottlenecks and methodological deficiencies. Traditional exhibition feedback mainly relies on post-exhibition questionnaires and simple observation records, which are not only untimely but also fail to capture the instantaneous emotional changes and subtle reactions of visitors during their experience. A common dilemma for museum staff is that while they can perceive that some exhibits lack appeal, they struggle to pinpoint the root cause: insufficient content depth, obscure explanatory text, or poor display angle. This subjectivity and lag in evaluation leads to a vague direction for optimization, often resulting in an inefficient cycle of "adjusting based on experience - insignificant results - readjusting again." Existing technologies generally lack an analytical framework that organically combines visitors' emotional experiences with spatial behavior, failing to reveal the underlying reasons behind complex phenomena such as "long dwell time but poor emotional experience" or "strong emotional reaction but quick departure." When exhibition managers face diverse visitor groups of different ages, cultural backgrounds, and education levels, existing static display models are even more inadequate, failing to meet personalized needs. Especially in large-scale special exhibitions, the stark contrast between the large visitor flow and limited feedback channels leaves museums with very little knowledge of the actual effectiveness of the exhibition. Even when adjustments are made, the lack of a scientific verification mechanism makes it impossible to determine whether the improvements are truly effective, let alone quantify the degree of optimization. Although the exhibition field in the digital age has introduced some technological means, it has still failed to fully leverage the potential of artificial intelligence in creative generation and content optimization, resulting in a significant gap between exhibition design and audience expectations, ultimately affecting the effectiveness of cultural dissemination and the quality of the visitor experience.

[0003] In view of this, the present invention proposes an exhibition dynamic optimization method based on generative AI and visitor emotional feedback to solve the above problems. Summary of the Invention

[0004] To overcome the aforementioned shortcomings of existing technologies and to achieve the above objectives, this invention provides the following technical solution: a method for optimizing exhibition dynamics based on generative AI and visitor emotional feedback, comprising:

[0005] Multimodal feedback data of visitors was collected at the exhibition venue during each time period. The multimodal feedback data included facial image sequences and spatial behavior trajectories.

[0006] Based on the temporal changes in facial expression features and the amplitude of micro-movements in the facial image sequence, the emotional response index of each visitor in front of each exhibit is obtained;

[0007] Based on the distribution of dwell time and movement path characteristics in the spatial behavior trajectory, the spatial attention of each visitor to each exhibit is obtained;

[0008] Based on the correlation between the emotional response index and spatial attention, a comprehensive attractiveness score is obtained for each exhibit; exhibits that need optimization are identified based on the comprehensive attractiveness score.

[0009] The display attributes of the exhibits to be optimized and the emotional response characteristics of the corresponding visitor groups are input into the generative AI model to generate targeted exhibition optimization solutions; the exhibition content is then adjusted based on the exhibition optimization solutions.

[0010] The technical effects and advantages of the exhibition dynamics optimization method based on generative AI and visitor emotional feedback of this invention are as follows:

[0011] This invention, through the seamless collection and analysis of visitors' genuine emotional fluctuations and behavioral patterns, allows exhibition managers to see beyond the surface and gain insight into the essence of the visitor experience. This deep understanding translates into precise optimization, enabling exhibitions to perceive, learn, and evolve. In practical applications, previously overlooked subtle experiential pain points are identified and intelligently resolved. Exhibitions are no longer static presentations but become dynamic systems capable of self-improvement, constantly adapting to the needs and expectations of different audience groups. For cultural institutions, this means moving beyond guesswork-based, blind adjustments and ushering in efficient management through precise policy implementation; for curators, it provides creative support and professional guidance; for visitors, it creates a more engaging and resonant visitor journey, elevating the cultural experience to a level of deep immersion and emotional connection. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the exhibition dynamics optimization method based on generative AI and visitor emotional feedback according to the present invention. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] This application provides a method for dynamic optimization of exhibitions based on generative AI and visitor emotional feedback. The implementers of this method include, but are not limited to, museum exhibition management systems, intelligent display optimization platforms, visitor experience enhancement systems, and smart cultural venue management systems.

[0015] Please see Figure 1 In this embodiment of the invention, the specific implementation process of the exhibition dynamic optimization method based on generative AI and visitor emotional feedback includes:

[0016] Multimodal feedback data of visitors was collected at each time point within the exhibition venue. This data included facial image sequences and spatial behavior trajectories. Facial image sequences were captured using a high-definition camera system within the exhibition hall, recording changes in visitors' facial expressions during their visit, including micro-expression features such as eyebrow and eye movements, mouth changes, and overall facial expression transitions. Spatial behavior trajectories were collected using a positioning sensor network or computer vision positioning system, accurately recording visitors' movement paths, stopping positions, and walking speeds within the exhibition space. This multimodal data provides rich raw material for subsequent analysis, ensuring the accuracy and comprehensiveness of sentiment analysis and spatial behavior research.

[0017] It is important to note that this invention employs a high-definition camera system to collect facial image sequences and a positioning sensor network or computer vision positioning system to collect spatial behavioral trajectories within the exhibition hall. The deployment of these technologies is a necessary measure to maintain public safety. The exhibition hall features prominent information collection notices, clearly informing visitors of the existence, purpose, and scope of data collection. All collected personal images and location information undergo real-time de-identification processing and are strictly limited to public safety maintenance and anonymized exhibition experience analysis, and will not be used for identity verification or other commercial purposes. This invention complies with relevant laws and regulations regarding personal information protection, employing strict data encryption, access control, and storage period management measures to ensure visitor privacy and security. Visitors entering the exhibition hall are deemed to have acknowledged and consented to the collection of relevant information. The exhibition hall also provides an opt-out mechanism, fully respecting individual wishes and protecting legitimate rights.

[0018] Based on the temporal changes and micro-movement amplitudes of facial expression features in facial image sequences, an emotional response index is obtained for each visitor in front of each exhibit. The emotional response index is a key indicator measuring the intensity of visitors' emotional responses to exhibits. It is calculated by analyzing the frequency and amplitude of changes in facial micro-movements, as well as the distribution characteristics of emotional types. The analysis process first performs facial keypoint detection on the facial image sequences, then calculates the dynamic change features between keypoints, extracts the intensity of emotional fluctuations and the positive emotional rate, and finally derives a comprehensive emotional response index, providing a quantitative basis for the emotional dimension of exhibit evaluation.

[0019] Based on the distribution of dwell time and movement path characteristics in spatial behavior trajectories, the spatial attention level of each visitor to each exhibit is obtained. Spatial attention reflects the spatial behavioral performance of visitors' attention to and interest in exhibits, and is quantified by analyzing dwell time and movement characteristics. The analysis process first determines the effective viewing area of ​​each exhibit, then calculates the cumulative dwell time and movement path characteristics of visitors within that area, and comprehensively evaluates the spatial attention level, providing quantitative indicators of behavioral dimensions for exhibit evaluation.

[0020] Based on the correlation between the emotional response index and spatial attention, a comprehensive attractiveness score is obtained for each exhibit. The comprehensive attractiveness score is the core indicator for evaluating exhibit effectiveness, integrating information from both emotional response and spatial behavior dimensions. The evaluation process first calculates the mean emotional score and mean attention for each exhibit, then analyzes the consistency between emotion and behavior, and finally generates a comprehensive attractiveness score, providing a quantitative basis for identifying exhibits that need optimization.

[0021] The system identifies exhibits requiring optimization based on a comprehensive attractiveness score. These exhibits typically have attractiveness scores significantly below average or exhibits with poor consistency between emotion and behavior. The identification process uses threshold screening or ranking methods to pinpoint the exhibits most in need of optimization, ensuring the effective use of optimization resources.

[0022] The display attributes of exhibits to be optimized, along with the emotional response characteristics of the corresponding visitor groups, are input into a generative AI model to generate targeted exhibition optimization solutions. Optimization solution generation is the core of this method. The generative AI model integrates exhibit characteristics with visitor feedback to create personalized optimization suggestions. The generation process first constructs structured input prompts, then processes them through a pre-trained generative AI model, outputting multi-dimensional optimization suggestions to provide professional guidance for exhibition adjustments.

[0023] The exhibition content is adjusted based on the exhibition optimization plan. Exhibition adjustment is the final step in the optimization loop, enhancing the exhibition's effectiveness by implementing AI-generated optimization suggestions. The adjustment process may include rearranging the display order, revising explanatory texts, adding interactive elements, and other aspects, ensuring the optimization suggestions are effectively implemented to provide visitors with a better experience.

[0024] In this embodiment of the invention, the detailed implementation steps for obtaining the emotional response index of each visitor in front of each exhibit include:

[0025] Facial keypoint detection is performed on a sequence of facial images to obtain the coordinates of a predetermined number of facial feature points in each frame. Facial keypoint detection is fundamental to expression analysis, quantifying expression changes by accurately locating key facial features. The detection process employs deep learning models such as an improved multi-task convolutional neural network (MTCNN) or a keypoint regression network to extract the precise coordinates of 68 or more facial keypoints. Keypoints include expression-sensitive areas such as the corners of the eyes, eyebrows, and corners of the mouth, providing foundational data for subsequent dynamic analysis. The detection algorithm is optimized for exhibition environments, adapting to different lighting and angle conditions to ensure stability and accuracy. The coordinate data of facial keypoints undergoes standardized processing to eliminate the influence of individual facial size differences, facilitating subsequent calculations and comparisons.

[0026] The displacement of the same facial feature point coordinates in adjacent frames is calculated, and the weighted sum of the displacements of all facial feature points is used as the expression dynamic value for each frame. The expression dynamic value reflects the instantaneous intensity of facial expression changes and is the foundation of emotion fluctuation analysis. The calculation process first aligns the facial regions of adjacent frames to eliminate interference from head movement; then, the Euclidean distance between corresponding feature points is calculated to obtain the displacement; finally, a weighted sum is performed based on the emotional expression importance of different facial regions. The facial region is divided into functional areas such as the eyes, mouth, and eyebrows, and different weight coefficients are assigned to highlight the contribution of important regions. The expression dynamic value intuitively reflects the instantaneous intensity of expression changes, providing time-series data for calculating the intensity of emotion fluctuations.

[0027] The standard deviation of facial expression dynamics values ​​for each visitor across all frames during their viewing period at each exhibit was calculated as the emotional fluctuation intensity. Emotional fluctuation intensity measures the magnitude of change in emotional response, reflecting the degree of emotional fluctuation during the viewing process. The calculation formula is as follows:

[0028] ;

[0029] in, Intensity of emotional fluctuations For the first The facial expression dynamics values ​​of the frame image. The average of the facial expression dynamics values ​​across all frames. The number of frames.

[0030] High emotional fluctuation intensity indicates dramatic changes in visitors' emotions, usually meaning that the exhibition content evoked strong emotional resonance; while low fluctuation intensity may reflect a lack of emotional appeal in the content. Emotional fluctuation intensity provides a quantitative indicator of the changing dimensions for calculating the emotional response index.

[0031] The basic emotion category is extracted from each frame of a facial image sequence, and the percentage of frames with positive emotion categories is used as the emotion positivity rate. The emotion positivity rate reflects the proportion of positive emotional experiences in the overall viewing process and is an important indicator for assessing emotion quality. The extraction process uses emotion recognition models such as Convolutional Neural Networks (CNN) combined with Long Short-Term Memory Networks (LSTM) to classify each frame into seven basic emotions (happiness, surprise, fear, disgust, anger, sadness, and neutral). Emotion categories are divided into three main categories: positive (happiness, surprise), neutral, and negative (fear, disgust, anger, sadness). The ratio of positive emotion frames to the total number of frames is calculated to obtain the emotion positivity rate. The emotion positivity rate provides a quantitative indicator of the emotion quality dimension in the calculation of the emotion response index.

[0032] The weighted product of the intensity of emotional fluctuation and the positive emotional rate is used as the emotional response index for the corresponding exhibit. The emotional response index integrates both the intensity and quality of emotional changes, comprehensively reflecting the emotional experience of the visitors. The calculation formula is as follows:

[0033] ;

[0034] in, Emotional Response Index Intensity of emotional fluctuations For positive emotional rate, This is the weighting coefficient (usually set to 0.5-0.7).

[0035] The formula design takes into account both fluctuation intensity and positive mass, with weighting coefficients... This can be dynamically adjusted based on the type of exhibition; art exhibitions may place greater emphasis on emotional intensity (higher emotional impact). (Value), while science exhibitions may focus more on positive experiences (lower). (Value). The Emotional Response Index, as a comprehensive indicator for measuring tourists' emotional responses, provides a key basis for subsequent evaluation of exhibits.

[0036] In this embodiment of the invention, the detailed implementation steps for obtaining each visitor's spatial attention to each exhibit include:

[0037] Based on location sampling points in the spatial behavior trajectory, the cumulative dwell time of each visitor within the effective viewing area of ​​each exhibit is calculated. Cumulative dwell time is the most direct manifestation of spatial attention, reflecting the continuity of visitors' interest in the exhibits. The calculation process first checks the continuity of location sampling points to identify stopping points and movement segments; then, it counts the duration of stopping points within the effective viewing area of ​​each exhibit, summing them up to obtain the total dwell time. Considering individual differences among visitors, relative dwell time (the proportion of total visit time) is used for standardization to eliminate the influence of different total visit times. Cumulative dwell time intuitively reflects the time visitors invest in the exhibits and is a crucial foundational variable for calculating spatial attention.

[0038] The path coverage rate is calculated as the ratio of the length of each visitor's movement path within the effective viewing area of ​​each exhibit to the length of the corresponding area's diagonal. Path coverage rate reflects the completeness of a visitor's exploration of the exhibit and measures the comprehensiveness of the viewing space. The calculation process first smooths the original location sampling points to eliminate positioning jitter; then, it calculates the actual movement path length within the effective viewing area; finally, it is calculated as a ratio to the area's diagonal length to obtain a standardized coverage rate index. The formula for calculating path coverage rate is:

[0039] ;

[0040] in, For path coverage, This is the actual movement path length. This is the length of the diagonal of the viewing area.

[0041] High path coverage indicates that visitors have fully explored all angles and parts of the exhibit, usually reflecting a high level of interest; while low coverage may mean that visitors quickly skipped over or only focused on localized content. Path coverage provides a quantitative indicator of spatial comprehensiveness in spatial attention calculations.

[0042] The product of the cumulative dwell time (after standardization) and the path coverage rate is used as the spatial attention level of each visitor to the corresponding exhibit. Spatial attention level comprehensively considers both time investment and spatial exploration, fully reflecting the spatial behavior characteristics of visitors. Standardization is performed using the Min-Max normalization method, mapping the original dwell time to the [0,1] interval to obtain the standardized dwell time. .

[0043] The final formula for calculating spatial attention is:

[0044] ;

[0045] in, For spatial attention, This represents path coverage.

[0046] The spatial attention value range is [0,1]. The higher the value, the higher the attention visitors pay to the exhibits, providing a quantitative basis for the behavioral dimension of exhibit attractiveness assessment.

[0047] In this embodiment of the invention, the detailed implementation steps for obtaining the overall attractiveness score of each exhibit include:

[0048] The mean of the emotional response index for all visitors corresponding to the same exhibit is calculated as the mean emotional score for that exhibit. The mean emotional score reflects the overall level of emotional response elicited by the exhibit and is a measure of the central tendency of the group's emotional experience. A weighted average method is used in the calculation process, taking into account the influence of sample size and observation duration to ensure the representativeness and reliability of the scores. Outliers are filtered using box plots to remove the influence of extreme emotional responses, ensuring the stability of the mean. The mean emotional score intuitively reflects the level of emotional appeal of the exhibit, providing basic data for the emotional dimension of the comprehensive score.

[0049] The mean spatial attention of all visitors for the same exhibit is calculated as the mean attention for that exhibit. The mean attention quantifies the exhibit's overall ability to attract visitors to stay and explore, serving as a measure of the central tendency of group spatial behavior. The calculation method is similar to that of the mean sentiment score, employing a weighted average and handling outliers to ensure data quality. The mean attention intuitively reflects the behavioral attractiveness level of an exhibit, providing fundamental data for the behavioral dimension of the comprehensive score.

[0050] The correlation coefficient between the emotional response index and spatial attention of each visitor for the same exhibit was calculated, and the mean of the correlation coefficients for all visitors was taken as the emotional-behavioral consistency of the exhibit. Emotional-behavioral consistency measures the degree of coordination between emotional experience and spatial behavior, reflecting the inherent consistency of the exhibit's attractiveness. The Pearson correlation coefficient was used for calculation, and the formula is:

[0051] ;

[0052] in, The correlation coefficient is... For the first The emotional response index of each tourist. For the first Spatial attention of tourists and These represent the average emotional response index and the average spatial attention of all tourists, respectively.

[0053] High emotional-behavioral consistency indicates that visitors with strong emotional responses typically spend more time exploring the space and engage fully, reflecting a high degree of unity between the exhibit's emotional and behavioral appeal. Conversely, low consistency may indicate a weakness in some aspect of the exhibit, such as visual appeal but dull content, or profound content but poor presentation. Emotional-behavioral consistency provides a quantitative indicator of consistency in the overall score.

[0054] The normalized weighted sum of the mean emotional score, the mean attention score, and the consistency of emotional behavior is used as the comprehensive attractiveness score for the corresponding exhibit. The comprehensive attractiveness score integrates three dimensions: emotional response, spatial behavior, and consistency, comprehensively reflecting the overall effect of the exhibit. The calculation process first normalizes the three indicators to ensure uniformity of measurement; then, weights are assigned according to the exhibition objectives, and a weighted sum is performed. Normalization uses the Z-score standardization method to convert the original values ​​into a standard normal distribution. The formula for calculating the comprehensive attractiveness score is:

[0055] ;

[0056] in, To score overall attractiveness, , and These are the normalized mean sentiment score, mean attention level, and consistency between sentiment and behavior, respectively. , and These are the weighting coefficients, and .

[0057] The weighting can be dynamically adjusted based on the type of exhibition and the evaluation objectives. For example, art exhibitions may place greater emphasis on emotional response (higher weighting). Interactive exhibitions may place more emphasis on behavioral participation (higher level). Education fairs, on the other hand, may place more emphasis on consistency (higher level). The overall attractiveness score provides a quantitative basis for identifying exhibits that need optimization, and exhibits with low scores are usually given priority for optimization.

[0058] In this embodiment of the invention, the detailed implementation steps for generating a targeted exhibition optimization scheme include:

[0059] The process involves constructing descriptive text for the exhibits to be optimized, including exhibit type, current layout, and a summary of explanatory content. This descriptive text forms the foundation for the generative AI model to understand the current state of the exhibits, providing a starting point and constraints for optimization. A structured template is used to ensure the completeness and consistency of the information. The exhibit type section details the exhibit's classification, material, age, and cultural background; the current layout section includes spatial arrangement information such as display height, angle, lighting conditions, and surrounding environment; and the summary of explanatory content provides key information points, narrative style, and presentation characteristics from existing explanatory texts. The descriptive text uses objective and neutral language to avoid subjective evaluation, providing accurate foundational data for the AI ​​model.

[0060] Emotional response characteristics of the visitor group corresponding to the exhibits to be optimized were extracted. These characteristics included the peak periods of negative emotions and the display segments corresponding to the inflection points of emotional decline. Emotional response characteristics are key to identifying problem areas of exhibits, revealing critical nodes and turning points in the visitor's emotional experience. The extraction process first performed time-series analysis on the emotional data to identify significant characteristic points of emotional changes; then, these characteristic points were precisely mapped to display segments to establish an emotion-display relationship. The peak periods of negative emotions refer to the time periods when the emotional index reaches a local minimum, typically indicating that the content evoked confusion, dissatisfaction, or boredom; the inflection points of emotional decline refer to the turning point where the emotional curve changes from rising to falling, typically indicating the critical point where interest begins to wane. Accurate location of these characteristic points provides precise problem localization for optimization, indicating the segments that need focused improvement.

[0061] The text describing the display attributes and the emotional response features are combined according to a preset prompt template to form the input prompts for the model. These input prompts are crucial for guiding generative AI to generate high-quality optimization solutions; well-structured prompts significantly improve the relevance and usability of the generated content. The combination process employs a task-oriented prompt engineering approach, clearly organizing the problem description, data input, and expected output. The preset prompt template typically includes core components such as task description, exhibit information, visitor feedback, problem identification, and optimization requirements, presented in natural language. The template design follows a "problem-evidence-solution" logical structure, ensuring the AI ​​model accurately understands the task requirements and contextual information. The final generated prompts provide the AI ​​model with a complete problem background and task guidance, forming the foundation for generating high-quality solutions.

[0062] The model inputs prompts into a pre-trained generative AI model to generate exhibition optimization solutions, including suggestions for adjusting the display order, rewriting explanatory text, and adding interactive elements. These optimization solutions are specific improvement suggestions generated by the AI ​​model based on data analysis and professional knowledge, providing practical guidance for exhibition adjustments. The generation process employs advanced generative AI technologies, such as large-scale language models (GPT series or similar), which are specifically fine-tuned using knowledge from fields like museology, exhibition design, and visitor psychology to ensure the professionalism and feasibility of the output content. The optimization solutions encompass three core dimensions: suggestions for adjusting the display order focus on optimizing content logic and visitor flow, such as prioritizing key content or aggregating related content; suggestions for rewriting explanatory text focus on improving language expression, such as simplifying technical terminology or changing narrative style; and suggestions for adding interactive elements focus on enhancing engagement, such as adding multi-sensory experiences or gamification elements. The AI ​​model generates highly targeted and innovative optimization suggestions by deeply learning the correlation between visitor emotional feedback and exhibition characteristics, providing professional guidance for exhibition adjustments.

[0063] In this embodiment of the invention, the weighted sum of the displacements of all facial feature points is used as the expression dynamic value for each frame of the image, including:

[0064] Facial feature points are categorized based on their corresponding facial regions, including the eye, mouth, and eyebrow regions. This functional area classification is fundamental to refined expression analysis, improving the accuracy of interpreting facial features. The process begins by establishing a standard facial coordinate system, using the midpoint of the line connecting the eyes as the origin, and creating an orthogonal coordinate system. Then, based on anatomical features, facial feature points are mapped to their corresponding functional areas. The eye region typically includes points on the upper and lower eyelids, the corners of the eyes, and the periorbital area, totaling approximately 20 feature points. The mouth region includes points on the lip contour, the corners of the lips, and the periorbital area, totaling approximately 18 feature points. The eyebrow region includes points on the eyebrow contour and the area between the eyebrows, totaling approximately 10 feature points. This functional area-based classification method aligns with research in the psychology of facial expression, where different regions have different functions and importance in emotional expression, providing a theoretical basis for subsequent weighted calculations.

[0065] Different emotional weight coefficients are assigned to feature points in different facial regions, with the emotional weight coefficients for the eye and mouth regions being greater than that for the eyebrow region. These emotional weight coefficients reflect the relative importance of each facial region in emotional expression and are determined based on research in facial expression psychology. The weighting is typically set at 0.4 for the eye region, 0.4 for the mouth region, and 0.2 for the eyebrow region, reflecting the dominant role of the eyes and mouth in emotional expression. Subtle changes in the eye region, especially the muscles around the eyes, can convey rich emotional details, including surprise, fear, and joy; the mouth region primarily expresses basic emotions such as happiness, disgust, and sadness; while the eyebrow region is relatively important in expressing anger and anxiety, its overall contribution to facial expression is slightly lower. This weighting allocation based on psychological research ensures the scientific rigor and accuracy of facial expression dynamic value calculation, enabling more precise capture of subtle emotional changes.

[0066] The sum of the products of the displacement of each facial feature point and its corresponding emotion weight coefficient is used as the expression dynamic value of the corresponding frame image. The expression dynamic value comprehensively considers the movement of feature points in each region, and the weighting reflects the importance of emotional expression in different regions. The calculation formula is:

[0067] ;

[0068] in, For facial expression dynamic values, For the region The emotional weighting coefficient, For the region The Middle Displacement of each feature point This represents the total number of feature points.

[0069] Displacement is calculated using Euclidean distance, measuring the distance a feature point moves between adjacent frames. To eliminate the influence of face size and camera distance, the displacement is typically normalized by dividing by a facial reference scale (such as the distance between the eyes). Facial dynamics values, as an instantaneous quantitative indicator of emotional changes, provide fundamental data for calculating the intensity of emotional fluctuations, reflecting the real-time state and intensity of facial expression changes.

[0070] In this embodiment of the invention, the method further includes:

[0071] During the validation period following the application of the exhibition optimization scheme, multimodal feedback data from visitors was re-collected, and the overall attractiveness score of the exhibits to be optimized was calculated. Optimization validation is a crucial step in ensuring the effectiveness of the optimization, and its success is verified through a scientific process of experimentation, measurement, and evaluation. The validation process first determines an appropriate validation period, typically 1-2 weeks after optimization implementation, to ensure a sufficient sample size. Then, the entire process of multimodal data collection and overall score calculation is repeated to ensure consistency in evaluation criteria. Finally, a new overall attractiveness score for the validation period is obtained, providing objective data for effectiveness evaluation. Validation sampling should carefully control time conditions (e.g., weekdays / weekends, mornings / afternoons) to be consistent with the initial sampling, reducing interference from non-optimization factors and improving the reliability of the validation results.

[0072] The difference between the overall attractiveness score during the verification period and the overall attractiveness score before optimization is calculated as the increment of the optimization effect. The increment of the optimization effect directly reflects the actual improvement brought about by the optimization measures and is a core indicator for evaluating the success of the optimization. The calculation uses a simple difference method, subtracting the original score from the new score to obtain the increment value. To enhance comparability, the increment is usually standardized as a percentage, and the calculation formula is as follows:

[0073] ;

[0074] in, To optimize the incremental effect (percentage). To verify the scoring of the time period, The score before optimization.

[0075] A positive increment value indicates that the optimization is effective, and the larger the value, the more significant the effect. A negative increment value indicates that the optimization may have had a counterproductive effect, requiring re-analysis and adjustment. Increment evaluation provides an objective basis for optimization loop closure and guides subsequent optimization decisions.

[0076] When the incremental improvement is less than a preset threshold, the incremental improvement is associated with and stored as the corresponding exhibition optimization plan, triggering the generative AI model to regenerate the exhibition optimization plan. The incremental threshold mechanism is a crucial self-correcting step in the optimization process, ensuring that optimization measures achieve the expected results. The threshold is typically set at 10-15%, dynamically adjusted based on the exhibition type and optimization objectives. The associated storage mechanism pairs and saves historical optimization plans with effect data, forming an optimization knowledge base that provides learning materials for the AI ​​model, improving the quality of future optimization suggestions. The regeneration process incorporates information about poor performance from previous optimizations into prompts, guiding the AI ​​model to avoid similar directions and explore new optimization paths. This effect-evaluation-based optimization iteration mechanism forms a complete closed-loop feedback system, ensuring the continuous effectiveness of the optimization process and ultimately achieving the expected improvement goals.

[0077] In this embodiment of the invention, the method for determining the effective viewing area is as follows:

[0078] A circular area is defined with the display center point of each exhibit as the center and a preset viewing radius. The display center point is the core location of the exhibit, usually the geometric center or the recommended position from the main viewing angle. The viewing radius is a reasonable viewing distance determined based on exhibition studies and is dynamically set according to the type, size, and display method of the exhibit. Generally speaking, the viewing radius of two-dimensional exhibits (such as paintings and photographs) is approximately 1.5-2 times the diagonal length of the exhibit; the viewing radius of three-dimensional exhibits (such as sculptures and installations) is approximately 2-3 times the maximum size of the exhibit. Special exhibits, such as large cultural relics or small and delicate items, may require a customized viewing radius. The setting of the circular area is based on research in visual psychology, reflecting the natural distribution characteristics of human visual attention, and providing a theoretically supported spatial definition for subsequent analysis.

[0079] The effective viewing area for each exhibit is obtained by intersecting the circular area with the accessible area of ​​the exhibition space. The effective viewing area represents the actual spatial range within which exhibits can be viewed, taking into account physical constraints. The accessible area is the space within the exhibition hall where visitors can move freely, excluding physical obstacles such as display cases, partitions, and other exhibits. The intersection calculation employs computational geometry methods, performing Boolean operations between the theoretical circular area and the accessible polygonal area to derive the actual usable viewing space. This process eliminates inaccessible parts of the theoretical area, such as spaces behind walls or areas occupied by other exhibits, ensuring the practicality and accuracy of the spatial analysis. The effective viewing area serves as the foundational area division for spatial behavior analysis, providing a clear spatial definition for calculating cumulative dwell time and path coverage.

[0080] This invention achieves dynamic optimization and personalized adjustment of exhibitions by collecting multimodal feedback data, analyzing emotional responses and spatial behavior characteristics, evaluating the attractiveness of exhibits, and utilizing generative AI to optimize exhibition content. The closed-loop optimization method of this invention can accurately capture visitors' emotional experiences, effectively improve exhibition results, and provide a systematic solution for optimizing the display of cultural and museum venues.

[0081] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0082] It should be noted that all formulas in this manual are calculated by removing dimensions and taking their numerical values. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0083] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for optimizing exhibition dynamics based on generative AI and visitor emotional feedback, characterized in that, include: Collect multimodal feedback data of visitors in the exhibition venue during each time period. The multimodal feedback data includes facial image sequences and spatial behavior trajectories. Based on the temporal changes in facial expression features and the amplitude of micro-movements in the facial image sequence, the emotional response index of each visitor in front of each exhibit is obtained; Based on the distribution of dwell time and movement path characteristics in the spatial behavior trajectory, the spatial attention of each visitor to each exhibit is obtained; Based on the correlation and matching degree between the emotional response index and the spatial attention, a comprehensive attractiveness score for each exhibit is obtained; Based on the comprehensive attractiveness score, identify exhibits that need optimization; The display attributes of the exhibits to be optimized and the emotional response characteristics of the corresponding visitor groups are input into the generative AI model to generate targeted exhibition optimization solutions; the exhibition content is then adjusted based on the exhibition optimization solutions.

2. The exhibition dynamic optimization method based on generative AI and visitor emotional feedback as described in claim 1, characterized in that, The acquisition of each visitor's emotional response index in front of each exhibit includes: Facial key point detection is performed on facial image sequences to obtain the coordinates of a preset number of facial feature points in each frame; Calculate the displacement of the same facial feature point coordinates in adjacent frames, and use the weighted sum of the displacements of all facial feature points as the expression dynamic value of each frame. The standard deviation of the facial expression dynamics values ​​of all frames of images for each visitor during the viewing period of each exhibit was calculated as the intensity of emotional fluctuations. The basic emotion category of each frame is extracted from the facial image sequence, and the proportion of frames with positive emotion category is used as the emotion positivity rate. The weighted product of the intensity of the emotional fluctuation and the positive rate of emotion is used as the emotional response index of the corresponding tourist in front of the corresponding exhibit.

3. The exhibition dynamic optimization method based on generative AI and visitor emotional feedback according to claim 1, characterized in that, The acquisition of each visitor's spatial attention to each exhibit includes: Based on the location sampling points in the spatial behavior trajectory, calculate the cumulative stay time of each visitor in the effective viewing area of ​​each exhibit; Calculate the ratio of the length of each visitor's movement path within the effective viewing area of ​​each exhibit to the length of the corresponding diagonal of the area, and use this as the path coverage rate; The product of the cumulative dwell time (after standardization) and the path coverage rate is used as the spatial attention level of the corresponding exhibit for each visitor.

4. The exhibition dynamic optimization method based on generative AI and visitor emotional feedback according to claim 1, characterized in that, The process of obtaining the overall attractiveness score for each exhibit includes: The mean of the emotional response index of all visitors corresponding to the same exhibit is calculated as the mean emotional score of that exhibit. The average spatial attention of all visitors to the same exhibit is calculated as the average attention of that exhibit. Calculate the correlation coefficient between the emotional response index and spatial attention of each visitor for the same exhibit, and take the mean of the correlation coefficients of all visitors as the emotional behavior consistency of the exhibit. The normalized weighted sum of the mean emotional score, the mean attention level, and the consistency of emotional behavior is used as the comprehensive attractiveness score of the corresponding exhibit.

5. The exhibition dynamic optimization method based on generative AI and visitor emotional feedback according to claim 1, characterized in that, The generation of targeted exhibition optimization solutions includes: Construct a description text for the display attributes of the exhibits to be optimized. The display attributes include exhibit type, current layout, and summary of explanatory content. Extract the emotional response characteristics of the visitor group corresponding to the exhibits to be optimized. The emotional response characteristics include the display segments corresponding to the peak period of negative emotions and the inflection point of emotional decline. The display attribute description text and the emotional response features are combined according to a preset prompt template to form the model input prompt words; The input prompts are fed into a pre-trained generative AI model to obtain an exhibition optimization scheme that includes suggestions for adjusting the display order, rewriting explanatory text, and adding interactive elements.

6. The exhibition dynamic optimization method based on generative AI and visitor emotional feedback according to claim 2, characterized in that, The weighted sum of the displacements of all facial feature points is used as the expression dynamic value for each frame of the image, including: Feature points are categorized according to the facial regions to which they belong, including the eye region, mouth region, and eyebrow region. Different emotional weight coefficients were assigned to feature points in different facial regions, with the emotional weight coefficients for the eye and mouth regions being greater than those for the eyebrow region. The sum of the products of the displacement of each facial feature point and the corresponding emotion weight coefficient is used as the expression dynamic value of the corresponding frame image.

7. The exhibition dynamic optimization method based on generative AI and visitor emotional feedback according to claim 1, characterized in that, The method further includes: During the verification period after applying the exhibition optimization scheme, multimodal feedback data of visitors were re-collected and the comprehensive attractiveness score of the exhibits to be optimized during the verification period was calculated. The difference between the overall attractiveness score during the verification period and the overall attractiveness score before optimization is calculated as the increment of the optimization effect; When the incremental improvement effect is less than the preset incremental threshold, the incremental improvement effect is associated with the corresponding exhibition optimization scheme and stored, and the generative AI model is triggered to regenerate the exhibition optimization scheme.

8. The exhibition dynamic optimization method based on generative AI and visitor emotional feedback according to claim 3, characterized in that, The method for determining the effective viewing area is as follows: A circular area is defined with the display center point of each exhibit as the center and a preset viewing radius. The effective viewing area for the corresponding exhibit is obtained by intersecting the circular area with the passable area of ​​the exhibition venue.

Images