Digital exhibition hall content optimization method and platform based on multi-dimensional interaction data analysis

Abstract

The invention relates to the technical field of digital exhibition halls, in particular to a digital exhibition hall content optimization method and platform based on multi-dimensional interaction data analysis, and the method comprises the steps: collecting audience interaction data, content data and environment data from all systems of a digital exhibition hall, cleaning and verifying the collected data, and storing the data in a database; the method comprises the following steps: performing standardized conversion on verified data, ensuring the accuracy and completeness of acquired multi-dimensional data by introducing a data cleaning and verification mechanism, adopting a deep learning architecture of Transform to improve the predictive ability of a model, and meanwhile, increasing the diversity and quantity of training data to improve the generalization ability of the model; a streaming data processing technology is adopted to realize real-time analysis and processing of data, and meanwhile, a configurable exhibition hall content optimization model and an optimization strategy are constructed, so that rapid adjustment can be performed according to actual requirements, and the optimization effect of the digital exhibition hall content is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of digital exhibition halls, and in particular relates to a digital exhibition hall content optimization method and platform based on multi-dimensional interactive data analysis. Background Art

[0002] Digital exhibition hall content optimization is to improve the display effect, audience experience and interactivity of the digital exhibition hall through a series of strategies and technical means. The optimized content can convey information more clearly and accurately, making it easier for the audience to understand and accept.

[0003] For example, the patent application number is 202311259824.6, which discloses a method and device for optimizing the content of a digital exhibition hall based on multi-dimensional interactive data analysis, including obtaining first information and second information, the first information including digital exhibition hall content data; performing data integration processing based on the first information and the second information, and performing data association on the integrated data to obtain third information; performing text and image information extraction processing based on the third information to obtain fourth information; performing sentiment analysis processing based on the fourth information to obtain fifth information; constructing a model based on the second information and a preset deep learning mathematical model to obtain sixth information; constructing a content optimization mathematical model based on the sixth information, and using the fifth information as the input value of the content optimization mathematical model to obtain seventh information. This patent optimizes digital display content through technologies such as sentiment analysis, topic modeling, and deep learning, thereby improving audience satisfaction.

[0004] However, since the method of optimizing the content of digital exhibition halls is to improve the content of digital exhibition halls through data analysis and model prediction to enhance the satisfaction of visitors, and the amount of digital exhibition hall content data and audience interaction data is relatively large, the probability of errors in the first information and the second information obtained increases. The errors will be amplified in subsequent steps such as data integration, text and image information extraction, and sentiment analysis, and ultimately cause the analysis results to deviate from the actual situation, affecting the effect of optimizing the content of digital exhibition halls. Therefore, we need to propose a digital exhibition hall content optimization method and platform based on multi-dimensional interactive data analysis to solve the above-mentioned problems, so that it can ensure the accuracy and completeness of data collection by introducing data cleaning and verification mechanisms, thereby improving the optimization effect of digital exhibition hall content. Summary of the invention

[0005] In view of the above problems, the present invention provides a method for optimizing the content of a digital exhibition hall based on multi-dimensional interactive data analysis, comprising the following steps: S1. Collect audience interaction data, content data, and environmental data from various systems in the digital exhibition hall, and clean and verify the collected data; S2, standardize and transform the verified data to obtain a multi-dimensional interaction data set; S3, using the Transformer architecture to build a deep learning model for analyzing and predicting multi-dimensional interactive data sets; The formula for establishing the deep learning model is as follows: ,in, , are query, key, and value matrices respectively, is the weight matrix of the first head, is the weight matrix of the h-th head, is the dimension of the u-th key vector, h is the number of heads, u is the sequence number of the head, for The calculation result of the hth head in the mechanism, for The specific calculation results of the u-th head in the mechanism; S4. Perform multimodal data fusion on the text, image, voice and physiological signal data in the multi-dimensional interactive data set, and use the fused multimodal data to train the deep learning model. The trained model is the exhibition hall content optimization model. S5. Use a streaming data processing framework to receive and process multi-dimensional interactive data in real time according to the content display requirements of the digital exhibition hall, and dynamically adjust the parameters and optimization strategies of the exhibition hall content optimization model according to the real-time analysis results; S6. Build a user feedback portal, collect user feedback information data through the user feedback portal, process the user feedback information data, and optimize the exhibition hall content optimization model according to the processing results; S7. Monitor the performance indicators of the exhibition hall content optimization model in real time, and optimize the structure, parameters and data processing methods of the exhibition hall content optimization model according to the monitoring results.

[0006] Furthermore, in step S1, the process of cleaning and verifying the collected data is as follows: A1. Use hash algorithm to check data duplication, find duplicate data, and delete the duplicate data; A2. Use the mean to fill in the missing data values. The formula for calculating the data mean is as follows: ,in, is the mean, is the data value, is the number of data; A3. According to the calculated mean Standard deviation of data Calculation, the calculation formula is as follows: ; A4. Based on the data standard deviation and mean Determine outliers in the data and use the mean to calculate outliers Replacement, the method of judging abnormal values ​​is: if the data value satisfy , it is considered as an outlier.

[0007] Furthermore, in step S2, the process of standardizing the verified data is as follows: B1. Use the standardized conversion formula to unify the data of different dimensions to the same scale to standardize the data. The standardized conversion formula is as follows: ,in, is the data standard deviation, is the mean, After verification, The data is processed after standardization; B2. Perform division interaction on data of different dimensions to generate multi-dimensional interaction feature data. The division interaction formula is as follows: , where m is the number of dimensional data types, , , , and They are different types of dimensional data after standardized transformation; B3. Divide the multi-dimensional interactive feature data into training set, validation set and test set, with the division ratio being 75% for training set, 15% for validation set and 10% for test set.

[0008] Furthermore, in step S4, the process of multimodal data fusion is as follows: C1, extracting text feature vector P, image feature vector I, speech feature vector M and physiological signal feature vector S from the multi-dimensional interaction data set; C2. Concatenate the text feature vector P, the image feature vector I, the speech feature vector M, and the physiological signal feature vector S into a fusion feature vector ,Right now =[P;I;M;S].

[0009] Furthermore, when training the deep learning model, the support vector machine function is used to predict the deep learning model according to the fusion feature vector F. The prediction formula is as follows: ,in, is the actual value of the i-th fused feature vector, b is the bias term, is the fusion feature weight vector, is the predicted value of the deep learning model; The loss function is then calculated based on the predicted value of the deep learning model, and the parameters of the deep learning model are updated through the back propagation algorithm based on the loss function calculation result. The formula of the loss function is as follows: , where L is the loss value calculated by the loss function, is the actual value of the fused feature vector F, is the predicted value of the deep learning model.

[0010] Furthermore, in step S5, the stream data processing framework uses the Apache Flink framework to receive and process the multi-dimensional interactive data in real time. The steps of multi-dimensional interactive data processing are as follows: D11. Use Apache Flink's data source connector to obtain multi-dimensional interactive data from different data sources in real time; D12. Clean the multi-dimensional interactive data accessed in real time to remove duplicate and abnormal data; D13. According to the display requirements of the exhibits, the time length is set to T, and the visitor stay time is , calculate the average audience stay time within the time length T , the calculation formula is: ,in, is the number of viewers within the time length T; D14. Statistics on the average stay time of visitors The total number of audience interactions during the time Number of interactions with a single exhibit , and calculate the popularity G of a single exhibit, the calculation formula is: .

[0011] Furthermore, the process of dynamically adjusting the parameters and optimization strategies of the exhibition hall content optimization model is as follows: D21. Calculate the accuracy and recall of the exhibition hall content optimization model based on the popularity of individual exhibits to confirm the performance indicators of the exhibition hall content optimization model; D22. Adjust the parameters and optimization strategies of the exhibition hall content optimization model based on performance indicators and real-time analysis results; Among them, the parameter adjustment formula of the exhibition hall content optimization model is: in, Optimize the model's initial learning rate for the exhibition hall content, is the diagonal element of the sum of squared gradients, To prevent small constants with zero denominators, is the adjusted learning rate; The optimization strategy is adjusted based on business logic and analysis results.

[0012] Furthermore, in step S6, the process of optimizing the exhibition hall content optimization model is as follows: E1. Collect user feedback information data from online questionnaire systems, message boards and comment areas; E2. Clean the collected text feedback data to remove special characters and stop words; E3. Use text mining algorithms to confirm the importance of each word in the cleaned text and extract keywords that can represent the main content of user feedback; E4, associating and integrating the processed keywords with the audience interaction data to form a comprehensive data set; E5. Analyze the factors that affect user feedback based on the data in the comprehensive data set. If a certain keyword is found to be related to the user's negative emotions, and the exhibition hall content corresponding to the keyword is related to the parameter setting of the model, adjust the relevant parameters in the model; if a certain keyword is not found to be related to the user's negative emotions, and the exhibition hall content corresponding to the keyword is related to the parameter setting of the model, enter E6; E6. Adjust the optimization strategy of exhibition hall content based on user feedback and data analysis results.

[0013] Furthermore, in step S7, the performance indicator monitoring and optimization process of the exhibition hall content optimization model is as follows: F1. Establish a monitoring system to regularly obtain data from the output results of the exhibition hall content optimization model and actual data, and calculate the values ​​of performance indicators; F2. Analyze the calculated performance indicators to determine whether the performance of the exhibition hall content optimization model meets the requirements; F21. If the performance of the exhibition hall content optimization model meets the requirements, there is no need to optimize the exhibition hall content optimization model; F22. If the performance of the hall content optimization model does not meet the requirements, further analysis of the hall content optimization model is required; F3. Confirm the existing problems of the model based on the analysis results, and optimize the exhibition hall content optimization model according to the existing problems; The optimization method for the exhibition hall content optimization model is: F31. If the model is too simple to capture complex audience behavior patterns, or if the model is too complex to cause overfitting, the model structure needs to be adjusted; F32. If the poor performance of the model is caused by unreasonable parameter settings, the model parameters need to be optimized; F33. If the model performance is caused by problems in the inspection data processing, such as incomplete data cleaning resulting in noise data affecting the model performance, or loopholes in the data cleaning rules, the cleaning rules need to be readjusted; F4. After optimizing the model structure, parameters and data processing methods, use the new model to predict the actual data, recalculate the performance indicators, evaluate the effect of optimization, and compare the optimized performance indicators with the previous indicators to determine whether the model performance has been improved. If the performance has not been significantly improved, repeat the above steps for further analysis and optimization.

[0014] Based on the above-described digital exhibition hall content optimization method based on multi-dimensional interactive data analysis, the present invention also provides a digital exhibition hall content optimization platform based on multi-dimensional interactive data analysis, including a data acquisition and preprocessing module, a deep learning model construction module, a multi-modal sentiment analysis fusion module, a real-time data analysis and processing module, a user feedback collection and fusion module for building a user feedback entrance, and a continuous monitoring and optimization module, which are electrically connected in sequence. The data acquisition and preprocessing module establishes a data interface with each system in the digital exhibition hall, extracts audience interaction data, exhibition hall content and environmental data from different data sources through the data interface, cleans and verifies the collected data, and then performs standardized transformation on the verified data to obtain a multi-dimensional interactive data set; The deep learning model building module uses the Transformer architecture to build a deep learning model for analyzing and predicting multi-dimensional interactive data sets; The multimodal sentiment analysis fusion module performs multimodal data fusion on the text, image, voice and physiological signal data in the multidimensional interaction data set, and uses the fused multimodal data to train the deep learning model; The real-time data analysis and processing module adopts a streaming data processing framework to receive and process multi-dimensional interactive data in real time according to the content display requirements of the digital exhibition hall, and dynamically adjusts the parameters and optimization strategies of the exhibition hall content optimization model according to the real-time analysis results; The user feedback collection and fusion module collects user feedback information data through the user feedback entrance, processes the user feedback information data, and optimizes the exhibition hall content optimization model according to the processing results; The continuous monitoring and optimization module monitors the performance indicators of the exhibition hall content optimization model in real time, and optimizes the structure, parameters and data processing method of the exhibition hall content optimization model according to the monitoring results.

[0015] The beneficial effects of the present invention are: The method provided by the present invention ensures the accuracy and completeness of the acquired multidimensional data by introducing a data cleaning and verification mechanism, adopts the deep learning architecture of Transformer to improve the prediction ability of the model, and at the same time, increases the diversity and quantity of training data to improve the generalization ability of the model; adopts streaming data processing technology to realize real-time analysis and processing of data, and at the same time, constructs a configurable exhibition hall content optimization model and optimization strategy so as to make rapid adjustments according to actual needs, thereby improving the optimization effect of the digital exhibition hall content; The method provided by the present invention establishes a user feedback channel to collect the opinions and suggestions of the audience on the content of the digital exhibition hall, and uses the user feedback as an important input of the optimization mathematical model to achieve more accurate personalized recommendations. By continuously monitoring the performance of the exhibition hall content optimization model and regularly evaluating its effect in practical applications, it is convenient to iteratively optimize the exhibition hall content optimization model according to the evaluation results to adapt to the ever-changing audience needs and digital exhibition hall environment.

[0016] The digital exhibition hall content optimization platform provided by the present invention, through the cooperation of the data collection and preprocessing module, the deep learning model construction module, the multimodal sentiment analysis and fusion module, the real-time data analysis and processing module, the user feedback collection and fusion module and the continuous monitoring and optimization module, through operations such as data cleaning, verification and standardization, ensures that the data used by the platform is accurate, complete and comparable, provides a reliable data foundation for subsequent analysis and modeling, and enables the platform to continuously adapt to the changes in the digital exhibition hall environment and the development of audience needs, and maintain good performance and effects.

[0017] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 A flowchart of a method for optimizing content in a digital exhibition hall according to an embodiment of the present invention is shown; Figure 2 A flowchart of cleaning and verifying collected data according to an embodiment of the present invention is shown; Figure 3A flowchart of standardization conversion of verified data according to an embodiment of the present invention is shown; Figure 4 A flowchart of multi-dimensional interactive data processing according to an embodiment of the present invention is shown; Figure 5 A flowchart of optimizing the exhibition hall content optimization model according to an embodiment of the present invention is shown; Figure 6 A system block diagram of a digital exhibition hall content optimization platform according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] The embodiment of the present invention provides a method for optimizing the content of a digital exhibition hall based on multi-dimensional interactive data analysis. Figure 1-5 As shown, the following steps are included: S1. Collect audience interaction data, content data, and environmental data from various systems in the digital exhibition hall, and clean and verify the collected data; like Figure 2 As shown in the figure, the process of cleaning and verifying the collected data is as follows: A1. Use hash algorithm to check data duplication, find duplicate data, and delete the duplicate data; A2. Use the mean to fill in the missing data values. The formula for calculating the data mean is as follows: ,in, is the mean, is the data value, is the number of data; A3. According to the calculated mean Standard deviation of data Calculation, the calculation formula is as follows: ; A4. Based on the data standard deviation and mean Determine outliers in the data and use the mean to calculate outliers Replacement, the method of judging abnormal values ​​is: if the data value satisfy , it is considered as an outlier; By introducing data cleaning and verification mechanisms, we ensure the accuracy and completeness of audience interaction data, content data, and environmental data, reduce the impact of extreme values ​​or outliers on deep learning models, and improve the robustness of the algorithm. At the same time, when data anomalies and missing data occur, we can correct the data in a timely manner to reduce errors caused by missing or anomalies, making it easier for subsequent analysis results to be close to the actual situation.

[0022] S2, standardize and transform the verified data to obtain a multi-dimensional interaction data set; like Figure 3 As shown in the figure, the process of standardizing the verified data is as follows: B1. Use the standardized conversion formula to unify the data of different dimensions to the same scale to standardize the data. The standardized conversion formula is as follows: ,in, is the data standard deviation, is the mean, After verification, The data is processed after standardization; B2. Perform division interaction on data of different dimensions to generate multi-dimensional interaction feature data. The division interaction formula is as follows: , where m is the number of dimensional data types, , , , and They are different types of dimensional data after standardized transformation; B3. Divide the multi-dimensional interactive feature data into training set, validation set and test set, with the ratio of training set 75%, validation set 15% and test set 10%; By first standardizing data of different dimensions, the data can be compared on the same scale, improving the convergence speed and performance of the deep learning model. Then, the data of different dimensions are interactively processed to capture the nonlinear relationship and interaction between the features of data of different dimensions, thereby improving the prediction ability of the deep learning model and reducing the deep learning model's excessive reliance on certain features.

[0023] S3, using the Transformer architecture to build a deep learning model for analyzing and predicting multi-dimensional interactive data sets; The formula for establishing the deep learning model is as follows: in, , are query, key, and value matrices respectively, is the weight matrix of the first head, is the weight matrix of the h-th head, is the dimension of the u-th key vector, h is the number of heads, u is the sequence number of the head, for The calculation result of the hth head in the mechanism, for The specific calculation results of the u-th head in the mechanism; S4. Multimodal data fusion is performed on the text, image, voice and physiological signal data in the multi-dimensional interactive data set, and the deep learning model is trained using the fused multimodal data. The trained model is the exhibition hall content optimization model. Text data refers to the introduction text of the exhibits, the display instructions and the comments of the audience in the digital exhibition hall, which can reflect the audience's understanding and feelings of the exhibition hall content from the language level. Image data refers to the images of exhibits in the exhibition hall, the facial expression images of the audience and the layout images of the exhibition hall. The exhibit images provide visual information such as the appearance and details of the exhibits. The facial expression images of the audience are collected by the cameras installed in the exhibition hall to analyze the emotional state of the audience. The layout images of the exhibition hall are used to understand the activities of the audience in different spatial areas. Language data refers to the voice communication of the audience in the exhibition hall and the voice content of the voice guide system. The voice content of the voice guide system can reflect whether the information conveyed by the exhibition hall is clear and easy to understand. Physiological signal data refers to the physiological signals of the audience collected through wearable devices (such as smart bracelets, smart watches, etc.). Physiological signals include heart rate, blood pressure and skin galvanic response. These physiological signals can more directly reflect the emotional and psychological state of the audience.

[0024] The process of multimodal data fusion is as follows: C1, extracting text feature vector P, image feature vector I, speech feature vector M and physiological signal feature vector S from the multi-dimensional interaction data set; C2. Concatenate the text feature vector P, image feature vector I, speech feature vector M, and physiological signal feature vector S into a fused feature vector Right now =[P;I;M;S]; When training the deep learning model, first use the support vector machine function to predict the deep learning model based on the fusion feature vector F. The prediction formula is as follows: ,in, is the actual value of the i-th fused feature vector, b is the bias term, is the fusion feature weight vector, is the predicted value of the deep learning model; The loss function is then calculated based on the predicted value of the deep learning model, and the parameters of the deep learning model are updated through the back propagation algorithm based on the loss function calculation result. The formula of the loss function is as follows: Among them, L is the loss value calculated by the loss function, is the actual value of the fused feature vector F, is the predicted value of the deep learning model; The Transformer architecture can improve the predictive ability of deep learning models. At the same time, it can increase the diversity and quantity of training data to improve the generalization ability of the model. In addition to text and image information, speech analysis and physiological signal data are also introduced to more comprehensively capture the emotional changes of the audience.

[0025] S5. Use a streaming data processing framework to receive and process multi-dimensional interactive data in real time according to the content display requirements of the digital exhibition hall, and dynamically adjust the parameters and optimization strategies of the exhibition hall content optimization model according to the real-time analysis results to improve the effect of digital exhibition hall content optimization; The streaming data processing framework uses the Apache Flink framework to receive and process multi-dimensional interactive data in real time, such as Figure 4 As shown, the steps of multi-dimensional interactive data processing are as follows: D11. Use Apache Flink's data source connector to obtain multi-dimensional interactive data from different data sources in real time; D12. Clean the multi-dimensional interactive data accessed in real time to remove duplicate and abnormal data; D13. According to the display requirements of the exhibits, the time length is set to T, and the visitor stay time is , calculate the average audience stay time within the time length T , the calculation formula is: ,in, is the number of viewers within the time length T; D14. Statistics on the average stay time of visitors The total number of audience interactions during the time Number of interactions with a single exhibit , and calculate the popularity G of a single exhibit, the calculation formula is: ; The process of dynamic adjustment of the parameters and optimization strategies of the exhibition hall content optimization model is as follows: D21. Calculate the accuracy and recall of the exhibition hall content optimization model based on the popularity of individual exhibits to confirm the performance indicators of the exhibition hall content optimization model; D22. Adjust the parameters and optimization strategies of the exhibition hall content optimization model based on performance indicators and real-time analysis results; Among them, the parameter adjustment formula of the exhibition hall content optimization model is: ,in, The initial learning rate of the exhibition hall content optimization model, is the diagonal element of the sum of squared gradients, To prevent small constants with zero denominators, is the adjusted learning rate; The optimization strategy is adjusted based on business logic and analysis results. Specifically, if it is found that certain types of interactive data have a greater impact on the model's prediction results, adjustments are made by increasing the collection and processing weights of these data; if the model performs poorly in certain time periods, adjustments are made by adjusting the model's training time window or the training data selection strategy. The display order of the exhibition hall content can also be adjusted based on audience feedback and real-time behavior data.

[0026] The streaming data processing framework enables real-time analysis and processing of data, while building configurable exhibition hall content optimization model parameters and optimization strategies to facilitate rapid adjustments based on actual needs.

[0027] S6. Build a user feedback portal, collect user feedback information data through the user feedback portal, process the user feedback information data, and optimize the exhibition hall content optimization model according to the processing results; Among them, the method of building the user feedback entrance is as follows: 1) Use HTML or CSS to build an online questionnaire system and design reasonable questionnaire questions, covering satisfaction with exhibition hall content, types of exhibits of interest, and suggestions for display methods; 2) Use Django or Flask to create a message board function, allowing visitors to freely enter text content to express their views and opinions on the digital exhibition hall; 3) Set up a comment area on the official website and social media accounts of the digital exhibition hall to facilitate visitors to comment on the relevant content of the exhibition hall.

[0028] like Figure 5 As shown in the figure, the process of optimizing the exhibition hall content optimization model is as follows: E1. Collect user feedback information data from online questionnaire systems, message boards and comment areas; E2. Clean the collected text feedback data to remove special characters and stop words; E3. Use text mining algorithms to confirm the importance of each word in the cleaned text and extract keywords that can represent the main content of user feedback; E4, associating and integrating the processed keywords with the audience interaction data to form a comprehensive data set; E5. Analyze the factors that affect user feedback based on the data in the comprehensive data set. If a keyword is found to be related to the user's negative emotions, and the exhibition hall content corresponding to the keyword is related to the parameter setting of the model, adjust the relevant parameters in the model; among which, the method of negative emotion judgment is: Assume the sentiment score of each word in the sentiment dictionary , the frequency of occurrence of word e in the text , then the sentiment score of the text The emotional tendency of the user feedback is judged by the calculated emotional score. If the score B is greater than 0, it indicates positive emotion, and if the score B is less than 0, it indicates negative emotion. If a certain keyword is not found to be related to the user's negative emotion, and the exhibition hall content corresponding to the keyword is related to the parameter setting of the model, enter E6.

[0029] E6. Adjust the optimization strategy of exhibition hall content based on user feedback and data analysis results. If users generally reflect that the display order of certain exhibits is unreasonable, the recommendation algorithm of exhibition hall content needs to be adjusted to change the display order of exhibits.

[0030] By establishing a user feedback portal, we collect audience opinions and suggestions on the content of the digital exhibition hall, and use user feedback as an important input for optimizing the exhibition hall content optimization model to achieve more accurate personalized recommendations.

[0031] S7. Monitor the performance indicators of the exhibition hall content optimization model in real time, and optimize the structure, parameters and data processing method of the exhibition hall content optimization model according to the monitoring results; The performance indicator monitoring and optimization process of the exhibition hall content optimization model is as follows: F1. Establish a monitoring system to regularly obtain data from the output results of the exhibition hall content optimization model and actual data, and calculate the values ​​of performance indicators; F2. Analyze the calculated performance indicators to determine whether the performance of the exhibition hall content optimization model meets the requirements; F21. If the performance of the exhibition hall content optimization model meets the requirements, there is no need to optimize the exhibition hall content optimization model; F22. If the performance of the hall content optimization model does not meet the requirements, further analysis of the hall content optimization model is required; F3. Confirm the existing problems of the model based on the analysis results, and optimize the exhibition hall content optimization model according to the existing problems; The optimization method for the exhibition hall content optimization model is: F31. If the model is too simple to capture complex audience behavior patterns, or if the model is too complex to cause overfitting, the model structure needs to be adjusted; F32. If the poor performance of the model is caused by unreasonable parameter settings, the model parameters need to be optimized; F33. If the model performance is caused by problems in the inspection data processing, such as incomplete data cleaning resulting in noise data affecting the model performance, or loopholes in the data cleaning rules, the cleaning rules need to be readjusted; F4. After optimizing the model structure, parameters and data processing methods, use the new model to predict the actual data, recalculate the performance indicators, evaluate the effect of optimization, and compare the optimized performance indicators with the previous indicators to determine whether the model performance has been improved. If the performance has not been significantly improved, repeat the above steps for further analysis and optimization.

[0032] By continuously monitoring the performance of the exhibition hall content optimization model, the effectiveness of the exhibition hall content optimization model in actual application is regularly evaluated. Based on the evaluation results, the exhibition hall content optimization model is iteratively optimized to adapt to the ever-changing audience needs and digital exhibition hall environment.

[0033] By introducing data cleaning and verification agencies to ensure the accuracy and completeness of the acquired multi-dimensional data, the Transformer deep learning architecture is adopted to improve the prediction ability of the model. At the same time, the diversity and quantity of training data are increased to enhance the generalization ability of the model. In addition to text and image information, speech analysis and physiological signal monitoring technology can also be considered to more comprehensively capture the emotional changes of the audience. Streaming data processing technology is used to achieve real-time analysis and processing of data. At the same time, a configurable exhibition hall content optimization model and optimization strategy are constructed to enable rapid adjustments based on actual needs. A user feedback channel is established to collect the opinions and suggestions of the audience on the content of the digital exhibition hall. User feedback is used as an important input for optimizing the mathematical model to achieve more accurate personalized recommendations. The performance of the exhibition hall content optimization model is continuously monitored and its effect in actual application is regularly evaluated. According to the evaluation results, the exhibition hall content optimization model can be iteratively optimized to adapt to the ever-changing audience needs and digital exhibition hall environment, thereby improving the optimization effect of the digital exhibition hall content.

[0034] Based on the above-described digital exhibition hall content optimization method based on multi-dimensional interactive data analysis, the present invention also provides a digital exhibition hall content optimization platform based on multi-dimensional interactive data analysis, such as Figure 6 As shown, it includes a data acquisition and preprocessing module, a deep learning model building module, a multimodal sentiment analysis and fusion module, a real-time data analysis and processing module, a user feedback collection and fusion module, and a continuous monitoring and optimization module, which are electrically connected in sequence. The data acquisition and preprocessing module establishes a data interface with various systems in the digital exhibition hall (such as an interactive device system, a content display system, and an environmental monitoring system), uses network communication technology to obtain data in real time or at a fixed time, extracts audience interaction data, exhibition hall content, and environmental data from different data sources through the data interface, cleans and verifies the collected data, and then performs standardized transformation on the verified data to obtain a multi-dimensional interactive data set; The deep learning model building module uses the Transformer architecture to build a deep learning model for analyzing and predicting multi-dimensional interactive data sets; The multimodal sentiment analysis fusion module performs multimodal data fusion on the text, image, voice and physiological signal data in the multidimensional interaction data set, and uses the fused multimodal data to train the deep learning model; The real-time data analysis and processing module adopts a streaming data processing framework to receive and process multi-dimensional interactive data in real time according to the content display requirements of the digital exhibition hall, and dynamically adjusts the parameters and optimization strategies of the exhibition hall content optimization model according to the real-time analysis results; The user feedback collection and fusion module is used to build a user feedback portal, collect user feedback information data through the user feedback portal, process the user feedback information data, and optimize the exhibition hall content optimization model according to the processing results; The continuous monitoring and optimization module monitors the performance indicators of the exhibition hall content optimization model in real time, and optimizes the structure, parameters and data processing method of the exhibition hall content optimization model according to the monitoring results.

[0035] Through the cooperation of data collection and preprocessing module, deep learning model building module, multimodal sentiment analysis and fusion module, real-time data analysis and processing module, user feedback collection and fusion module and continuous monitoring and optimization module, through data cleaning, verification and standardization, the data used by the platform is ensured to be accurate, complete and comparable, providing a reliable data foundation for subsequent analysis and modeling; the deep learning model based on Transformer architecture and multimodal data fusion technology can conduct in-depth analysis and accurate prediction of multi-dimensional interactive data, better understand the audience's behavior, needs and emotional changes, and provide strong support for the optimization of exhibition hall content; the use of streaming data processing technology Technology is used to realize real-time analysis and processing of data, and dynamically adjust model parameters and optimization strategies based on real-time analysis results, so that the exhibition hall content can adapt to the audience's needs and behavioral changes in a timely manner, and improve the audience's visiting experience and satisfaction; combined with user feedback data, the exhibition hall content optimization model is continuously optimized to achieve more accurate personalized recommendations, provide the audience with exhibition hall content that is more in line with their interests and needs, and enhance the audience's participation and stickiness; the performance of the model is continuously monitored and evaluated through the continuous monitoring and optimization module, and the various modules and methods of the platform are iteratively optimized according to the evaluation results, so that the platform can continuously adapt to the changes in the digital exhibition hall environment and the development of audience needs, and maintain good performance and effects.

[0036] Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein; and these modifications or substitutions do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A digital exhibition hall content optimization method based on multi-dimensional interactive data analysis, characterized by: The steps include: S1. Collect audience interaction data, content data, and environmental data from various systems in the digital exhibition hall, and clean and verify the collected data; S2, standardize and transform the verified data to obtain a multi-dimensional interaction data set; S3, using the Transformer architecture to build a deep learning model for analyzing and predicting multi-dimensional interactive data sets; The formula for establishing the deep learning model is as follows: ,in, , are query, key, and value matrices respectively, is the weight matrix of the first head, is the weight matrix of the h-th head, is the dimension of the u-th key vector, h is the number of heads, u is the sequence number of the head, for The calculation result of the hth head in the mechanism, for The specific calculation results of the u-th head in the mechanism; S4. Perform multimodal data fusion on the text, image, voice and physiological signal data in the multi-dimensional interactive data set, and use the fused multimodal data to train the deep learning model. The trained model is the exhibition hall content optimization model. S5. Use a streaming data processing framework to receive and process multi-dimensional interactive data in real time according to the content display requirements of the digital exhibition hall, and dynamically adjust the parameters and optimization strategies of the exhibition hall content optimization model according to the real-time analysis results; S6. Build a user feedback portal, collect user feedback information data through the user feedback portal, process the user feedback information data, and optimize the exhibition hall content optimization model according to the processing results; S7. Monitor the performance indicators of the exhibition hall content optimization model in real time, and optimize the structure, parameters and data processing methods of the exhibition hall content optimization model according to the monitoring results.

2. The method for optimizing digital exhibition hall content based on multi-dimensional interactive data analysis according to claim 1, characterized in that: In step S1, the process of cleaning and verifying the collected data is as follows: A1. Use hash algorithm to check data duplication, find duplicate data, and delete the duplicate data; A2. Use the mean to fill in the missing data values. The formula for calculating the data mean is as follows: ,in, is the mean, is the data value, is the number of data; A3. According to the calculated mean Standard deviation of data Calculation, the calculation formula is as follows: ； A4. Based on the data standard deviation and mean Determine outliers in the data and use the mean to calculate outliers Replacement, the method of judging abnormal values ​​is: if the data value satisfy , it is considered as an outlier.

3. The method for optimizing digital exhibition hall content based on multi-dimensional interactive data analysis according to claim 2 is characterized in that: In step S2, the process of standardizing the verified data is as follows: B1. Use the standardized conversion formula to unify the data of different dimensions to the same scale to standardize the data. The standardized conversion formula is as follows: ,in, is the data standard deviation, is the mean, To verify the data, The data is processed after standardization; B2. Perform division interaction on data of different dimensions to generate multi-dimensional interaction feature data. The division interaction formula is as follows: , where m is the number of dimensional data types, , , , and They are different types of dimensional data after standardized transformation; B3. Divide the multi-dimensional interactive feature data into training set, validation set and test set, with the division ratio being 75% for training set, 15% for validation set and 10% for test set.

4. The method for optimizing digital exhibition hall content based on multi-dimensional interactive data analysis according to claim 3 is characterized in that: In step S4, the process of multimodal data fusion is as follows: C1, extracting text feature vector P, image feature vector I, speech feature vector M and physiological signal feature vector S from the multi-dimensional interaction data set; C2. Concatenate the text feature vector P, image feature vector I, speech feature vector M, and physiological signal feature vector S into a fused feature vector ,Right now =[P;I;M;S].

5. The method for optimizing digital exhibition hall content based on multi-dimensional interactive data analysis according to claim 4 is characterized in that: When training the deep learning model, first use the support vector machine function to predict the deep learning model based on the fusion feature vector F. The prediction formula is as follows: ,in, is the actual value of the i-th fused feature vector, b is the bias term, is the fusion feature weight vector, is the predicted value of the deep learning model; The loss function is then calculated based on the predicted value of the deep learning model, and the parameters of the deep learning model are updated through the back propagation algorithm based on the loss function calculation result. The formula of the loss function is as follows: , where L is the loss value calculated by the loss function, is the actual value of the fused feature vector F, is the predicted value of the deep learning model.

6. The method for optimizing digital exhibition hall content based on multi-dimensional interactive data analysis according to claim 5 is characterized in that: In step S5, the streaming data processing framework uses the Apache Flink framework to receive and process the multi-dimensional interactive data in real time. The steps of multi-dimensional interactive data processing are as follows: D11. Use Apache Flink's data source connector to obtain multi-dimensional interactive data from different data sources in real time; D12. Clean the multi-dimensional interactive data accessed in real time to remove duplicate and abnormal data; D13. According to the display requirements of the exhibits, the time length is set to T, and the visitor stay time is , calculate the average audience stay time within the time length T , the calculation formula is: ,in, is the number of viewers within the time length T; D14. Statistics on the average stay time of visitors The total number of audience interactions during the time Number of interactions with a single exhibit , and calculate the popularity G of a single exhibit, the calculation formula is: .

7. The method for optimizing digital exhibition hall content based on multi-dimensional interactive data analysis according to claim 6 is characterized in that: The process of dynamic adjustment of the parameters and optimization strategies of the exhibition hall content optimization model is as follows: D21. Calculate the accuracy and recall of the exhibition hall content optimization model based on the popularity of individual exhibits to confirm the performance indicators of the exhibition hall content optimization model; D22. Adjust the parameters and optimization strategies of the exhibition hall content optimization model based on performance indicators and real-time analysis results; Among them, the parameter adjustment formula of the exhibition hall content optimization model is: in, Optimize the model's initial learning rate for the exhibition hall content, is the diagonal element of the sum of squared gradients, To prevent small constants with zero denominators, is the adjusted learning rate; The optimization strategy is adjusted based on business logic and analysis results.

8. The method for optimizing digital exhibition hall content based on multi-dimensional interactive data analysis according to claim 7 is characterized in that: In step S6, the process of optimizing the exhibition hall content optimization model is as follows: E1. Collect user feedback information data from online questionnaire systems, message boards and comment areas; E2. Clean the collected text feedback data to remove special characters and stop words; E3. Use text mining algorithms to confirm the importance of each word in the cleaned text and extract keywords that can represent the main content of user feedback; E4, associating and integrating the processed keywords with the audience interaction data to form a comprehensive data set; E5. Analyze the factors that affect user feedback based on the data in the comprehensive data set. If a keyword is found to be related to the user's negative emotions, and the exhibition hall content corresponding to the keyword is related to the parameter setting of the model, adjust the relevant parameters in the model; If a keyword is not found to be related to the user's negative emotions, and the exhibition hall content corresponding to the keyword is related to the parameter setting of the model, then enter E6; E6. Adjust the optimization strategy of exhibition hall content based on user feedback and data analysis results.

9. The method for optimizing digital exhibition hall content based on multi-dimensional interactive data analysis according to claim 8 is characterized in that: In step S7, the performance indicator monitoring and optimization process of the exhibition hall content optimization model is as follows: F1. Establish a monitoring system to regularly obtain data from the output results of the exhibition hall content optimization model and actual data, and calculate the values ​​of performance indicators; F2. Analyze the calculated performance indicators to determine whether the performance of the exhibition hall content optimization model meets the requirements; F21. If the performance of the exhibition hall content optimization model meets the requirements, there is no need to optimize the exhibition hall content optimization model; F22. If the performance of the hall content optimization model does not meet the requirements, further analysis of the hall content optimization model is required; F3. Confirm the existing problems of the model based on the analysis results, and optimize the exhibition hall content optimization model according to the existing problems; The optimization method for the exhibition hall content optimization model is: F31. If the model is too simple to capture complex audience behavior patterns, or if the model is too complex to cause overfitting, the model structure needs to be adjusted; F32. If the poor performance of the model is caused by unreasonable parameter settings, the model parameters need to be optimized; F33. If the model performance is caused by problems in the inspection data processing, such as incomplete data cleaning resulting in noise data affecting the model performance, or loopholes in the data cleaning rules, the cleaning rules need to be readjusted; F4. After optimizing the model structure, parameters and data processing methods, use the new model to predict the actual data, recalculate the performance indicators, evaluate the effect of optimization, and compare the optimized performance indicators with the previous indicators to determine whether the model performance has been improved. If the performance has not been significantly improved, repeat the above steps for further analysis and optimization.

10. A digital exhibition hall content optimization platform based on multi-dimensional interactive data analysis, based on a digital exhibition hall content optimization method based on multi-dimensional interactive data analysis according to any one of claims 1 to 9, characterized in that: It includes a data collection and preprocessing module, a deep learning model building module, a multimodal sentiment analysis fusion module, a real-time data analysis and processing module, a user feedback collection and fusion module for building a user feedback portal, and a continuous monitoring and optimization module, which are electrically connected in sequence. The data collection and preprocessing module establishes a data interface with each system in the digital exhibition hall, extracts audience interaction data, exhibition hall content and environmental data from different data sources through the data interface, cleans and verifies the collected data, and then performs standardized transformation on the verified data to obtain a multi-dimensional interactive data set; The deep learning model building module uses the Transformer architecture to build a deep learning model for analyzing and predicting multi-dimensional interactive data sets; The multimodal sentiment analysis fusion module performs multimodal data fusion on the text, image, voice and physiological signal data in the multidimensional interaction data set, and uses the fused multimodal data to train the deep learning model; The real-time data analysis and processing module adopts a streaming data processing framework to receive and process multi-dimensional interactive data in real time according to the content display requirements of the digital exhibition hall, and dynamically adjusts the parameters and optimization strategies of the exhibition hall content optimization model according to the real-time analysis results; The user feedback collection and fusion module collects user feedback information data through the user feedback entrance, processes the user feedback information data, and optimizes the exhibition hall content optimization model according to the processing results; The continuous monitoring and optimization module monitors the performance indicators of the exhibition hall content optimization model in real time, and optimizes the structure, parameters and data processing method of the exhibition hall content optimization model according to the monitoring results.

Images