A method and system for predicting emotional dynamics based on graph neural networks and time series

Abstract

The present application relates to the field of sentiment analysis technology, and in particular to a method and system for predicting sentiment dynamics based on graph neural networks and time series. The method includes: acquiring and compressing an image to be analyzed to obtain a compressed image, and extracting features from the compressed image to obtain image features; analyzing the image features for each image feature, and assigning emotional characteristics based on the feature analysis results; randomly combining at least two image features, and determining the emotional triggering situation based on the combination results and the emotional characteristics; and determining whether to shield the compressed image based on the emotional triggering situation. Reducing the time and resources required for image processing, by combining different features and analyzing the emotional triggering situation under their combined action, helps to identify the overall emotional tendency of the image. Based on the emotional triggering situation, decide whether to shield the image to protect the user from negative emotional influences.

Description

Technical Field

[0001] The present application relates to the field of sentiment analysis technology, and in particular to a method and system for predicting sentiment dynamics based on graph neural networks and time series. Background Art

[0002] In the information and digital age, the dissemination and sharing of image content has permeated every aspect of life, particularly through social media, video platforms, and online communities. While images themselves possess powerful communication capabilities, their potential emotional triggers, social impact, and potential psychological and physiological reactions are often overlooked. To address these challenges, the review, filtering, and dynamic prediction of image content are crucial.

[0003] However, traditional image review technologies typically rely on manual review or rule-based screening mechanisms. This approach is not only inefficient when faced with massive amounts of images but is also susceptible to human bias, subjective judgment, and time constraints. As the volume of image data continues to grow, manual review can no longer meet the demands for speed, accuracy, and comprehensiveness. Summary of the Invention

[0004] This application provides a method and system for predicting emotional dynamics based on graph neural networks and time series to solve the above problems.

[0005] In a first aspect, the present application provides a method for predicting emotional dynamics based on graph neural networks and time series, the method comprising:

[0006] Acquiring and compressing an image to be analyzed to obtain a compressed image, and performing feature extraction on the compressed image to obtain image features;

[0007] For each image feature, analyzing the image feature, and assigning an emotional characteristic based on the feature analysis result;

[0008] Randomly combining at least two image features, and determining an emotional triggering situation based on the combination result and the emotional characteristics;

[0009] According to the emotional triggering situation, it is determined whether to mask the compressed image.

[0010] This solution reduces the time and resources required for image processing while preserving the image's key information. It helps extract image features, identifies key structures within the image, and provides a basis for assigning emotional characteristics. Each image feature is assigned a corresponding emotional label, such as happiness, sadness, or anger. By combining different features and analyzing their combined emotional triggers, it helps identify the overall emotional tendency of the image. Based on the emotional triggers, it decides whether to block the image to protect users from negative emotional influences.

[0011] Optionally, analyzing each image feature includes:

[0012] Analyzing the image features to determine the overall image structure;

[0013] Analyzing the overall structure of the image and determining a feature analysis area;

[0014] For the feature analysis area, feature analysis is performed on each image feature.

[0015] This solution analyzes image features such as color, shape, and texture to identify the overall structure of an image, including faces, objects, and backgrounds. This helps identify the overall layout and content of an image. Based on the overall image structure, feature areas requiring focused analysis can be identified. For example, in face recognition, the areas of focus for analysis might be the eyes, nose, and mouth. This helps improve the efficiency and accuracy of feature analysis. Detailed feature analysis of each feature analysis area can extract more targeted feature information. For example, analyzing the face area can extract features such as the eyes, nose, and mouth, and further analyze the expressions and shapes of these features. This helps to more accurately analyze and describe image content.

[0016] Optionally, assigning emotional characteristics based on the feature analysis results includes:

[0017] Analyzing the image features to determine image characteristics of each image feature;

[0018] Inputting the image features into a preset analysis model to obtain feature analysis results;

[0019] According to the characteristic analysis results, the emotional characteristics of each image feature are determined.

[0020] This solution identifies key features in images, such as color, shape, and texture. These features help identify the image's visual content and style. Using a pre-set analysis model, we conduct in-depth analysis of these image features, obtaining detailed data on their emotional orientation and intensity. Based on the analysis results, each image feature is assigned an emotional label or description, such as happiness, sadness, or anger. These emotional characteristics facilitate sentiment classification and emotional state prediction.

[0021] Optionally, determining the emotional triggering situation based on the combination result and the emotional characteristics includes:

[0022] Determining the combined color and combined composition of the random combination features according to the combination result;

[0023] Analyzing the combined colors and the combined composition to determine a characteristic atmosphere of the random combination feature;

[0024] Determine whether there is an emotional conflict based on several emotional characteristics of the random combination features;

[0025] If there is an emotional conflict, the emotional triggering situation is determined based on the characteristic atmosphere and the several emotional features.

[0026] This solution analyzes randomly combined features within an image to determine its overall color scheme and composition, helping to identify the image's visual style and layout. Based on the characteristics of the combined colors and composition, it can determine the image's characteristic atmosphere, such as harmony, conflict, and tension, helping to grasp the image's emotional tendencies. By comparing the emotional characteristics of different features, such as the contrast between happiness and sadness, it can determine whether there is emotional conflict within the image, helping to identify inconsistent or contradictory emotions within the image. If emotional conflict is detected, the characteristic atmosphere and emotional characteristics can be combined to determine the specific circumstances of the emotional trigger, such as the degree, type, and direction of emotional evolution.

[0027] Optionally, before determining whether to shield the compressed image according to the emotional triggering situation, the method further includes:

[0028] determining whether an abnormal structure exists in the compressed image according to the feature analysis result;

[0029] If an irrational structure exists, analyzing the irrational structure and determining the structural properties;

[0030] Conduct ethical analysis on the irrational structure based on the structural attributes;

[0031] According to the result of the ethics analysis, it is determined whether to shield the compressed image.

[0032] This solution captures detailed feature data about compressed images. It identifies structures within the image that are unconventional or expected, potentially containing negative content or ethical issues. It analyzes the size, shape, texture, and other attributes of these unusual structures in detail. This information helps identify the nature and impact of these structures. It assesses the ethical risks posed by these unusual structures, providing an ethical basis for decision-making. Combined with sentiment analysis results, it determines whether these unusual structures trigger emotions such as fear and anger. Combining ethical risk assessment with emotional triggers yields a comprehensive ethical analysis. Based on the ethical analysis results, a decision is made on whether to block the compressed image to ensure that the content adheres to social ethical standards.

[0033] Optionally, determining whether to shield the compressed image according to the emotion triggering situation includes:

[0034] Assigning an emotion to the random combination feature according to the emotion triggering situation to obtain an emotion score for the random combination feature;

[0035] determining whether the random combination feature has content inappropriateness based on the emotion score and a preset emotion intensity level;

[0036] If it exists, the compressed image is masked.

[0037] This solution assigns emotional labels such as happiness, sadness, and anger to randomly combined features. These emotional labels are converted into specific emotional scores to quantify emotional intensity. A standard emotional intensity rating system is established to provide a reference for evaluating emotional scores. By comparing emotional scores with pre-set ratings, content that causes discomfort can be identified. Emotional dissonance between different features in an image, such as the coexistence of happiness and sadness, indicates emotional conflict. The overall emotional impact of an image is evaluated by comprehensively considering the emotional score, emotional conflict, and potential ethical issues. Based on this comprehensive evaluation, a decision is made as to whether the image should be blocked to prevent the spread of inappropriate content.

[0038] Optionally, the method further includes:

[0039] After shielding the compressed image, analyzing the compressed image to determine the image type;

[0040] Perform an online search based on the image type to obtain online pictures that match the image type;

[0041] Analyzing image features of the online image and the compressed image to determine an image evolution portion;

[0042] Predicting the evolution possibility within a preset time period based on the image evolution portion;

[0043] According to the evolution possibility, a number of prediction images are generated and transmitted to the preset analysis model to train the preset analysis model.

[0044] Through this solution, the category of the image can be determined by analyzing the compressed image after shielding. By performing online retrieval based on the image type, a large amount of relevant image data can be collected, providing more reference samples for image evolution analysis. By analyzing the image features of online pictures and compressed images, key information that helps to identify the image evolution trend can be extracted. By determining the image evolution part, it is helpful to identify the area where the image changes over time, providing a basis for predicting future evolution trends. Predicting the possible evolution within a preset time period can identify the changing trend of image content in advance, which is helpful for content review and sentiment analysis. Generating predicted images can simulate possible future image states and provide forward-looking information for content review and sentiment analysis. By training the preset analysis model, the model's ability to predict future image evolution trends can be improved, and the model's accuracy can be enhanced.

[0045] Optionally, analyzing the image features of the online image and the compressed image to determine the image evolution portion includes:

[0046] Analyzing the online picture to determine online image features;

[0047] Analyzing the network image features and the image features to determine common features;

[0048] Determine the characteristic differences of the remaining characteristics based on the same characteristics;

[0049] According to the feature distinction, the image evolution portion is determined.

[0050] This solution analyzes online images to obtain a rich set of image features. By comparing the features of online and compressed images, shared features can be identified, helping to identify similarities and differences between images. After identifying common features, it becomes easier to identify and distinguish differences between remaining features, including changes in feature intensity, position, or context, which helps reveal details of image evolution. By analyzing feature differences, it is possible to identify areas of change in the image, such as increases or decreases in feature intensity or changes in feature position. These areas are key components of image evolution.

[0051] Optionally, analyzing the network image features and the image features to determine the same features includes:

[0052] Analyze the online picture to determine the upload time;

[0053] Sort the networked images in time series according to the upload time to obtain a plurality of ordered images;

[0054] For each ordered image, analyzing the network image features and the image features according to the time sequence to determine the same features;

[0055] The determining of the image evolution portion based on the feature distinction includes:

[0056] Based on the time series, the feature distinctions are analyzed to determine the image evolution portion.

[0057] This solution determines the upload time of each online image, providing a timeline for time series analysis and image evolution tracking. Images are sorted chronologically to create a time series, facilitating the observation and analysis of image changes over time. By analyzing the features of each image, we can identify features between images at different time points. These features can serve as reference points for image evolution analysis. By comparing the differences between these features, we can identify the parts of the image that evolve over time, revealing the dynamic changes in the image. By analyzing feature differences over time, we can more accurately track the evolution of image features.

[0058] In a second aspect, the present application provides a system for predicting emotional dynamics based on graph neural networks and time series, the system comprising:

[0059] A feature extraction module is used to obtain and compress the image to be analyzed to obtain a compressed image, and perform feature extraction on the compressed image to obtain image features;

[0060] A feature analysis module is used to analyze each image feature and assign emotional characteristics based on the feature analysis results;

[0061] An emotion trigger analysis module, configured to randomly combine at least two image features and determine an emotion trigger condition based on the combination result and the emotion characteristics;

[0062] The shielding judgment module is used to determine whether to shield the compressed image according to the emotional triggering situation.

[0063] Optionally, the feature analysis module is used to analyze each image feature to:

[0064] Analyzing the image features to determine the overall image structure;

[0065] Analyzing the overall structure of the image and determining a feature analysis area;

[0066] For the feature analysis area, feature analysis is performed on each image feature.

[0067] Optionally, when assigning emotional characteristics based on the feature analysis results, the feature analysis module is used to:

[0068] Analyzing the image features to determine image characteristics of each image feature;

[0069] Inputting the image features into a preset analysis model to obtain feature analysis results;

[0070] According to the characteristic analysis results, the emotional characteristics of each image feature are determined.

[0071] Optionally, when the emotion trigger analysis module determines the emotion trigger situation based on the combination result and the emotion characteristics, it is used to:

[0072] Determining the combined color and combined composition of the random combination features according to the combination result;

[0073] Analyzing the combined colors and the combined composition to determine a characteristic atmosphere of the random combination feature;

[0074] Determine whether there is an emotional conflict based on several emotional characteristics of the random combination features;

[0075] If there is an emotional conflict, the emotional triggering situation is determined based on the characteristic atmosphere and the several emotional features.

[0076] Optionally, the emotion dynamics prediction system based on graph neural network and time series further includes a structure analysis module for:

[0077] determining whether an abnormal structure exists in the compressed image according to the feature analysis result;

[0078] If an irrational structure exists, analyzing the irrational structure and determining the structural properties;

[0079] Conduct ethical analysis on the irrational structure based on the structural attributes;

[0080] According to the result of the ethics analysis, it is determined whether to shield the compressed image.

[0081] Optionally, when the shielding determination module determines whether to shield the compressed image according to the emotion triggering condition, it is configured to:

[0082] Assigning an emotion to the random combination feature according to the emotion triggering situation to obtain an emotion score for the random combination feature;

[0083] determining whether the random combination feature has content inappropriateness based on the emotion score and a preset emotion intensity level;

[0084] If it exists, the compressed image is masked.

[0085] Optionally, the graph neural network and time series-based emotion dynamics prediction system further includes a model training module for:

[0086] After shielding the compressed image, analyzing the compressed image to determine the image type;

[0087] Perform an online search based on the image type to obtain online pictures that match the image type;

[0088] Analyzing image features of the online image and the compressed image to determine an image evolution portion;

[0089] Predicting the evolution possibility within a preset time period based on the image evolution portion;

[0090] According to the evolution possibility, a number of prediction images are generated and transmitted to the preset analysis model to train the preset analysis model.

[0091] Optionally, when the model training module analyzes the image features of the online image and the compressed image and determines the image evolution portion, it is used to:

[0092] Analyzing the online picture to determine online image features;

[0093] Analyzing the network image features and the image features to determine common features;

[0094] Determine the characteristic differences of the remaining characteristics based on the same characteristics;

[0095] According to the feature distinction, the image evolution portion is determined.

[0096] Optionally, the model training module analyzes the network image features and the image features, and when identical features are determined, is used to:

[0097] Analyze the online picture to determine the upload time;

[0098] Sort the networked images in time series according to the upload time to obtain a plurality of ordered images;

[0099] For each ordered image, analyzing the network image features and the image features according to the time sequence to determine the same features;

[0100] When determining the image evolution portion based on the feature distinction, it is used to:

[0101] Based on the time series, the feature distinctions are analyzed to determine the image evolution portion. BRIEF DESCRIPTION OF THE DRAWINGS

[0102] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to 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 application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0103] Figure 1 A schematic diagram of an application scenario provided in one embodiment of the present application;

[0104] Figure 2 A flowchart of a method for predicting emotional dynamics based on graph neural networks and time series provided in one embodiment of the present application;

[0105] Figure 3 A schematic diagram of the structure of an emotion dynamics prediction system based on graph neural networks and time series is provided in one embodiment of the present application. DETAILED DESCRIPTION

[0106] To make the purpose, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0107] In this document, the term "and / or" simply describes a relationship between related objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. Furthermore, the character " / " in this document, unless otherwise specified, generally indicates an "or" relationship between the related objects.

[0108] The embodiments of the present application are described in further detail below with reference to the accompanying drawings.

[0109] In the information and digital age, the dissemination and sharing of image content has permeated every aspect of life, especially through social media, video platforms, and online communities. While images themselves possess powerful communication capabilities, their potential emotional triggering, social impact, and potential psychological or physiological reactions are often overlooked. To address these challenges, the review, filtering, and dynamic prediction of image content are crucial. However, traditional image review techniques typically rely on manual review or rule-based screening mechanisms, which are inefficient when faced with massive amounts of images and are susceptible to human bias, subjective judgment, and time constraints. As the volume of image data continues to grow, manual review can no longer meet the demands for speed, accuracy, and comprehensiveness.

[0110] Based on this, the present application provides a method and system for predicting dynamic emotions based on graph neural networks and time series, which acquires and compresses the image to be analyzed to obtain a compressed image, and performs feature extraction on the compressed image to obtain image features; for each image feature, analyzes the image feature, and assigns emotional characteristics based on the feature analysis results; randomly combines at least two image features, and determines the emotional triggering situation based on the combination results and the emotional characteristics; and determines whether to block the compressed image based on the emotional triggering situation. Reduces the time and resources required for image processing while retaining the main information of the image. Helps to extract the features of the image. Identifies the key structures in the image and provides a basis for assigning emotional characteristics. Assigns corresponding emotional labels such as happiness, sadness, anger, etc. to each image feature. By combining different features and analyzing the emotional triggering situation under their joint action, it helps to identify the overall emotional tendency of the image. Based on the emotional triggering situation, decides whether to block the image to protect users from negative emotional influences.

[0111] Figure 1 This is a schematic diagram of an application scenario provided by this application. When performing image analysis to determine the emotional state of an image, the method provided by this application is applied.

[0112] Specifically, the method provided in the present application is applied to any server, and the server interacts with the user end to obtain and analyze the image to be analyzed uploaded by the user end, thereby reducing the time and resources required for image processing while retaining the main information of the image. It is helpful to extract the features of the image. It identifies the key structures in the image and provides a basis for assigning emotional characteristics. Each image feature is assigned a corresponding emotional label such as happiness, sadness, anger, etc. By combining different features and analyzing the emotional triggering conditions under their joint action, it is helpful to identify the overall emotional tendency of the image. Based on the emotional triggering conditions, it is decided whether to shield the image to protect the user from adverse emotional influences. For specific implementation methods, please refer to the following embodiments.

[0113] Figure 2 This is a flowchart of a method for predicting emotional dynamics based on graph neural networks and time series provided in one embodiment of the present application. The method of this embodiment can be applied to the server in the above scenario. Figure 2 As shown, the method includes:

[0114] S201, acquiring and compressing an image to be analyzed to obtain a compressed image, and performing feature extraction on the compressed image to obtain image features;

[0115] The image to be analyzed may be an original image that may have problems and is uploaded or captured by the user and needs to be analyzed.

[0116] The compressed image may be an image obtained by compressing the original image.

[0117] Specifically, we obtain original images from user-uploaded or web-crawled sources. We use image compression algorithms to reduce the image file size. We then apply convolutional neural networks (CNNs) or other feature extraction techniques to extract features such as color, texture, shape, and edges from the compressed images.

[0118] S202, analyzing each image feature, and assigning an emotional characteristic based on the feature analysis result;

[0119] Image features can be information extracted from an image that can describe the image content.

[0120] The feature analysis result may be a result obtained by analyzing the extracted image features.

[0121] Specifically, a graph neural network (GNN) is used to further analyze the extracted features and identify the overall structure of the image. The facial, body, or other prominent areas to be analyzed are identified. Each feature area is analyzed to determine its image characteristics. These image characteristics are then fed into a pre-defined analysis model, such as a deep learning model, for feature analysis. Based on the analysis results, each feature area is assigned an emotional characteristic, such as happiness, sadness, or anger.

[0122] In a specific implementation, the above-mentioned deep learning model can be trained using image feature areas of several known image characteristics and their corresponding emotional characteristics, so that the deep learning model can analyze the input image characteristics and determine the corresponding emotional characteristics.

[0123] S203, randomly combining at least two image features, and determining an emotional triggering situation based on the combination result and the emotional characteristics;

[0124] The combination result may be a result obtained by randomly combining at least two image features.

[0125] Emotional features can be emotional labels or emotional descriptions assigned to image features based on image feature analysis results.

[0126] Specifically, at least two of the aforementioned features are randomly selected and combined. The combined colors and composition of the combined features are analyzed. The characteristic atmosphere of the combined features is determined. Based on the emotional characteristics of the combined features, it is determined whether there is an emotional conflict. If an emotional conflict exists, the emotional trigger is determined based on the characteristic atmosphere and emotional characteristics.

[0127] S204: Determine whether to shield the compressed image according to the emotional triggering situation.

[0128] The emotional triggering situation can be an emotional response triggered by an image determined based on a feature combination result and emotional characteristics.

[0129] Specifically, to facilitate analysis of emotional triggers, a preset emotional intensity level can be set in advance. This preset emotional intensity level is primarily based on the emotions displayed by the image to be blocked, and therefore is set to favor negative and dark emotions. A higher level indicates a more negative and uncomfortable feeling.

[0130] Based on the emotional trigger, the combined features are assigned an emotional value to obtain an emotional score. The emotional score is compared with the preset emotional intensity level to determine the emotional intensity level that matches the current emotional trigger. This determines whether the combined features contain inappropriate content. If so, the compressed image is blocked.

[0131] This solution reduces the time and resources required for image processing while preserving the image's key information. It helps extract image features, identifies key structures within the image, and provides a basis for assigning emotional characteristics. Each image feature is assigned a corresponding emotional label, such as happiness, sadness, or anger. By combining different features and analyzing their combined emotional triggers, it helps identify the overall emotional tendency of the image. Based on the emotional triggers, it decides whether to block the image to protect users from negative emotional influences.

[0132] In some embodiments, image features are analyzed to determine the overall structure of the image; the overall structure of the image is analyzed to determine a feature analysis area; and within the feature analysis area, feature analysis is performed on each image feature.

[0133] The overall structure of an image can be the sum of the relationships between various parts of the image, such as layout, hierarchy, symmetry, and proportion.

[0134] The feature analysis area can be a key part of the image that needs to be feature analyzed, such as a face, limbs, background, or a significant area with prominent colors and obvious textures.

[0135] Specifically, convolutional neural networks (CNNs) or other feature extraction techniques are used to extract features such as color, texture, shape, and edges from the compressed image. Graph neural networks (GNNs) are used to further analyze the extracted features to identify the overall structure of the image. Feature analysis is performed on facial, limb, or other prominent areas. Each feature area is analyzed to determine its image characteristics. These image characteristics are then input into a pre-defined analysis model for feature analysis.

[0136] In a specific implementation, the training method of the analysis model capable of performing feature analysis may refer to the above embodiment, wherein the samples may be adjusted according to the functions to be achieved.

[0137] This solution analyzes image features such as color, shape, and texture to identify the overall structure of an image, including faces, objects, and backgrounds. This helps identify the overall layout and content of an image. Based on the overall image structure, feature areas requiring focused analysis can be identified. For example, in face recognition, the areas of focus for analysis might be the eyes, nose, and mouth. This helps improve the efficiency and accuracy of feature analysis. Detailed feature analysis of each feature analysis area can extract more targeted feature information. For example, analyzing the face area can extract features such as the eyes, nose, and mouth, and further analyze the expressions and shapes of these features. This helps to more accurately analyze and describe image content.

[0138] In some embodiments, image features are analyzed to determine the image characteristics of each image feature; the image features are input into a preset analysis model to obtain feature analysis results; and based on the feature analysis results, the emotional characteristics of each image feature are determined.

[0139] Image features can be representative features of the image, such as color, shape, texture, layout, etc.

[0140] The preset analysis model can be an analysis model that has been pre-trained based on machine learning or deep learning technology, and is used to analyze image characteristics and generate corresponding analysis results.

[0141] The feature analysis result may be information representing the emotional characteristics, such as the emotion label and emotion intensity, outputted after a preset analysis model analyzes the input data, i.e., the image features.

[0142] Specifically, the image features identified, such as color histograms, shape descriptors, and texture features, are computationally quantified to obtain image features. These quantified image features are then fed into a pre-set analysis model. The analysis model is run to infer the image features and obtain feature analysis results. Based on these feature analysis results, the image's emotional labels, such as happiness, sadness, and anger, are predicted. The predicted emotional labels are then evaluated to determine their strength. Based on the emotional label and its strength, each image feature is assigned a corresponding emotional characteristic.

[0143] This solution identifies key features in images, such as color, shape, and texture. These features help identify the image's visual content and style. Using a pre-set analysis model, we conduct in-depth analysis of these image features, obtaining detailed data on their emotional orientation and intensity. Based on the analysis results, each image feature is assigned an emotional label or description, such as happiness, sadness, or anger. These emotional characteristics facilitate sentiment classification and emotional state prediction.

[0144] In some embodiments, based on the combination results, the combination color and combination composition of the random combination feature are determined; the combination color and combination composition are analyzed to determine the characteristic atmosphere of the random combination feature; based on several emotional characteristics of the random combination feature, it is determined whether there is an emotional conflict; if there is an emotional conflict, the emotional triggering situation is determined based on the characteristic atmosphere and several emotional characteristics.

[0145] Random combination features can be a combination of multiple features such as color, shape, texture, layout, etc. randomly selected from the image.

[0146] Composite colors can be a combination of different color features in an image.

[0147] Composition can be the spatial arrangement and layout of various elements such as shapes, lines, objects, etc. in an image.

[0148] Feature atmosphere can be the overall emotional atmosphere or emotional tone created by image features.

[0149] Emotional characteristics can be the emotional properties or emotional labels expressed by image features.

[0150] Emotional conflict can be the existence of two or more contradictory emotional characteristics when the emotional elements in an image oppose each other.

[0151] Specifically, features of random combinations, such as color and shape combinations, are extracted from the image. The extracted features are analyzed to determine their color matching and composition. Based on the color and composition combinations, the image's atmosphere is analyzed for characteristics such as harmony, conflict, and tension. The emotional characteristics assigned to each random combination of features, such as happiness, sadness, and anger, are analyzed. Whether there is a conflict in the emotional characteristics of different features is checked. If an emotional conflict is detected, a determination is made as to whether an emotional change has been triggered based on the feature atmosphere and emotional characteristics.

[0152] This solution analyzes randomly combined features within an image to determine its overall color scheme and composition, helping to identify the image's visual style and layout. Based on the characteristics of the combined colors and composition, it can determine the image's characteristic atmosphere, such as harmony, conflict, and tension, helping to grasp the image's emotional tendencies. By comparing the emotional characteristics of different features, such as the contrast between happiness and sadness, it can determine whether there is emotional conflict within the image, helping to identify inconsistent or contradictory emotions within the image. If emotional conflict is detected, the characteristic atmosphere and emotional characteristics can be combined to determine the specific circumstances of the emotional trigger, such as the degree, type, and direction of emotional evolution.

[0153] In some embodiments, based on the feature analysis results, it is determined whether there is an irrational structure in the compressed image; if there is an irrational structure, the irrational structure is analyzed and the structural attributes are determined; based on the structural attributes, an ethical analysis is performed on the irrational structure; based on the ethical analysis results, it is determined whether to shield the compressed image.

[0154] Irrational structures can be structures or elements that appear in an image that are unconventional or counterintuitive.

[0155] Structural properties can be various features and characteristics of non-physical structures such as size, shape, texture, etc.

[0156] The results of ethical analysis can be the results of ethical risk assessment of irrational structures.

[0157] Specifically, feature analysis results are obtained from the image feature extraction and analysis process. The feature analysis results are analyzed to detect any irrational structures in the image, such as unusual shapes, colors, or unnatural layouts. If irrational structures are detected, their attributes, such as size, position, and texture, are analyzed. Based on the attributes of these irrational structures, ethical risks, such as the potential for violence, terror, or psychological contamination, are assessed. In conjunction with emotional triggers, whether these irrational structures lead to negative emotional reactions is determined. The ethical risk assessment results are combined with the emotional triggers for a comprehensive analysis. Based on the ethical analysis results, a decision is made on whether to block the compressed image.

[0158] This solution captures detailed feature data about compressed images. It identifies structures within the image that are unconventional or expected, potentially containing negative content or ethical issues. It analyzes the size, shape, texture, and other attributes of these unusual structures in detail. This information helps identify the nature and impact of these structures. It assesses the ethical risks posed by these unusual structures, providing an ethical basis for decision-making. Combined with sentiment analysis results, it determines whether these unusual structures trigger emotions such as fear and anger. Combining ethical risk assessment with emotional triggers yields a comprehensive ethical analysis. Based on the ethical analysis results, a decision is made on whether to block the compressed image to ensure that the content adheres to social ethical standards.

[0159] In some embodiments, based on the emotional triggering situation, the random combination features are assigned an emotional value to obtain an emotional score of the random combination features; based on the emotional score and the preset emotional intensity level, it is determined whether the random combination features have inappropriate content; if so, the compressed image is shielded.

[0160] Sentiment assignment can be the process of assigning sentiment labels or scores to features or elements in an image.

[0161] The sentiment score may be a numerical representation that quantifies the sentiment intensity of an image feature.

[0162] The preset emotion intensity level may be a set of standards defined in advance for measuring emotion intensity.

[0163] Inappropriate content can be caused by violent and horrific scenes, inappropriate symbols or images in the image, etc., which cause discomfort or negative emotional reactions in the audience.

[0164] Specifically, sentiment is assigned to the randomly combined features, using a sentiment analysis model to assign a sentiment score to each feature. Based on the sentiment assignment results, a sentiment score is calculated for each randomly combined feature. Three levels of sentiment intensity are set: low, medium, and high. The sentiment scores are compared with the preset sentiment intensity levels to determine whether the randomly combined features contain content inappropriate content. The sentiment scores of different randomly combined features are examined to determine whether there are any sentiment conflicts. A comprehensive assessment of the compressed image is conducted based on the sentiment scores, sentiment conflicts, and ethical risks. Based on the comprehensive assessment results, a determination is made as to whether the compressed image needs to be shielded.

[0165] This solution assigns emotional labels such as happiness, sadness, and anger to randomly combined features. These emotional labels are converted into specific emotional scores to quantify emotional intensity. A standard emotional intensity rating system is established to provide a reference for evaluating emotional scores. By comparing emotional scores with pre-set ratings, content that causes discomfort can be identified. Emotional dissonance between different features in an image, such as the coexistence of happiness and sadness, indicates emotional conflict. The overall emotional impact of an image is evaluated by comprehensively considering the emotional score, emotional conflict, and potential ethical issues. Based on this comprehensive evaluation, a decision is made as to whether the image should be blocked to prevent the spread of inappropriate content.

[0166] In some embodiments, after the compressed image is shielded, the compressed image is analyzed to determine the image type; based on the image type, an online search is performed to obtain an online picture that matches the image type; the image features of the online picture and the compressed image are analyzed to determine the image evolution part; based on the image evolution part, the possible evolution within a preset time period is predicted; based on the evolution possibility, several predicted images are generated and transmitted to a preset analysis model to train the preset analysis model.

[0167] Image types can be classified according to visual features, content themes, or usage of images.

[0168] Networked images can be images retrieved from the Internet that match the compressed image type being analyzed.

[0169] The image evolution portion may be a portion of the image that changes over time.

[0170] The preset time period may be a pre-set time period, and changes of the image within the future time period are predicted.

[0171] Evolution may be based on current image features and time series analysis to predict the possible evolution of the image in the future.

[0172] The predicted image may be an image that may be generated according to the evolution.

[0173] Specifically, the masked compressed image is analyzed to determine image types, such as ordinary photographs, artwork, and advertising images. Based on the determined image type, online images matching that type are retrieved through internet searches. Image features, such as color, shape, and texture, of the online and compressed images are analyzed. Regions of change, i.e., the evolving portions of the images, are identified. Based on these evolving portions of the images, combined with time series analysis, the image evolution trend within a preset time period is predicted. Based on the predicted evolution trend results, several predicted images are generated, simulating future evolution scenarios. These predicted images are then transferred to a preset analysis model for training and optimization.

[0174] Through this solution, the category of the image can be determined by analyzing the compressed image after shielding. By performing online retrieval based on the image type, a large amount of relevant image data can be collected, providing more reference samples for image evolution analysis. By analyzing the image features of online pictures and compressed images, key information that helps to identify the image evolution trend can be extracted. By determining the image evolution part, it is helpful to identify the area where the image changes over time, providing a basis for predicting future evolution trends. Predicting the possible evolution within a preset time period can identify the changing trend of image content in advance, which is helpful for content review and sentiment analysis. Generating predicted images can simulate possible future image states and provide forward-looking information for content review and sentiment analysis. By training the preset analysis model, the model's ability to predict future image evolution trends can be improved, and the model's accuracy can be enhanced.

[0175] In some embodiments, the networked pictures are analyzed to determine the networked image features; the networked image features and image features are analyzed to determine the same features; based on the same features, the feature differences of the remaining features are determined; based on the feature differences, the image evolution part is determined.

[0176] Internet image features can be a series of attributes such as color, shape, texture, size, etc. extracted from images obtained from the Internet.

[0177] The same features may be common attributes present in the networked image and the compressed image.

[0178] The remaining features may be attributes unique to each of the networked image and the compressed image in addition to the same features.

[0179] Feature distinction may be the difference between features in the networked image and the compressed image.

[0180] Specifically, image recognition technology based on deep learning models is used to extract features such as color, shape, and texture from online images. Image recognition technology is also used to extract corresponding features from compressed images for comparison with the features of the online image. The features of the online and compressed images are compared to identify common features. The differences between the common features in the online and compressed images are analyzed. Based on these feature differences, the evolutionary trends of the image features are inferred. Based on these evolutionary trends, regions of the image that have changed are identified, representing the areas of image evolution.

[0181] This solution analyzes online images to obtain a rich set of image features. By comparing the features of online and compressed images, shared features can be identified, helping to identify similarities and differences between images. After identifying common features, it becomes easier to identify and distinguish differences between remaining features, including changes in feature intensity, position, or context, which helps reveal details of image evolution. By analyzing feature differences, it is possible to identify areas of change in the image, such as increases or decreases in feature intensity or changes in feature position. These areas are key components of image evolution.

[0182] In some embodiments, the networked images are analyzed to determine the upload time; based on the upload time, the networked images are sorted in time series to obtain a number of ordered images; for each ordered image, the networked image features and image features are analyzed based on the time series to determine the same features; based on the feature differences, the image evolution part is determined; based on the time series, the feature differences are analyzed to determine the image evolution part.

[0183] The upload time may be a specific time point when the image is uploaded to the Internet.

[0184] Online images can be image files such as photos, illustrations, and graphics stored on the Internet.

[0185] A time series can be a collection of data points arranged in chronological order.

[0186] An ordered image may be a collection of images arranged in chronological order.

[0187] Specifically, relevant online images are retrieved and obtained from the internet. Upload time information is extracted from the original image data. Based on the extracted upload time, the online images are sorted in time series to obtain an ordered image set. Image processing and computer vision techniques are used to extract image features such as color, texture, and shape for each ordered image. The extracted features are converted into feature vectors or matrices. Image features at different time points are compared to identify common features. The differences between the same features at different time points are analyzed to determine how the features change over time. Based on the feature differences, the parts of the image that evolve over time are identified. Based on the time series, the feature differences are analyzed to determine the evolution trend of the image features.

[0188] This solution determines the upload time of each online image, providing a timeline for time series analysis and image evolution tracking. Images are sorted chronologically to create a time series, facilitating the observation and analysis of image changes over time. By analyzing the features of each image, we can identify features between images at different time points. These features can serve as reference points for image evolution analysis. By comparing the differences between these features, we can identify the parts of the image that evolve over time, revealing the dynamic changes in the image. By analyzing feature differences over time, we can more accurately track the evolution of image features.

[0189] Figure 3 This is a structural diagram of an emotion dynamic prediction system based on graph neural network and time series provided in one embodiment of the present application, such as Figure 3 As shown, the emotion dynamic prediction system 300 based on graph neural network and time series of this embodiment includes: a feature extraction module 301, a feature analysis module 302, an emotion trigger analysis module 303, and a shielding judgment module 304.

[0190] The feature extraction module 301 is used to obtain and compress the image to be analyzed to obtain a compressed image, and perform feature extraction on the compressed image to obtain image features;

[0191] A feature analysis module 302 is configured to analyze each image feature and assign an emotional characteristic based on the feature analysis result;

[0192] The emotion trigger analysis module 303 is used to randomly combine at least two image features and determine the emotion trigger situation based on the combination result and the emotion characteristics;

[0193] The shielding determination module 304 is configured to determine whether to shield the compressed image according to the emotion triggering condition.

[0194] Optionally, the feature analysis module 302 is configured to:

[0195] Analyzing the image features to determine the overall image structure;

[0196] Analyzing the overall structure of the image and determining a feature analysis area;

[0197] For the feature analysis area, feature analysis is performed on each image feature.

[0198] Optionally, when assigning emotional characteristics based on the feature analysis results, the feature analysis module 302 is used to:

[0199] Analyzing the image features to determine image characteristics of each image feature;

[0200] Inputting the image features into a preset analysis model to obtain feature analysis results;

[0201] According to the characteristic analysis results, the emotional characteristics of each image feature are determined.

[0202] Optionally, when the emotion trigger analysis 303 module determines the emotion trigger situation based on the combination result and the emotion characteristics, it is used to:

[0203] Determining the combined color and combined composition of the random combination features according to the combination result;

[0204] Analyzing the combined colors and the combined composition to determine a characteristic atmosphere of the random combination feature;

[0205] Determine whether there is an emotional conflict based on several emotional characteristics of the random combination features;

[0206] If there is an emotional conflict, the emotional triggering situation is determined based on the characteristic atmosphere and the several emotional features.

[0207] Optionally, the emotion dynamics prediction system 300 based on graph neural network and time series further includes a structure analysis module 305 for:

[0208] determining whether an abnormal structure exists in the compressed image according to the feature analysis result;

[0209] If an irrational structure exists, analyzing the irrational structure and determining the structural properties;

[0210] Conduct ethical analysis on the irrational structure based on the structural attributes;

[0211] According to the result of the ethics analysis, it is determined whether to shield the compressed image.

[0212] Optionally, when the shielding determination module 304 determines whether to shield the compressed image according to the emotion triggering condition, it is configured to:

[0213] Assigning an emotion to the random combination feature according to the emotion triggering situation to obtain an emotion score for the random combination feature;

[0214] determining whether the random combination feature has content inappropriateness based on the emotion score and a preset emotion intensity level;

[0215] If it exists, the compressed image is masked.

[0216] Optionally, the graph neural network and time series based emotion dynamics prediction system 300 further includes a model training module 306 for:

[0217] After shielding the compressed image, analyzing the compressed image to determine the image type;

[0218] Perform an online search based on the image type to obtain online pictures that match the image type;

[0219] Analyzing image features of the online image and the compressed image to determine an image evolution portion;

[0220] Predicting the evolution possibility within a preset time period based on the image evolution portion;

[0221] According to the evolution possibility, a number of prediction images are generated and transmitted to the preset analysis model to train the preset analysis model.

[0222] Optionally, the model training module 306 analyzes the image features of the online image and the compressed image to determine the image evolution portion, and is used to:

[0223] Analyzing the online picture to determine online image features;

[0224] Analyzing the network image features and the image features to determine common features;

[0225] Determine the characteristic differences of the remaining characteristics based on the same characteristics;

[0226] According to the feature distinction, the image evolution portion is determined.

[0227] Optionally, the model training module 306 analyzes the network image features and the image features, and when identical features are determined, is used to:

[0228] Analyze the online picture to determine the upload time;

[0229] Sort the networked images in time series according to the upload time to obtain a plurality of ordered images;

[0230] For each ordered image, analyzing the network image features and the image features according to the time sequence to determine the same features;

[0231] When determining the image evolution portion based on the feature distinction, it is used to:

[0232] Based on the time series, the feature distinctions are analyzed to determine the image evolution portion.

[0233] The system of this embodiment can be used to execute the method of any of the above embodiments. Its implementation principles and technical effects are similar and will not be described in detail here.

Claims

1. A method for predicting emotional dynamics based on graph neural networks and time series, characterized by: include: Acquiring and compressing an image to be analyzed to obtain a compressed image, and performing feature extraction on the compressed image to obtain image features; For each image feature, analyzing the image feature, and assigning an emotional characteristic based on the feature analysis result; Randomly combining at least two image features, and determining an emotional triggering situation based on the combination result and the emotional characteristics; determining whether to shield the compressed image according to the emotional triggering situation; The emotional characteristics are assigned according to the feature analysis results, including: Analyzing the image features to determine image characteristics of each image feature; Inputting the image features into a preset analysis model to obtain feature analysis results; Determining the emotional characteristics of each image feature based on the characteristic analysis results; After shielding the compressed image, analyzing the compressed image to determine the image type; Perform an online search based on the image type to obtain online pictures that match the image type; Analyzing image features of the online image and the compressed image to determine an image evolution portion; Predicting the evolution possibility within a preset time period based on the image evolution portion; generating a plurality of predicted images according to the evolution possibility, and transmitting the predicted images to the preset analysis model to train the preset analysis model; The step of analyzing each image feature includes: Applying a graph neural network to analyze the image features and determine the overall structure of the image; Analyzing the overall structure of the image and determining a feature analysis area; For the feature analysis area, performing feature analysis on each image feature; The analyzing the image features of the online picture and the compressed image to determine the image evolution portion includes: Analyzing the online picture to determine online image features; Analyzing the network image features and the image features to determine common features; Determine the characteristic differences of the remaining characteristics based on the same characteristics; determining an image evolution portion based on the feature distinction; The analyzing the network image features and the image features to determine the same features includes: Analyze the online picture to determine the upload time; Sort the networked images in time series according to the upload time to obtain a plurality of ordered images; For each ordered image, analyzing the network image features and the image features according to the time sequence to determine the same features; The determining of the image evolution portion based on the feature distinction includes: Analyzing the characteristic differences based on the time series to determine the image evolution portion; Determining the emotional triggering situation based on the combination result and the emotional characteristics includes: Determining the combined color and combined composition of the random combination features according to the combination result; Analyzing the combined colors and the combined composition to determine a characteristic atmosphere of the random combination feature; Determine whether there is an emotional conflict based on several emotional characteristics of the random combination features; If there is an emotional conflict, the emotional triggering situation is determined based on the characteristic atmosphere and the several emotional features.

2. The method according to claim 1, characterized in that Before determining whether to shield the compressed image according to the emotional triggering situation, the method further includes: determining whether an abnormal structure exists in the compressed image according to the feature analysis result; If an irrational structure exists, analyzing the irrational structure and determining the structural properties; Conduct ethical analysis on the irrational structure based on the structural attributes; According to the result of the ethics analysis, it is determined whether to shield the compressed image.

3. The method according to claim 2, characterized in that The determining whether to shield the compressed image according to the emotion triggering situation includes: Assigning an emotion to the random combination feature according to the emotion triggering situation to obtain an emotion score for the random combination feature; determining whether the random combination feature has content inappropriateness based on the emotion score and a preset emotion intensity level; If it exists, the compressed image is masked.

4. A system for predicting emotional dynamics based on graph neural networks and time series, applied to the method according to any one of claims 1 to 3, characterized in that: include: A feature extraction module is used to obtain and compress the image to be analyzed to obtain a compressed image, and perform feature extraction on the compressed image to obtain image features; A feature analysis module is used to analyze each image feature and assign emotional characteristics based on the feature analysis results; An emotion trigger analysis module, configured to randomly combine at least two image features and determine an emotion trigger condition based on the combination result and the emotion characteristics; The shielding judgment module is used to determine whether to shield the compressed image according to the emotional triggering situation.

Images