Group collaborative music sharing and emotion synchronizing system

Abstract

The invention discloses a group collaborative music sharing and emotion synchronization system, which belongs to the technical field of information retrieval and recommendation systems and comprises a multi-source data acquisition module, a group emotion state fusion module, a music sequence collaborative generation module and a feedback learning and adaptive optimization module. Physiological signals and social interaction data of group members are collected, a group emotion influence relation graph is constructed based on a graph neural network, a unified emotion vector is generated through fusion, the collaboration degree of music and group emotion is calculated, an optimal music sequence is selected, real-time monitoring and feedback are conducted, and system parameters are optimized in a closed-loop mode. Accurate recognition, effective fusion, collaborative optimization and continuous improvement of group emotions are achieved, the problem that music recommendation in a group scene is difficult to balance multi-member emotion demands is solved, and the method is suitable for multi-person scenes such as family parties, team cooperation and social activities.

Description

Technical Field

[0001] This invention relates to the field of information retrieval and recommendation system technology, and in particular to a group collaborative music sharing and emotion synchronization system, which can be applied to music recommendation and emotion management in multi-person scenarios such as family gatherings, team collaborations, and social activities. Background Technology

[0002] With the rapid development of digital music platforms and the widespread application of intelligent recommendation technology, music recommendation systems have become an indispensable service in people's daily lives. Traditional music recommendation systems are mainly geared towards individual users, recommending suitable music by analyzing a user's historical playback records, music preferences, or current emotional state. Existing single-user music recommendation technology is relatively mature and can well meet individual listening needs.

[0003] However, in real life, people often enjoy music together in group settings, such as family gatherings where everyone listens to music together, teamwork with background music, or friends sharing music at parties. In these group settings, different members often have different musical preferences and emotional states, making it a significant challenge to select music that meets the needs of all members. Traditional single-user music recommendation methods cannot effectively handle the diverse needs in group settings, and simply playing different members' favorite music in turn is unlikely to create a good shared experience.

[0004] Existing technology CN106202073A discloses a music recommendation method and system. This method analyzes music in a music library to determine its emotional characteristics, identifies users' emotional states to determine their own emotional characteristics, and matches these user emotional characteristics with the music's emotional characteristics to generate a music recommendation list. While this system can effectively achieve music recommendation based on a single user's emotional state, it has the following shortcomings: First, this technology only performs emotional analysis and music recommendation for a single user, failing to handle the different emotional states of multiple members in a group setting, and thus cannot balance the emotional needs of each member when multiple people listen to music together. Second, the emotional recognition method used in this technology is mainly based on users' social application signatures or touchscreen operations. These data sources are difficult to accurately reflect the real-time emotional state of each member in a group setting, and cannot capture the emotional interactions and influences between members. Third, this technology only considers a simple match between user emotions and music emotions, lacking modeling of the dynamic evolution of group emotions and the influence of emotions between members, thus failing to achieve the coordination and synchronization of group emotions. Fourth, this technology does not consider the feedback effect of music playback on group emotions, lacks a closed-loop optimization mechanism, and is difficult to continuously improve the recommendation effect.

[0005] Group music recommendation differs fundamentally from single-user recommendation. In a group setting, the emotional states of each member influence one another; a change in one member's emotion can resonate with or adjust the emotions of others. Furthermore, music, as a vehicle for emotion, can facilitate emotional transmission and synchronization among group members. Therefore, a group music recommendation system needs to consider not only the individual emotions of each member but also the emotional relationships between them, selecting music that promotes emotional synergy and synchronization within the group to create a positive shared music experience.

[0006] Currently, the main technical challenges in the field of group recommendation are as follows: First, how to accurately acquire and integrate the emotional data of multiple members. Existing emotion recognition technologies are mostly designed for single users and are difficult to extend to group scenarios. Second, how to model the emotional influence relationship among group members. Traditional collaborative filtering methods cannot capture the dynamic emotional interaction among members. Third, how to select music sequences that can promote group emotional collaboration. Simple voting or averaging methods often result in recommendation results that do not satisfy anyone. Fourth, how to achieve closed-loop optimization of music recommendation and group emotional feedback to continuously improve recommendation performance.

[0007] Therefore, there is an urgent need for a new technological solution to address the above problems, enabling music recommendation and emotion synchronization for group scenarios, and meeting the needs of multiple people sharing music. Summary of the Invention

[0008] To address the shortcomings of existing technologies, the purpose of this invention is to provide a collaborative music sharing and emotion synchronization system for groups. By acquiring multi-source data, modeling the influence of group emotions using graph neural networks, generating collaborative music sequences, and optimizing closed-loop feedback, this system solves the problem of balancing the emotional needs of multiple members in music recommendations within a group setting, achieving collaborative and synchronized group emotions and enhancing the shared music experience.

[0009] To achieve the above objectives, the technical solution adopted by the present invention is: a group collaborative music sharing and emotion synchronization system, including a multi-source data acquisition module, a group emotion state fusion module, a music sequence collaborative generation module, and a feedback learning and adaptive optimization module.

[0010] The multi-source data acquisition module is used to collect physiological signal data, social interaction data, and environmental perception data of each member in the group, providing multi-dimensional information sources for subsequent emotion recognition.

[0011] The group emotion state fusion module constructs a group emotion influence relationship graph based on a graph neural network. It uses group members as graph nodes and the intensity of emotion influence between members as edge weights. Through multi-layer graph convolution propagation operations, it fuses the emotional features of group members to generate a unified emotion vector that can represent the overall emotional state of the group.

[0012] The music sequence collaborative generation module calculates the degree of synergy between each candidate music and the group's emotions through an attention mechanism based on the unified emotion vector of the group and the music emotion attribute library, and selects music sequences that can promote the synergy of group emotions.

[0013] The feedback learning and adaptive optimization module monitors the emotional changes of group members in real time during music playback, evaluates the collaborative effect of music sequences based on a reinforcement learning framework, feeds the evaluation results back to the group emotional state fusion module to update the graph neural network parameters, and feeds them back to the music sequence collaborative generation module to adjust the music selection strategy, forming a complete closed-loop optimization system.

[0014] Compared with the prior art, the present invention has the following beneficial effects: First, this invention breaks through the limitations of traditional single-user music recommendation and innovatively proposes a music sharing and emotion synchronization technology for groups. By collecting physiological signals, social interactions and environmental perception data of each member in the group through multi-source data collection, it provides a rich data foundation for group emotion analysis. Compared with existing emotion recognition methods based only on social signatures or touch operations, it can more accurately and comprehensively capture the real-time emotional state of group members.

[0015] Second, this invention uses graph neural networks to model the emotional influence relationships among group members, mapping group members as graph nodes and emotional influence intensity as edge weights. Through the propagation mechanism of multi-layer graph convolutional networks, it learns the emotional propagation patterns and influence relationships among members. Compared with traditional collaborative filtering methods that only consider user similarity, this invention can explicitly model the dynamic emotional interaction among members, and better capture the correlation and propagation of group emotions.

[0016] Third, this invention innovatively proposes a method for calculating the degree of group emotional synergy, which comprehensively considers three dimensions: compliance, intensity difference, and guidance. It not only meets the current emotional needs of the group, but also promotes the transition of group emotions to the target state through the emotional guidance effect of music. Compared with the simple emotion matching method of the prior art, this invention can select music sequences with more synergistic effects to achieve the synchronization and convergence of group emotions.

[0017] Fourth, this invention constructs a complete closed-loop feedback optimization system. By monitoring the emotional changes of group members in real time during music playback, it evaluates the synergistic effect of music sequences based on a reinforcement learning framework, and feeds the evaluation results back to the graph neural network parameter update and music selection strategy optimization, thereby achieving adaptive adjustment and continuous improvement of the system. Compared with the lack of feedback mechanism in existing technologies, this invention can continuously improve recommendation accuracy and group satisfaction.

[0018] Fifth, this invention achieves a non-linear synergy greater than the sum of its parts through deep coupling and collaborative optimization among the modules. The multi-source data acquisition module provides multi-dimensional data input for emotion fusion, the output of the group emotion state fusion module directly serves as the key basis for music sequence generation, and the playback effect of the music sequence influences the parameters and strategies of the preceding modules through the feedback module, forming a complete closed loop from data acquisition to emotion fusion to music generation and then to feedback optimization. The modules promote each other and have a synergistic effect, and the overall system performance is significantly better than the sum of the effects of each module running independently.

[0019] Sixth, this invention is applicable to various group scenarios such as family gatherings, team collaborations, and social activities. It can adaptively adjust the emotion integration strategy and music selection preferences according to the characteristics of different scenarios. Compared with the single recommendation strategy of existing technologies, this invention has stronger scenario adaptability and practical value, and can be widely applied in smart homes, office environments, entertainment venues and other fields to provide more intelligent and considerate music services for groups. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall architecture of the group collaborative music sharing and emotion synchronization system of the present invention.

[0021] Figure 2 This is a schematic diagram of the structure of the multi-source data acquisition module of the present invention.

[0022] Figure 3 This is a schematic diagram of the workflow of the group emotional state fusion module of the present invention.

[0023] Figure 4 This is a schematic diagram of the workflow of the music sequence collaborative generation module of the present invention.

[0024] Figure 5 This is a schematic diagram of the workflow of the feedback learning and adaptive optimization module of this invention. Detailed Implementation

[0025] Please refer to the attached document. Figures 1-5 The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0026] Reference Figure 1 The group collaborative music sharing and emotion synchronization system provided by this invention includes: a multi-source data acquisition module 1, a group emotion state fusion module 2, a music sequence collaborative generation module 3, and a feedback learning and adaptive optimization module 4. These four core modules are deeply coupled to form a complete closed-loop collaborative system, realizing accurate identification, effective fusion, collaborative optimization, and continuous improvement of group emotions.

[0027] Reference Figure 2 The multi-source data acquisition module 1 is responsible for collecting multi-dimensional data from each member of the group, providing a data foundation for subsequent emotion recognition and analysis. This module includes a physiological signal acquisition unit, a social interaction acquisition unit, an environmental perception acquisition unit, and a data preprocessing unit.

[0028] The physiological signal acquisition unit collects physiological signal data from group members via wearable devices, including electroencephalogram (EEG), heart rate (HR), and conductance on skin (GSR). These physiological signals objectively reflect the emotional state of the members, are not subject to conscious control, and have high reliability. In one embodiment of the invention, a wireless EEG acquisition device is used to acquire EEG signals from the user's forehead region at a sampling frequency of 256 Hz; a smart bracelet or smartwatch is used to acquire heart rate signals at a sampling frequency of 1 Hz; and a wristband sensor is used to acquire conductance on skin signals at a sampling frequency of 4 Hz.

[0029] The social interaction acquisition unit collects dialogue content, facial expressions, and body movements among group members. Dialogue content is captured via a microphone array and converted into text, then analyzed using natural language processing technology to assess emotional inclination. Facial expressions are captured by a camera, and facial emotion recognition algorithms are used to extract facial features. Body movements are captured by a depth camera or posture recognition sensor, analyzing the activity level and emotional expression of the movements.

[0030] The environmental perception and acquisition unit collects environmental parameters of the current scene, including scene type (family gathering, team collaboration, social activities, etc.), ambient noise level, and light intensity. These environmental parameters can provide contextual information for emotion analysis and help determine the emotional state of group members. Preferably, the scene type is selected manually by the user or automatically identified based on location information, ambient noise is measured by a sound level meter, and light intensity is measured by a light sensor.

[0031] The data preprocessing unit performs filtering, noise reduction, normalization, and time alignment on the acquired multi-source data. For EEG signals, a bandpass filter is used to remove noise below 0.5Hz and above 50Hz, and baseline drift correction is performed. For heart rate signals, median filtering is used to remove outliers. For electrodermal signals, low-pass filtering is used to smooth signal fluctuations. Normalization maps different types of data to a uniform numerical range, facilitating subsequent fusion analysis. Time alignment ensures that data acquired by different devices remain synchronized on the time axis, with sampling timestamp errors controlled within 50ms.

[0032] Reference Figure 3The group emotion state fusion module 2 is one of the core innovations of this invention. This module constructs a group emotion influence relationship graph based on a graph neural network, and fuses the emotional features of group members through multi-layer graph convolution operations to generate a unified emotion vector representing the overall emotional state of the group. This module includes a graph construction unit, a multi-layer graph convolution unit, an attention weighting unit, and an emotion vector fusion unit.

[0033] The graph construction unit maps group members to graph nodes and constructs edge connections based on the strength of social relationships and the frequency of historical interactions among members. Specifically, for the group... Each member constructs a Undirected graph of nodes ,in Represents a set of nodes. This represents a set of edges. The existence and weight of an edge are determined by the strength of the social relationship between the members. If two members have a social relationship (such as family members, colleagues, friends, etc.), an edge is established between the corresponding nodes, and the edge weight is specified. Member and members The strength of the emotional influence between members. In a preferred embodiment, the edge weight is calculated based on the intimacy of the social relationship between members and the frequency of historical interactions. The intimacy is preset by the user or extracted from social network data, and the interaction frequency is obtained by statistically analyzing the number of historical conversations and the duration of interactions. The edge weight value ranges from 0 to 1.

[0034] Multi-layer graph convolutional units aggregate the features of each node's neighboring nodes through a multi-layer graph convolutional network (GCN) to learn the emotion propagation pattern among group members. The core idea of ​​graph convolution operation is to aggregate the features of each node with the features of its neighboring nodes, thereby capturing the relationships between nodes. In this invention, the following innovative graph convolution update formula is adopted: , in, For the first The node feature matrix of the layer has a dimension of , For the number of nodes, For the first The feature dimensions of the layer; For the first The node feature matrix of the layer; To add the adjacency matrix after adding self-loops, This is the original adjacency matrix. It is the identity matrix; for The degree matrix, where ; For the first The learnable weight matrix of the layer has dimensions of ; For the activation function, LeakyReLU activation function is used; the last term Residual connections are used to preserve original node information and prevent oversmoothing. This formula achieves weighted aggregation of neighbor features through a symmetrically normalized Laplacian matrix, while maintaining the node's own features through residual connections, effectively avoiding information loss in deep networks.

[0035] In practical applications, a three-layer graph convolutional network is used. The input feature dimension of the first layer is 128 (emotional features extracted from multi-source data), and the feature dimensions of the second and third layers are 64 and 32, respectively. Through multi-layer stacking, the system can capture the multi-hop emotion propagation relationship between members, that is, a member's emotion is not only affected by directly connected members, but also by indirectly connected members.

[0036] The attention-weighted unit assigns different weights to different members based on an attention mechanism, highlighting the influence of core members on group emotions. Within a group, different members have varying degrees of influence on group emotions. For example, in a family gathering, the emotional state of elders often significantly impacts the overall atmosphere, while in team collaboration, the team leader's emotions may affect other members. This invention employs a self-attention mechanism to calculate the influence weight of each member, and the innovative attention calculation formula is as follows: , in, For members Attention weights Members after graph convolution eigenvectors, The attention weight matrix is ​​a learnable matrix. For bias terms, This represents the total number of group members. The attention weights are normalized using a softmax function to ensure that the sum of the weights for all members is 1. In a preferred embodiment, the attention weights also consider the social roles of members; for example, increasing the weight of elders in a family setting and increasing the weight of leaders in a team setting.

[0037] The emotion vector fusion unit fuses the weighted emotion features of all members to generate a unified emotion vector representing the overall emotional state of the group. The fusion operation uses a weighted summation method, and the fusion formula is as follows: , in, To create a unified emotional vector for the group, For members Attention weights For members The emotion feature vector, The total number of group members. This unified emotion vector comprehensively considers the emotional state and influence of all members, and can fully reflect the overall emotion of the group. In a preferred embodiment of the present invention, the unified group emotion vector adopts a three-dimensional representation, including three dimensions: pleasure, activity, and dominance, which correspond to the three axes of the Pleasure-Arousal-Dominance (PAD) emotional space model, and the values ​​range from -1 to +1.

[0038] The group emotion state fusion module leverages the powerful modeling capabilities of graph neural networks to not only consider the individual emotions of each member but also explicitly model the emotional influence relationships between members. Compared to simple emotion averaging or voting methods, it can more accurately capture the overall emotional state of the group. Furthermore, the introduction of an attention mechanism enables the system to identify core members within the group and assign them higher weights, which aligns with the dynamic characteristics of group emotions in reality.

[0039] Reference Figure 4 The music sequence collaborative generation module 3 selects music sequences that can promote group emotional collaboration based on the group's unified emotion vector and music emotion attribute library. This module includes a music emotion modeling unit, a collaboration degree calculation unit, a sequence optimization unit, and a conflict mediation unit.

[0040] The music emotion modeling unit extracts emotion feature vectors from a music emotion attribute database of candidate songs. This database pre-stores a large number of music emotion features, obtained through audio signal analysis and lyric sentiment analysis. Specifically, for each song, audio features are extracted, including acoustic features such as pitch, rhythm, loudness, and timbre; lyric features are extracted, including textual features such as lexical sentiment and semantic themes. These features are then mapped to the PAD emotion space using an emotion mapping algorithm to obtain the music's emotion feature vector. ,in Indicates level of pleasure. Indicates activity level. The degree of dominance is indicated. In a preferred embodiment, the sentiment mapping algorithm uses a convolutional neural network to extract features from the audio signal, employs a BERT model to perform sentiment analysis on the lyrics text, and finally maps the multimodal features to the PAD sentiment space through a fully connected layer.

[0041] The synergy calculation unit calculates the synergy between the unified emotion vector of the group and the emotion feature vector of each candidate piece of music. Synergy is an innovative concept in this invention. Unlike simple similarity matching, synergy comprehensively considers three dimensions: compliance, intensity difference, and guidance, aiming to select music that both matches the current emotion of the group and promotes emotional synergy. The innovative synergy calculation formula is as follows: , in, To score the degree of synergy, To adapt, For the degree of intensity difference, To guide the degree, , , Let be the weighting coefficient, satisfying In the preferred embodiments, the values ​​are 0.5, 0.2, and 0.3, respectively.

[0042] compliance The formula for measuring the directional consistency between musical emotion and collective emotion is as follows: , in, To create a unified emotional vector for the group, For musical emotion feature vectors, Represents the vector dot product. This represents the Euclidean norm of a vector. The formula essentially calculates the cosine similarity between two vectors, with a value ranging from -1 to +1. A larger value indicates a greater alignment between the musical sentiment and the overall collective sentiment.

[0043] Strength difference The formula for measuring the similarity between the intensity of musical emotion and the intensity of collective emotion is as follows: , in, Let be the magnitude of the group emotion vector, representing the intensity of the group emotion. Let be the magnitude of the musical emotion vector, representing the intensity of the musical emotion. This formula calculates the relative difference between two intensities, with a value ranging from 0 to 1. A larger value indicates that the intensity of the musical emotion is closer to the intensity of the collective emotion.

[0044] Guiding degree The formula for measuring the positive guiding effect of music on group emotions is as follows: , in, Let be the target emotion vector, representing the ideal state to which the group's emotions are expected to be guided. In a preferred embodiment, the target emotion vector is set to a positive and pleasant emotional state, i.e. , representing high pleasure, moderate activity, and moderate to high dominance. This formula calculates the consistency between the direction of the music's emotional shift relative to the group's current emotional state and the target direction, with a value ranging from -1 to +1. A larger value indicates that the music can better guide the group's emotions toward the target state.

[0045] The sequence optimization unit uses a dynamic programming algorithm to select music sequences from candidate music that can smoothly guide the emotional transition of the group. In practical applications, music playback is a sequence decision process that needs to consider the emotional transition between multiple consecutively played songs. The goal of sequence optimization is to maximize the cumulative coherence of the entire sequence while ensuring a smooth emotional transition between adjacent songs. The state transition equation of the dynamic programming is as follows: , in, Indicates at time Select music Cumulative synergy over time For a moment A collection of candidate music, For a moment The collective emotion vector For music The degree of alignment with the current group sentiment For music and The emotional distance between them indicates the intensity of the emotional transition. The smoothness penalty coefficient is set to 0.3 in the preferred embodiment. This equation maximizes the degree of cooperation while mitigating the impact of the penalty term. Constrain the emotional differences between adjacent musical pieces to ensure a smooth emotional transition.

[0046] The conflict resolution unit is used to handle situations where there are significant differences in the emotional states of group members. When the emotional states of group members differ significantly, simple emotional fusion may result in recommended music failing to satisfy anyone. The conflict resolution unit determines the degree of emotional difference by analyzing the variance of the group's emotional vector. When the variance exceeds a preset threshold, the system selects music with an emotional neutralizing effect, i.e., music whose emotional characteristics are located in the neutral region of the PAD space, to promote the convergence of group emotions. In a preferred embodiment, the variance threshold is set to 0.3, and the neutral emotional region is defined as the area where the pleasantness level is between -0.2 and +0.2, the activity level is between 0 and 0.5, and the dominance level is between 0.3 and 0.7. By playing music with neutral emotions, the emotional differences among group members can be gradually reduced, creating conditions for subsequent emotional collaboration.

[0047] The music sequence collaborative generation module, through innovative collaborative degree calculation methods and sequence optimization algorithms, can not only select music that matches the current mood of the group, but also promote the transition of the group's mood to a positive state through the emotional guidance effect of music. At the same time, it handles situations where there are large differences in the emotions of members through conflict mediation mechanisms, thus realizing intelligent and humanized music recommendation.

[0048] Reference Figure 5The feedback learning and adaptive optimization module 4 monitors the emotional changes of group members in real time during music playback, evaluates the synergistic effect of the music sequence, and feeds the evaluation results back to the front-end module for parameter updates and strategy optimization, forming a complete closed-loop optimization system. This module includes an emotion change tracking unit, a synergistic effect evaluation unit, a parameter update unit, and a strategy optimization unit.

[0049] The emotion change tracking unit monitors the emotional changes of group members in real time during music playback. During music playback, the multi-source data acquisition module 1 continuously collects physiological signals and social interaction data from group members. The emotion change tracking unit analyzes this data, extracts the real-time emotion vector for each member, and calculates the emotion offset and rate of change relative to before playback. The emotion offset is defined as the Euclidean distance between the current emotion vector and the emotion vector before playback, and the rate of change is defined as the derivative of the emotion offset with respect to time. These indicators can quantify the degree and speed of the music's influence on the emotions of group members.

[0050] The collaborative effect evaluation unit assesses the collaborative effect of music sequences based on a reinforcement learning framework, using group emotional convergence as the reward signal. In this framework, the music recommendation system is modeled as an agent, the group emotional state as the environment state, music selection as the action, and the collaborative effect as the reward. The collaborative effect evaluation unit defines an innovative reward function that comprehensively considers three aspects: emotional convergence, emotional guidance effect, and member satisfaction. , in, The total reward value, Rewards for emotional convergence. Reward for the effect of emotional guidance. Rewards for member satisfaction , , The weighting coefficients are 0.4, 0.3, and 0.3 in the preferred embodiments.

[0051] Emotional Convergence Reward The formula for measuring the degree of emotional convergence among group members after music playback is as follows: , in, Let Variance be the variance of the emotion vector of group members before the music is played. This represents the variance after the music was played. If the variance decreases, it indicates that the members' emotions are becoming more aligned, and a positive reward should be given; if the variance increases, it indicates that the members' emotions are becoming more dispersed, and a negative reward should be given.

[0052] Emotional guidance effect reward The formula for measuring the degree to which a group's emotions shift toward a target emotional state is: , in, The distance between the unified emotion vector of the group and the target emotion vector before the music is played. This represents the distance after the music has played. If the distance decreases, it indicates that the group's emotions are moving closer to the target state, and a positive reward is given; if the distance increases, it indicates that the group's emotions are deviating from the target state, and a negative reward is given.

[0053] Member satisfaction reward Based on the subjective feedback of group members, a simple rating button allows members to rate the currently playing music. The rating range is 1 to 5 points, and the satisfaction reward is the average of all members' ratings minus 3, with a value range of -2 to +2.

[0054] The parameter update unit updates the network weights and attention parameters of the graph neural network using gradient descent based on the collaborative effect evaluation results. Specifically, it updates the total reward value... The negative value of the loss function is used to calculate the gradient of the loss with respect to the graph neural network parameters via backpropagation, and then the parameters are updated using the Adam optimizer. The parameter update formula is: , in, This represents all learnable parameters of the graph neural network (including the weight matrix of the graph convolutional layer, the attention weight matrix, etc.). For loss function, The learning rate is 0.001 in the preferred embodiment. The loss function represents the gradient of the parameters. Through continuous parameter updates, graph neural networks can continuously optimize the fusion strategy of group emotions, more accurately capturing the emotional influence relationships among members.

[0055] The strategy optimization unit adjusts the synergy weighting coefficients in the music selection strategy to optimize the generation of future music sequences. In the music sequence synergy generation module 3, the synergy calculation involves three weighting coefficients: compliance, intensity difference, and guidance. , , These coefficients determine music selection preferences. The strategy optimization unit dynamically adjusts these weight coefficients based on the synergy evaluation results. If emotional convergence is rewarded... A higher compliance rate indicates that the current strategy effectively promotes convergence of group sentiment, maintaining or increasing the compliance weight. If the emotional guidance effect is rewarded A higher value indicates that the emotion guidance strategy is effective, thus increasing the weight of the guidance level. The weight coefficients are updated using a soft update strategy, with each adjustment not exceeding 0.1 to ensure the stability of the strategy.

[0056] The feedback learning and adaptive optimization module constructs a complete closed-loop feedback system through real-time emotion monitoring, scientific effect evaluation, and automatic parameter optimization. The output of this module directly influences the work of the group emotion state fusion module and the music sequence collaborative generation module, forming a closed-loop process from data acquisition to emotion fusion to music generation and then to feedback optimization, thus realizing the system's adaptive learning and continuous improvement.

[0057] In a preferred embodiment of the invention, the system further includes a scene adaptation module, used to identify the current application scenario type and adjust the group emotion fusion strategy and music selection preferences according to the scenario type. Different application scenarios have different requirements for music recommendations, and the scene adaptation module enables the system to better adapt to actual applications.

[0058] Scene types are categorized into three main types: family gatherings, team collaborations, and social activities. Scene recognition can be achieved through manual user selection, automatic recognition based on location information, or intelligent recognition based on environmental parameters. In a preferred embodiment, an intelligent recognition method based on environmental parameters is used, which analyzes features such as the age distribution of group members, social relationship types, and environmental noise levels, and then uses a support vector machine classifier to automatically identify the scene type.

[0059] For different scenarios, the scenario adaptation module adopts different emotion fusion strategies: For family gatherings, considering the hierarchical social structure among family members, the system increases the attention weight coefficient of core members (such as elders) to prioritize satisfying the emotional preferences of elders or key members. Specifically, when calculating attention weights, members over 60 years of age receive an additional weight of 0.2, and the main decision-maker in the family (determined through user presets or historical data analysis) receives an additional weight of 0.15.

[0060] For team collaboration scenarios, considering the relatively equal status of team members, the system adopts an equal-weight fusion strategy to ensure that the emotional needs of each member are balanced. In this scenario, the initial attention weights of all members are set to equal values, and are only fine-tuned based on the members' emotional characteristics and graph convolution results to avoid excessive bias towards any one member.

[0061] For social activities, considering that the purpose of these activities is to create a lively atmosphere and promote interaction, the system dynamically adjusts weighting coefficients, prioritizing music that can enhance the atmosphere. Specifically, the weight of the activity dimension in the music synergy calculation is increased, selecting music with high activity and high enjoyment levels, while also increasing the weight of guidance. This helps to guide the group's emotions towards a positive and active state.

[0062] The scene adaptation function enables the system of this invention to flexibly adjust the recommendation strategy according to the actual application scenario, thereby improving the system's practicality and adaptability.

[0063] The complete workflow of the system of the present invention will be described in detail below with reference to specific application scenarios.

[0064] Imagine a family gathering of four: a father (60 years old), a mother (58 years old), a son (30 years old), and a daughter-in-law (28 years old). They want to listen to music together.

[0065] First, the multi-source data acquisition module 1 collects physiological signal data from four family members via wearable devices, including EEG, heart rate, and electrodermal signals. It also captures facial expressions and conversations between members through a camera and microphone, and collects scene parameters through environmental sensors. After filtering, denoising, and normalization, the collected raw data extracts a 128-dimensional emotion feature vector for each member. Assuming the initial extracted emotion features show: the father is relatively calm (pleasure level 0.6, activity level 0.3); the mother is relatively happy (pleasure level 0.7, activity level 0.5); the son is slightly tired (pleasure level 0.3, activity level 0.2); and the daughter-in-law is relatively active (pleasure level 0.5, activity level 0.6).

[0066] Secondly, the group emotional state fusion module 2 constructs a graph of the emotional influence relationships among family members based on a graph neural network. The four members serve as four nodes in the graph, with edges connected according to family relationships: edges are established between parents, father and son, mother and son, and son and daughter-in-law. Edge weights are set based on intimacy and interaction frequency, with higher weights (0.9) between parents and between son and daughter-in-law, and medium weights (0.7) between father and son and mother and son and mother and son. Through a three-layer graph convolutional network, the system learns that fathers and mothers have a greater influence on the family atmosphere, and that there is emotional resonance between son and daughter-in-law. The attention weighting unit assigns a weight of 0.35 to the father, 0.3 to the mother, 0.15 to the son, and 0.2 to the daughter-in-law, highlighting the central role of parents in family gatherings. The final fused unified group emotional vector is (pleasure 0.55, activity 0.38, dominance 0.5), indicating that the family as a whole is in a relatively peaceful and pleasant but not very active state.

[0067] Then, the music sequence collaborative generation module 3 selects suitable music from the music emotion attribute library based on the unified emotion vector of the group. The system calculates the synergy score of each piece of music in the library. Considering that the current group emotion is relatively peaceful but the activity level is not high, the system selects music that both matches the current emotion and can moderately increase the activity level. After synergy calculation and sequence optimization, the system generates a playback sequence of 5 pieces of music: the first piece is peaceful and soothing light music (pleasure level 0.6, activity level 0.3), the second piece is a slightly more lively folk song (pleasure level 0.65, activity level 0.45), the third piece is a warm and cheerful pop song (pleasure level 0.75, activity level 0.5), the fourth piece is a cheerful children's song (pleasure level 0.8, activity level 0.65), and the fifth piece is a classic family warmth song (pleasure level 0.7, activity level 0.5). This sequence achieves a smooth transition from peaceful to moderately active to warm, respecting the current emotional state of family members while gradually increasing the activity level of the atmosphere through music.

[0068] During music playback, the feedback learning and adaptive optimization module 4 continuously monitored the emotional changes of family members. After playing the first song, the system detected a slight increase in the parents' happiness and a reduction in the son's fatigue. When playing the third song, the happiness and activity levels of all members significantly increased, and the family atmosphere became more harmonious. After playing the entire sequence, the system calculated the synergistic effect: the emotional convergence reward was a positive value of 0.15 (the variance of member emotions decreased from 0.25 to 0.10), the emotional guidance effect reward was a positive value of 0.12 (the group's emotions moved closer to the target state), and the member satisfaction reward was 1.2 (average score of 4.2). The total reward value was 0.45, indicating that the music sequence achieved a good synergistic effect.

[0069] Based on this positive feedback, the system optimizes the graph neural network parameters through the parameter update unit, enhancing the modeling of the impact of parents' emotions on the family atmosphere; it also adjusts the music selection strategy through the strategy optimization unit, appropriately increasing the guidance weight to better guide group emotions through music in the future. After multiple gatherings and feedback learning, the system gradually grasps the family's emotional patterns and music preferences, resulting in continuously improving recommendation effectiveness.

[0070] To verify the technical effectiveness of this invention, a comparative experiment was conducted. Thirty different groups of four (including families, teams, and friends) participated in the test. Each group experienced both the system of this invention and a traditional single-user recommendation system three times each, with each listening session lasting 30 minutes. Satisfaction ratings and emotional measurement data were collected after the experiment.

[0071] Experimental results show that after using the system of this invention, the average satisfaction score of group members was 4.35 (out of 5), significantly higher than the 3.15 score of the traditional system; the average variance of group members' emotions decreased by 42%, indicating greater emotional convergence; and the average distance the group's unified emotions moved towards a positive state increased by 35%, indicating that the system of this invention can more effectively guide group emotions in a positive direction through music. These data fully demonstrate the significant advantages of the system of this invention in group music recommendation and emotional synchronization.

[0072] Furthermore, the system of this invention demonstrates excellent adaptability in various application scenarios. In a family setting, the system respects the emotional needs of elders, creating a warm and harmonious family atmosphere; in a team setting, the system balances the emotions of all members, enhancing team cohesion; and in a social setting, the system selects upbeat music to promote interaction and communication among members.

[0073] The main technical innovations of this invention are reflected in the following aspects: First, it innovatively introduces the concepts of group collaboration and emotional synchronization into the field of music recommendation, breaking through the limitations of traditional single-user recommendation and proposing a music sharing technology solution for groups.

[0074] Second, it innovatively uses graph neural networks to model the emotional influence relationships among group members. By learning the emotional propagation patterns among members through multi-layer graph convolutional networks and combining attention mechanisms to identify core members, it can more accurately integrate group emotions compared to traditional methods.

[0075] Third, it innovatively proposes the concept and calculation method of music synergy, which comprehensively considers three dimensions: compliance, intensity difference, and guidance. It not only meets the current emotional needs of the group, but also promotes the transition of the group's emotions to a positive state through the emotional guidance effect of music.

[0076] Fourth, an innovative closed-loop feedback optimization system was constructed, which enables the system to learn adaptively and continuously improve through real-time emotion monitoring, reinforcement learning evaluation, and automatic parameter optimization.

[0077] Fifth, a deeply coupled closed-loop collaborative system is formed among the modules. Multi-source data collection provides input for emotion fusion, the result of emotion fusion directly determines music selection, and the feedback of music playback effect affects the parameters and strategies of the preceding modules, achieving a non-linear synergy where 1 plus 1 is greater than 2.

[0078] The group collaborative music sharing and emotion synchronization system provided by this invention effectively solves the problem of balancing the emotional needs of multiple members in group scenarios by using four core technologies: multi-source data acquisition, graph neural network emotion fusion, collaborative music sequence generation, and closed-loop feedback optimization. It realizes the collaboration and synchronization of group emotions, and provides an intelligent and humanized solution for multiple people to share music. It has broad application prospects and important practical value.

[0079] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. For those skilled in the art, the present invention can be modified and varied in various ways. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of protection of the present invention.

Claims

1. A group-based collaborative music sharing and emotion synchronization system, characterized in that: include: The multi-source data acquisition module is used to collect physiological signal data, social interaction data, and environmental perception data of each member in the group. The physiological signal data includes electroencephalogram (EEG) signals, heart rate signals, and skin conductance signals. The group emotion state fusion module is connected to the multi-source data acquisition module and is used to construct a group emotion influence relationship graph based on a graph neural network. The group emotion influence relationship graph uses group members as nodes and the intensity of emotion influence between members as edge weights. It fuses the emotional features of group members through multi-layer graph convolution propagation operation to generate a unified group emotion vector. The music sequence collaborative generation module is connected to the group emotional state fusion module. It is used to calculate the degree of synergy between each candidate music and the group emotion based on the unified group emotion vector and the music emotion attribute library through an attention mechanism, and select music sequences that can promote the synergy of group emotions. The feedback learning and adaptive optimization module is connected to the multi-source data acquisition module and the music sequence collaborative generation module. It is used to monitor the emotional change trajectory of group members in real time during music playback, evaluate the collaborative effect of the music sequence based on the reinforcement learning framework, feed the evaluation results back to the group emotional state fusion module to update the graph neural network parameters, and feed them back to the music sequence collaborative generation module to adjust the music selection strategy, forming a closed-loop optimization.

2. The system according to claim 1, characterized in that, The multi-source data acquisition module includes: The physiological signal acquisition unit is used to acquire electroencephalogram (EEG) signals, heart rate signals, and skin conductance signals of group members through wearable devices; The social interaction data collection unit is used to collect dialogue content, facial expressions, and body language among group members. The environmental sensing and acquisition unit is used to collect scene type, environmental noise, and light intensity. The data preprocessing unit is used to filter, denoise, normalize, and time-align the collected multi-source data.

3. The system according to claim 1, characterized in that, The group emotion state fusion module includes: Graph building unit, used to map group members to graph nodes, and build edge connections based on the strength of social relationships and the frequency of historical interactions among members; Multi-layer graph convolutional units are used to aggregate the features of each node's neighboring nodes through a multi-layer graph convolutional network to learn the emotion propagation pattern among group members. Attention-weighted units are used to assign different weights to different members based on attention mechanisms, highlighting the influence of core members on group emotions. The emotion vector fusion unit is used to fuse the weighted emotion features of all members to generate a unified vector representing the overall emotional state of the group.

4. The system according to claim 3, characterized in that, The multi-layer graph convolutional unit updates the node representation in the following way: First, calculate the product of the node features and the adjacency matrix to obtain the aggregated neighbor features; Then, a nonlinear transformation is performed on the aggregated features to extract higher-order sentiment association patterns; Finally, residual connections are used to preserve the original node information and prevent oversmoothing.

5. The system according to claim 1, characterized in that, The music sequence collaborative generation module includes: The music emotion modeling unit is used to extract emotion feature vectors of candidate music from the music emotion attribute library. The emotion feature vectors include three dimensions: pleasure, activity, and dominance. The synergy calculation unit is used to calculate the synergy between the unified emotion vector of the group and the emotion feature vector of each candidate music. The sequence optimization unit is used to select music sequences from candidate music that can smoothly guide the emotional transition of the group based on a dynamic programming algorithm. The conflict mediation unit is used to select music with an emotional neutralizing effect to promote emotional convergence among group members when there are significant differences in their emotions.

6. The system according to claim 5, characterized in that, The collaboration degree calculation unit calculates the collaboration degree in the following way: The cosine similarity between the group's unified emotion vector and the music's emotion feature vector is calculated as the compliance score. The difference in magnitude between two vectors is calculated as the intensity difference. The directional offset of the music emotion vector relative to the group emotion vector is calculated as the guiding degree. The final synergy score is obtained by weighting and combining compliance, intensity difference, and guidance.

7. The system according to claim 1, characterized in that, The feedback learning and adaptive optimization module includes: The emotion change tracking unit is used to monitor the emotion changes of group members in real time during music playback, and to calculate the emotion offset and rate of change. The collaborative effect evaluation unit is used to evaluate the collaborative effect of music sequences based on a reinforcement learning framework, using the convergence of group emotions as a reward signal. The parameter update unit is used to update the network weights and attention parameters of the graph neural network using the gradient descent method based on the evaluation results of the collaborative effect. The strategy optimization unit is used to adjust the synergy weight coefficient in the music selection strategy to optimize the generation of future music sequences.

8. The system according to claim 7, characterized in that, The synergy effect evaluation unit defines the reward function in the following manner: Calculate the variance change of the emotion vector of group members before and after music playback; a decrease in variance is rewarded positively. Calculate the average offset direction of the emotion vector of group members, and give a positive reward if it is towards the target emotion region; Collect satisfaction feedback from group members and award positive rewards for high satisfaction levels; The three types of reward signals are weighted and summed to obtain the total reward value, which is then used for strategy optimization.

9. The system according to claim 1, characterized in that, The system also includes: The scene adaptation module is used to identify the current application scene type and adjust the group emotion integration strategy and music selection preference according to the scene type. The scene types include family gatherings, team collaborations and social activities.

10. The system according to claim 9, characterized in that, The scene adaptation module employs different emotion fusion strategies for different scenes: For family gatherings, increase the weighting of core members and prioritize satisfying the emotional preferences of elders or key members. For team collaboration scenarios, an equal weighting fusion strategy is adopted to ensure that the emotional needs of each member are considered in a balanced manner. For social events, the weighting coefficients are dynamically adjusted, prioritizing music that can liven up the atmosphere to promote social interaction.

Images