A dynamic group division method and system based on social sensor online social network

Abstract

The present invention belongs to the field of social networks and discloses a method and system for dynamically segmenting groups based on an online social network using social sensors. The system includes a data collection module, a user node screening module, a social sensor module, a group fusion module, a social network construction module, and a group segmentation module. The segmentation method includes constructing a social network module, linking nodes based on obtained weights and the positions of each node, constructing an online social network, and performing sentiment analysis on text, voice, and emoticons received and sent by the nodes and social sensors. By dynamically updating link weights, node attributes are dynamically adjusted based on user needs, and groups in the online social network are dynamically adjusted and segmented, thereby accurately segmenting the groups. The present invention constructs an online social network using data collected by social sensors and dynamically segments the groups based on user needs.

Description

Technical Field

[0001] The present invention belongs to the field of social networks, and more particularly relates to a method and system for dynamic group segmentation based on online social networks of social sensors. Background Art

[0002] Research on social networks and online social networks is rapidly developing, and online social activities have become an integral part of daily life for communities. Commonly used social platforms include WeChat, QQ, Weibo, and TikTok. People connect with each other through online chat software, and chat content generally falls into three categories: text, voice, and emoticons.

[0003] There are numerous traditional social network modeling methods, but they all rely on extensive sample data. Given the current dynamics of social environments and the complexity of social phenomena, researchers lack the time and effort to acquire accurate sample data. This leads to the risk of data leakage and modification, resulting in incomplete network structures. Consequently, traditional modeling methods have limitations, lacking accuracy despite the abundance of sample data.

[0004] How to accurately divide the network structure in small online social networks such as WeChat and QQ groups is an urgent problem that needs to be solved. Summary of the Invention

[0005] In response to the shortcomings of existing group segmentation methods in online social networks, the present invention proposes a dynamic group segmentation method and system based on social sensors in online social networks. Social sensors are introduced to collect reliable sample data through the social sensors, and an online social network is constructed based on the data collected by the social sensors. Then, groups are dynamically divided according to user needs, which is of great social value.

[0006] In order to achieve the above object, the present invention is achieved through the following technical solutions:

[0007] The present invention is a method for dynamic group segmentation based on an online social network of social sensors, the method comprising the following steps:

[0008] Step 1: Obtain a questionnaire survey report on a new product from a social group through online methods;

[0009] Step 2: Filter the user nodes with activity data and select the nodes that are related to each other;

[0010] Step 3: Strategically select a person in the social group as a human social sensor and obtain the attribute value of the person's direct social environment, that is, the social circle in which the social sensor is located. The selected human social sensor actively or passively receives and sends messages, and the user node actively receives and sends messages. The messages include text, voice, and emoticons;

[0011] Step 4: Calculate the social group questionnaire report obtained in step 1 and the social circle attribute value based on the social sensor obtained in step 3, and calculate the weight value of the user node and the social sensor node;

[0012] Step 5: Construct an online small social network based on the weight values ​​obtained in step 4, and perform sentiment analysis on the text, voice, and emoticons received and sent by user nodes and social sensors to achieve sentiment classification;

[0013] Step 6: By dynamically updating the link weights, the attributes of the user nodes are dynamically adjusted according to the user's needs, thereby dynamically and accurately dividing the social network groups.

[0014] A further improvement of the present invention is that in step 2, the user nodes with activity data are screened using a binary tree, which specifically includes the following steps:

[0015] Step 2.1: Determine whether the user node is empty. If it is empty, it means that the user node has no connection with other nodes in the activity data, and the user node is deleted.

[0016] Step 2.2: Determine whether the user node is a leaf node. If the user node is a leaf node, retain the user node.

[0017] Step 2.3, recursively calculate the number of nodes in the left subtree and the right subtree, and add the number of nodes in the left subtree and the right subtree to get the total number of leaf nodes;

[0018] Step 2.4: If the total number of leaf nodes is greater than 1, it means that the user node is connected to other nodes in the activity data; otherwise, the user node has no adjacent nodes.

[0019] Step 2.5: Complete the screening of user nodes by calculating the total number of leaf nodes.

[0020] A further improvement of the present invention is that in step 3, the strategic selection is represented by selecting users with active behaviors and high activity frequency as social sensors.

[0021] A further improvement of the present invention is that: Step 4 is specifically as follows: Step 2 uses a binary tree to calculate the number of nodes in the left subtree and right subtree of the user node, which is the edge value between the user node and other adjacent nodes, and then adds up the weights of all edges of each user node to obtain the weight value of the user node and the social sensor.

[0022] A further improvement of the present invention is that: Step 5 uses the node distance and clustering algorithm NSCAN to construct an online small social network, and the clustering algorithm NSCAN specifically includes the following steps:

[0023] Step 5.1. Initialize the user nodes and determine two parameters: the correlation w and distance d between the user nodes. The correlation w represents the similarity between the two user nodes. If the activity attributes of the user nodes are similar, w is assigned a value of 1. Otherwise, if the activity attributes of the user nodes are dissimilar, w is assigned a value of -1. The distance d represents the position distance between the two user nodes. The interval of distance d is [-1, 1]. The positive direction of the user node is the right. When distance d is close to -1, it indicates that user node a is to the left of user node b and is far away. Similarly, when distance d is close to 1, it indicates that user node a is to the right of user node b and is far away.

[0024] Step 5.2: Find adjacent nodes: Given the initial positions of user nodes a and b and the correlation w between the two nodes, if the correlation w is 1, it indicates that the two nodes are highly correlated, i.e., highly similar. Therefore, connect user nodes a and b. The distance between them determines whether user node a is to the left or right of user node b.

[0025] Step 5.3: Cluster related nodes: Use a binary tree to determine the links between the user node and other adjacent nodes. Use the weights of the node edges to determine the similarity w of the user nodes and the distance d between the user nodes. This process is performed recursively.

[0026] Step 5.4, Mark pending nodes: Mark irrelevant user nodes as pending nodes until the information sent by the irrelevant nodes is relevant to the content of other user nodes, then unmark and connect with other user nodes;

[0027] Step 5.5: Output clustering results: The clustering results between user nodes are calculated by calculating the weights of the edges between binary tree nodes and connecting them by judging the distance d, forming an online small social network. The sentiment analysis of the text, voice, and emoticons received and sent by the user nodes is based on the neural network.

[0028] A further improvement of the present invention lies in that, in step 6, node attributes are dynamically adjusted based on user needs through dynamic updating of link weights, thereby dynamically and accurately segmenting social network groups. Specifically, the graph neural network represents each node as a vector, where each element in the vector represents a specific attribute or feature of the node. The graph neural network then uses graph convolution operations to transfer and aggregate the relationships and similarities between nodes, thereby clustering the nodes and dynamically segmenting social network groups.

[0029] The present invention also provides a dynamic group division system, which includes a data collection module, a user node screening module, a social sensor module, a group fusion module, a social network construction module, and a group division module, wherein:

[0030] The data collection module collects survey reports from social groups online;

[0031] The user node screening module screens the user nodes with activity data and selects the nodes that are related to each other;

[0032] The social sensor module strategically selects a sensor, actively or passively sends information, and calculates the size and weight coefficient of the social circle corresponding to the sensor;

[0033] The group fusion module processes the social groups and the social circles corresponding to the sensors to obtain the location of each node;

[0034] The social network construction module links user nodes according to the weights obtained by weight calculation, builds an online social network, and performs sentiment analysis on the text, voice and emoticons received and sent by user nodes and social sensors to achieve sentiment classification;

[0035] In the group division module, users can dynamically adjust and divide the groups of the online social network by searching for keywords or adjusting attribute values, and then accurately divide them into groups.

[0036] The beneficial effects of the present invention are as follows: the present invention eliminates the "self-island" bias to screen out social sensors, calculates the node weights of social groups and social circles based on social sensors through binary tree weights, and constructs a small online social network. The text, voice, and emoticons received and sent by the groups and social sensor nodes are analyzed for semantic sentiment. Finally, according to the specific needs of users, the online social network is divided into different groups by adjusting the attribute values ​​of the nodes, thereby achieving the goals of precise advertising, public opinion control, etc.

[0037] To more accurately segment online groups, this method uses semantic sentiment analysis of messages sent and received by nodes to generate sentiment classification results, improving the accuracy of network group segmentation methods. When promoting and advertising products, node clustering can be used to identify social groups that are closest to target customers, understanding their interests, hobbies, and needs, thereby enabling better product positioning and promotion. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a structural diagram of the group division system of the present invention.

[0039] Figure 2 It is a flow chart of the method implemented by the present invention.

[0040] Figure 3 is a schematic diagram of a social sensor. DETAILED DESCRIPTION

[0041] The following diagrams illustrate embodiments of the present invention. For clarity, many practical details are included in the following description. However, it should be understood that these practical details are not intended to limit the present invention. In other words, in some embodiments of the present invention, these practical details are not essential.

[0042] like Figure 1 As shown, a dynamic group segmentation system based on an online social network using social sensors includes a group data collection module, a user node screening module, a social sensor module, a group fusion module, a social network construction module, and a group segmentation module. The data collection module collects online survey reports from social groups on a company's new products. The user node screening module preprocesses the collected questionnaire data, screening user nodes with active data and selecting interconnected nodes. The social sensor module strategically selects a social sensor, actively or passively sends information, and calculates the size and weight coefficient of the sensor's corresponding social circle. The group fusion module processes the social group and the sensor's corresponding social circle to determine the position and weight of each node. The social network construction module links nodes based on the weights calculated. The online social network construction module performs sentiment analysis on text, voice, and emoticons received and sent by user nodes and social sensors to achieve sentiment classification. The group segmentation module dynamically adjusts and divides the online social network into groups by searching for keywords or adjusting attribute values.

[0043] like Figure 2 As shown, the present invention provides a dynamic group segmentation method based on a social sensor online social network, which specifically includes the following steps:

[0044] Step 1: Obtain a questionnaire survey report on social groups’ opinions on a company’s new product through online channels such as WeChat, QQ, Weibo, and Douyin;

[0045] Step 2: Further pre-process the collected questionnaire data, filter the user nodes with activity data, and select the nodes that are related to each other.

[0046] Because the present invention needs to establish a small online social network and find user nodes that are related to each other, the present invention needs to filter users with activity data. The present invention uses a binary tree to filter user nodes with activity data, which specifically includes the following steps:

[0047] Step 2.1: Determine whether the user node is empty. If it is empty, it means that the user node has no connection with other nodes in the activity data, and the user node is deleted.

[0048] Step 2.2: Determine whether the user node is a leaf node. If the user node is a leaf node, retain the user node.

[0049] Step 2.3, recursively calculate the number of nodes in the left subtree and the right subtree, and add the number of nodes in the left subtree and the right subtree to get the total number of leaf nodes;

[0050] Step 2.4: If the total number of leaf nodes is greater than 1, it means that the user node is connected to other nodes in the activity data; otherwise, the user node has no adjacent nodes.

[0051] Step 2.5: Complete the screening of user nodes by calculating the total number of leaf nodes.

[0052] Step 3: Strategically select a person in the social group as a human social sensor. Obtain the attribute values ​​of the person's immediate social environment, i.e., the social circle in which the social sensor resides. The selected human social sensor actively or passively receives and sends messages, while the user node actively receives and sends messages. These messages may include text, voice, and emoticons. The selection of social sensors is also important because biases will always occur when using human social perception methods. Therefore, it is important to select the right social sensor.

[0053] like Figure 3 This is a schematic diagram of a social sensor. The present invention requires strategically selecting a person in a social group as a social sensor, and actively or passively receiving messages, which include text, voice and emoticons.

[0054] In the present invention, there is a kind of network bias called "self-island". A user node believes that it is an islander of information island. Its own opinions are the opinions of the entire island and it refuses to communicate with others. They believe that the bias they insist on is the "truth". In the present invention, it is necessary to eliminate nodes such as "self-island" because the present invention requires each node to effectively receive and send information to realize the connection of small networks. The present invention needs to use sensors as representative samples to describe a wide range of people. Therefore, if it is a "prejudiced individual" P (Prejudiced), the degree of its social circle is calculated as follows

[0055]

[0056] in The meaning is the sum of the number of edges between node i and other adjacent nodes j, ω ab is the correlation between the characteristics of user node a and user node b, so the degree of the “biased individual” p can be obtained Assume that the other individuals are collectively referred to as O(other), if Then the "biased individual" p is the selected social sensor. Then individual O will replace p and become the new social sensor, where It is expressed as the maximum individual degree of the user node. In the “self-island”, the correlation ω between user a and user b is determined by ab The value of

[0057]

[0058] Get the degree of its adjacent nodes if This means that node p is a biased individual "self-island" and should be removed. This means that node p is linked to other nodes and is not a biased individual.

[0059] Step 4: Conduct a survey report on the social group, which is for the entire social circle. Calculate weight values ​​for the social groups and the social circles of the social sensors. There are many methods for weight calculation, and different methods have different characteristics and applicability. However, since the present invention divides online social networks into different groups based on user needs, it may involve a hot topic or opinions on a certain product, so a method combining fuzzy comprehensive evaluation and binary tree is selected to perform weight calculation. The fuzzy comprehensive evaluation and binary tree methods are specifically as follows:

[0060] First, the fuzzy evaluation matrix is ​​constructed by the fuzzy comprehensive evaluation method. The binary tree weight calculation method is to calculate the weight of each user node and other adjacent nodes, and the weight matrix of the tree node ω can be obtained.ab Then the fuzzy evaluation matrix obtained by the fuzzy comprehensive evaluation method and the weight tree matrix obtained by the binary tree method are combined through H(x)=kλμ+ω ab The final weight value H can be obtained ab .

[0061] Step 5: Construct a small online social network based on the weight values ​​obtained in step 4, and perform sentiment analysis on the text, voice, and emoticons received and sent by user nodes and social sensors to achieve sentiment classification. The network model selected in the present invention is a small-world network model, and the present invention uses the NSCAN clustering algorithm based on node distance and correlation to construct the small online social network.

[0062] The clustering algorithm NSCAN specifically includes the following steps:

[0063] Step 5.1. Initialize the user nodes and determine two parameters: the correlation w and distance d between the user nodes. The correlation w represents the similarity between the two user nodes. If the activity attributes of the user nodes are similar, w is assigned a value of 1. Otherwise, if the activity attributes of the user nodes are dissimilar, w is assigned a value of -1. The distance d represents the position distance between the two user nodes. The interval of distance d is [-1, 1]. The positive direction of the user node is the right. When distance d is close to -1, it indicates that user node a is to the left of user node b and is far away. Similarly, when distance d is close to 1, it indicates that user node a is to the right of user node b and is far away.

[0064] Step 5.2: Find adjacent nodes: Given the initial positions of user nodes a and b and the correlation w between the two nodes, if the correlation w is 1, it indicates that the two nodes are highly correlated, i.e., highly similar. Therefore, connect user nodes a and b. The distance between them determines whether user node a is to the left or right of user node b.

[0065] Step 5.3: Cluster related nodes: Use a binary tree to determine the links between the user node and other adjacent nodes. Use the weights of the node edges to determine the similarity w of the user nodes and the distance d between the user nodes. This process is performed recursively.

[0066] Step 5.4, Mark pending nodes: Mark irrelevant user nodes as pending nodes until the information sent by the irrelevant nodes is relevant to the content of other user nodes, then unmark and connect with other user nodes;

[0067] Step 5.5, Output the clustering result: Output the clustering result among user nodes, that is, the result calculated by the weights of the edges between the binary tree nodes and connected by judging the distance d, to form a small online social network.

[0068] Perform semantic sentiment orientation analysis on the text, voice, and emojis received and sent by user nodes and social sensors to obtain the sentiment classification result; since short texts are common in small online social networks but context is also needed, a model that combines the Attention mechanism and BIGRU is used. The first layer of the model is the word vector layer. First, convert the text into a fixed-length vector representation of features, using the distributed representation method, and the vector corresponding to each character is [λ1, λ2, ……, λ n ; The second layer is the BIGRU layer, which processes the input vector through the forward and backward GRUs to remove context information and obtain the preliminary prediction result [μ1, μ2, ……, μ n ; The third layer is the Attention layer. The Attention mechanism is to obtain the similarity probability value between the target character λ t and the input text vector [λ1, λ2, ……, λ n ; The fourth layer is the combination layer, which combines the output of the BIGRU layer and the output of the Attention layer. The combined value is the predicted classification result value. For the problem of sentiment classification of voice, we can regard the audio as a piece of text. The present invention can use the same method. First, convert the voice information into text information through speech recognition technology, and then follow the same steps as the text sentiment recognition method. The predicted classification result value obtained is the sentiment classification result of the voice. <​​​​This invention can analyze the sentiment of text, voice, and emoticons in group chats to understand the preferences of social groups. It can dynamically divide social networks into different groups based on different user needs, laying a solid foundation for companies to better promote new products.

[0071] Dynamic group segmentation in online social networks based on social sensors plays a vital role in many areas. In our daily interactions, social sensors, by collecting chat data from social groups, can effectively control the spread of public opinion and enable businesses to implement targeted advertising. Building online social networks using data collected by social sensors and dynamically segmenting groups based on user needs is of great social value.

[0072] The foregoing is merely an embodiment of the present invention and is not intended to limit the present invention. It will be apparent to those skilled in the art that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention are intended to be included within the scope of the claims of the present invention.

Claims

1. A dynamic group segmentation method based on online social networks of social sensors, characterized by: The dynamic group division method comprises the following steps: Step 1: Obtain a questionnaire survey report on a new product from a social group through online methods; Step 2: Filter the user nodes with activity data and select the nodes that are related to each other; Step 3: Strategically select a person in the social group as a human social sensor and obtain the attribute value of the person's direct social environment, that is, the social circle in which the social sensor is located. The selected human social sensor actively or passively receives and sends messages, and the user node actively receives and sends messages. The messages include text, voice, and emoticons; Step 4: Calculate the social group questionnaire report obtained in step 1 and the social circle attribute value based on the social sensor obtained in step 3, and calculate the weight value of the user node and the social sensor node; Step 5: Construct an online small social network based on the weight values ​​obtained in step 4, and perform sentiment analysis on the text, voice, and emoticons received and sent by user nodes and social sensors to achieve sentiment classification; Step 6: By dynamically updating the link weights, the attributes of the user nodes are dynamically adjusted according to the user's needs, thereby dynamically and accurately dividing the social network groups; wherein, In step 2, the user nodes with activity data are screened using a binary tree, which specifically includes the following steps: Step 2.1: Determine whether the user node is empty. If it is empty, it means that the user node has no connection with other nodes in the activity data, and the user node is deleted. Step 2.2: Determine whether the user node is a leaf node. If the user node is a leaf node, retain the user node. Step 2.3, recursively calculate the number of nodes in the left subtree and the right subtree, and add the number of nodes in the left subtree and the right subtree to get the total number of leaf nodes; Step 2.4: If the total number of leaf nodes is greater than 1, it means that the user node is connected to other nodes in the activity data; otherwise, the user node has no adjacent nodes. Step 2.5: Filter the user nodes by calculating the total number of leaf nodes. The step 5 uses the node distance and clustering algorithm NSCAN to build an online small social network. The clustering algorithm NSCAN specifically includes the following steps: Step 5.

1. Initialize the user nodes and determine two parameters: the correlation w and distance d between the user nodes. The correlation w represents the similarity between the two user nodes. If the activity attributes of the user nodes are similar, w is assigned a value of 1. Otherwise, if the activity attributes of the user nodes are dissimilar, w is assigned a value of -1. The distance d represents the position distance between the two user nodes. The interval of distance d is [-1, 1]. The positive direction of the user node is the right. When distance d is close to -1, it indicates that user node a is to the left of user node b and is far away. Similarly, when distance d is close to 1, it indicates that user node a is to the right of user node b and is far away. Step 5.2: Find adjacent nodes: Given the initial positions of user nodes a and b and the correlation w between the two nodes, if the correlation w is 1, it indicates that the two nodes are highly correlated, i.e., highly similar. Therefore, connect user nodes a and b. The distance between them determines whether user node a is to the left or right of user node b. Step 5.3: Cluster related nodes: Use a binary tree to determine the links between the user node and other adjacent nodes. Use the weights of the node edges to determine the similarity w of the user nodes and the distance d between the user nodes. This process is performed recursively. Step 5.4, Mark pending nodes: Mark irrelevant user nodes as pending nodes until the information sent by the irrelevant nodes is relevant to the content of other user nodes, then unmark and connect with other user nodes; Step 5.5: Output clustering results: The clustering results between user nodes are calculated by calculating the weights of the edges between binary tree nodes and connecting them by judging the distance d, forming an online small social network. The sentiment analysis of the text, voice, and emoticons received and sent by the user nodes is based on the neural network.

2. The method for dynamic group segmentation based on social sensor online social network according to claim 1, characterized in that: In step 3, the strategic selection is represented by selecting users with active behaviors and high activity frequency as social sensors.

3. The method for dynamic group segmentation based on social sensor online social network according to claim 2, characterized in that: Specifically, step 4 includes: using a binary tree to calculate the number of nodes in the left subtree and right subtree of the user node in step 2, which is the edge value between the user node and other adjacent nodes; then adding up the weights of all edges of each user node to obtain the weight value of the user node and the social sensor.

4. The method for dynamic group segmentation based on social sensor online social network according to claim 1, characterized in that: The dynamic group division method is implemented according to a dynamic group division system, which includes a data collection module, a user node screening module, a social sensor module, a group fusion module, a social network construction module, and a group division module, wherein: The data collection module collects survey reports from social groups online; The user node screening module screens the user nodes with activity data and selects the nodes that are related to each other; The social sensor module strategically selects a sensor, actively or passively sends information, and calculates the size and weight coefficient of the social circle corresponding to the sensor; The group fusion module processes the social groups and the social circles corresponding to the sensors to obtain the location of each node; The social network construction module links user nodes according to the weights obtained by weight calculation, builds an online social network, and performs sentiment analysis on the text, voice and emoticons received and sent by user nodes and social sensors to achieve sentiment classification; In the group division module, users can dynamically adjust and divide the groups of the online social network by searching for keywords or adjusting attribute values, and then accurately divide them into groups.

Images