Search

CN-121980332-A - Method and system for identifying key people of silver-colored user

CN121980332ACN 121980332 ACN121980332 ACN 121980332ACN-121980332-A

Abstract

The invention relates to a method and a system for identifying key persons of a silver-colored user, wherein the method is used for carrying out feature extraction and feature fusion on acquired historical multi-mode data of the silver-colored user to obtain historical fusion data, constructing a knowledge graph which takes the silver-colored user as a central node, takes contacts of the silver-colored user as associated nodes and takes multi-factor weighted association degree of the central node and the associated nodes as an associated side based on the historical fusion data. And inputting the knowledge graph into a family relationship community division model constructed based on a weighted label propagation algorithm, and automatically identifying and dividing the family relationship communities according to the weighted label propagation rule. The method and the system combine the current scene classification result obtained according to the real-time environment perception data and the real-time communication interaction data of the silver-colored users with the constructed multidimensional dynamic weight system, dynamically calculate the importance of each family node in the family relation community to the family core, thereby obtaining the key person, and realize dynamic and accurate identification of the key person based on the importance degree.

Inventors

  • QIAN QINGWEN
  • LIN HUOWANG

Assignees

  • 福建福诺移动通信技术有限公司

Dates

Publication Date
20260505
Application Date
20251223

Claims (9)

  1. 1. A method of identifying a key person of a silvered user, comprising: Acquiring historical multi-modal data of a silver-sending user, and performing feature extraction and feature fusion on the historical multi-modal data by adopting a cross-modal feature fusion technology to obtain historical fusion data, wherein the historical multi-modal data comprises historical identity association data, historical communication interaction data, historical image social data and historical environment perception data; Constructing a knowledge graph by taking the silver-colored user as a central node and the contact of the silver-colored user as an association node based on the history fusion data, taking the multi-factor weighted association degree of the central node and the association node as an association side, inputting the knowledge graph into a family relationship community division model constructed based on a weighted label propagation algorithm, and automatically identifying and dividing a family relationship community containing family nodes by taking the silver-colored user as a family core according to a weighted label propagation rule; collecting real-time environment sensing data and real-time communication interaction data of the silver-colored user, inputting the real-time environment sensing data and the real-time communication interaction data into a pre-trained scene classification model to classify a current scene, and obtaining a current scene classification result; combining the current scene classification result with the constructed multidimensional dynamic weight system, dynamically calculating the importance of each home node in the family relation community to the home core, and obtaining key people containing priority sequence according to the importance.
  2. 2. The method for identifying key persons of a silver-colored user according to claim 1, wherein the step of performing feature extraction and feature fusion on the historical multi-modal data by using a cross-modal feature fusion technique to obtain fusion data comprises the following steps: Respectively extracting historical identity related characteristics comprising family address information and household relatives from the historical identity related data by adopting a cross-modal characteristic fusion technology, extracting historical communication interaction characteristics comprising text content of short messages, text emotion of short messages, call distribution trend and call emotion from the historical communication interaction data, extracting historical image social characteristics comprising human face similarity from the historical image social data, and extracting historical environment perception characteristics comprising intelligent wearing physiological data, positioning data, activity data and intelligent household equipment operation records from the historical environment perception data; and carrying out feature fusion on the historical identity association features, the historical communication interaction features, the historical image social features and the historical environment perception features by adopting an attention weighted fusion strategy according to preset modal feature weights to obtain fusion data comprising four major topic domains, wherein the four major topic domains are identity topics, communication topics, social topics and environment topics.
  3. 3. The method of claim 1, wherein constructing a knowledge graph with the silvery user as a center node, the silvery user's contact as an associated node, and the multi-factor weighted association of the center node and the associated node as an associated edge based on the history fusion data comprises: Extracting a user entity, a relation entity and an attribute tag based on the history fusion data, wherein the user entity comprises a silver-colored user and contacts of the silver-colored user, the relation entity comprises a relative relation, a high-frequency contact relation and a career relation, the attribute tag comprises basic information of the silver-colored user, fusion data of the silver-colored user, basic information of the contacts, a household relationship between the contacts and the silver-colored user, a communication relationship between the contacts and the silver-colored user, fusion data of the contacts, the operation times and time attenuation factors of the current month-average intelligent household equipment of the contacts, and the communication relationship comprises current month-average actual communication frequency, historical month-average communication frequency, current month actual call time and historical total call time; Based on the attribute tag, respectively calculating the relative association degree of the relative relationship between the silvery user and the contact person of the silvery user, the high-frequency association degree of the high-frequency contact person relationship and the care association degree of the care person relationship through a first association formula, a second association formula and a third association formula, wherein the first association formula is as follows: ; ; Wherein, the The degree of relatives' association is indicated, Representing the household relationship of the silvery user and the contact person of the silvery user, A time-decay factor is represented and, Representing the similarity between the fuse data of the silvered user and the fuse data of the contact, The influence factor is represented by a factor of influence, Representing a difference in days between the time of acquiring the historical multimodal data and the current time; The second association formula is: ; Wherein, the The degree of correlation at a high frequency is indicated, Indicating the actual communication frequency of the current month, Represents the historical frequency of the month-to-month communication, Indicating the actual talk time of the current month, Representing a historical total call duration; The third association formula is: ; Wherein, the The degree of correlation of the care is indicated, Indicating the operation times of the smart home equipment in the current month, Representing the operational weights; And constructing a knowledge graph taking the silver-colored user as a center node, the contact of the silver-colored user as an association node, and the relatives association degree, the high-frequency association degree and the care association degree as association sides of the center node and the association node based on a graph database.
  4. 4. The method for identifying key people of a silver-colored user according to claim 1, wherein inputting the knowledge graph into a family relationship community division model constructed based on a weighted label propagation algorithm, automatically identifying and dividing a family relationship community containing family nodes with the silver-colored user as a family core according to a weighted label propagation rule comprises: Inputting the knowledge graph into a family relationship community division model constructed based on a weighted label propagation algorithm, labeling a center node and an associated node in the knowledge graph, labeling the center node as a family core label, distributing family label weight to the family core label, and labeling the associated node as an initial node label to obtain a labeled center node and a labeled associated node; Taking the center node marked by the label and the associated node marked by the label as matrix elements, generating a weighted adjacent matrix with N multiplied by N dimensions based on the matrix elements and the associated edges of the knowledge graph, and distributing corresponding self-loop weights for the matrix elements in the weighted adjacent matrix, wherein N represents the total number of the matrix elements; sequentially taking each matrix element in the weighted adjacent matrix as a target node, taking all matrix elements which are in a neighbor relation with the target node as neighbor nodes, judging whether the number of the neighbor nodes is larger than 0, if so, calculating the label contribution value of each neighbor node to the target node through a first contribution value formula, and if not, calculating the label contribution value of each neighbor node to the target node through a second contribution value formula so as to obtain the label contribution value of all neighbor nodes, wherein the first contribution value formula is as follows: ; Wherein G (i, j) represents a label contribution value of a neighbor node j to a target node i, W (i, j) represents a multi-factor weighted association degree of the neighbor node j and the target node i in the knowledge graph, ω (j) represents a current node label weight of the neighbor node j, Representing the attenuation coefficient; The second contribution formula is: Wherein, the A tag contribution value representing the target node i, Representing the self-loop weight of the target node i, Representing a target node Is used to determine the current node tag weight of (c), Representing the attenuation coefficient; summarizing the label contribution values of the neighbor nodes of which all labels are marked as home core labels to obtain a home total contribution value, and summarizing the label contribution values of the neighbor nodes of which all labels are marked as initial node labels to obtain a non-home total contribution value; Judging whether the home total contribution value is larger than the non-home total contribution value according to a weighted tag propagation rule, if so, updating the tag label of the target node into a home core tag, otherwise, not updating the tag label of the target node; Calculating node tag weight of the target node according to a third contribution value formula, taking the node tag weight of the target node as a tag contribution value in the next round of iterative computation, realizing dynamic transfer of the node tag weight until an iterative threshold is reached, marking the tag as a home core tag and the target node as a central node as a home core, simultaneously marking the tag as a home core tag and the target node as an associated node as a home node, and obtaining a home relationship community based on the home core and the home node, wherein the third contribution value formula is as follows: ; Wherein, the The node tag weight representing the target node, Representing the total contribution value of the same tag type as the target node, Representing the total contribution value of a different tag type than the target node.
  5. 5. The method for identifying a key person of a silver-back user as recited in claim 4, wherein the obtaining a family relationship community based on the family core and the family node comprises: Calculating a quality score of the family relation community through a first quality formula, judging whether the quality score is lower than a first threshold value, if yes, performing down-regulation on an attenuation coefficient in the first contribution value formula to obtain a down-regulated attenuation coefficient, re-calculating a tag contribution value based on the down-regulated attenuation coefficient to obtain a new tag contribution value, and re-dividing the family relation community according to the new tag contribution value until the quality score of the new family relation community exceeds the first threshold value, wherein the first quality formula is as follows: ; ; ; ; Wherein Q represents a quality score, m represents the sum of all multi-factor weighted associations in the family relationship community, W (p, Q) represents the multi-factor weighted association of the family node p and the family node Q in the knowledge graph, A sum of multi-factor weighted associations representing all neighboring edges of home node p, A sum of multi-factor weighted associations representing all neighboring edges of home node q, An exponential function with respect to home node p and home node q is shown.
  6. 6. The method for identifying key persons of a silver-sending user according to claim 1, wherein the multidimensional dynamic weighting system comprises a basic weight and a scene weight, the step of combining the current scene classification result with the constructed multidimensional dynamic weighting system, and the step of dynamically calculating the importance of each home node in the family relation community to the family core, and obtaining the key persons comprising the priority order according to the importance comprises the steps of: acquiring a history communication record of each home node and the home core, wherein the history communication record comprises history communication duration and history communication frequency, inputting the history communication duration and the history communication duty ratio into a first weight formula for calculation to obtain a basic weight of each home node, and the first weight formula is as follows: ; Wherein, the Representing the base weight of the home node p, Representing a duration of the historical communication, A weight coefficient indicating the duration of the historical communication, Representing the frequency of the historical communications, A weight coefficient that affects the historical communication frequency; Acquiring a history interaction record of each home node and the home core, wherein the history interaction record comprises a history emergency response rate in a history emergency scene and a history daily response rate in a history daily scene, the history emergency response rate and the history daily response rate are input into a second weight formula to calculate, and the scene weight of each home node is generated, and the second weight formula is as follows: ; Wherein, the The scene weight of the home node p is represented, Representing a historical emergency response rate of the vehicle, A weight coefficient representing the impact of the historical emergency response rate, Representing the historical daily response rate of the device, A weight coefficient which affects the historical daily response rate; Combining the basic weight and the corresponding scene weight of each home node with the current scene classification result by adopting a characteristic attention mechanism, dynamically calculating the importance degree of each home node to the home core, and sorting all home nodes in a descending order according to the importance degree to obtain all home nodes after descending order, screening out the first M home nodes from all home nodes after descending order as key people, and taking the descending order sorting order corresponding to the key people as priority order.
  7. 7. The method of claim 6, wherein the multi-dimensional dynamic weighting system includes address weights, wherein the employing a feature attention mechanism to combine the base weight and corresponding scene weights for each home node with the current scene classification result, and wherein dynamically calculating the importance of each home node to the home core comprises: collecting current positioning data of each home node, simultaneously acquiring the current positioning data of the silver-colored user from the real-time environment sensing data, inputting the current positioning data of the silver-colored user and the current positioning data of each home node into a third weight formula for calculation, and generating address weight of each home node, wherein the third weight formula is as follows: ; Wherein, the Representing the address weight of the home node p, D P represents the current location data of the home node p, Current positioning data representing a silvery user; The address weight, the corresponding basic weight, the corresponding scene weight and the current scene classification result of each home node are input into an attention fusion formula in a characteristic attention mechanism to be dynamically calculated, so that the importance of each home node is obtained, wherein the attention fusion formula is as follows: ; ; ; Wherein, the The importance of the home node p is indicated, Representation of Is used for the concentration ratio of (a), Representation of Is a duty cycle of attention of (c).
  8. 8. The method for identifying key persons of a silver-back user according to claim 7, wherein obtaining the importance of each home node comprises: Introducing a temperature scaling mechanism to sequentially input the importance degree of each home node into a first calibration formula for calibration to obtain the importance degree of each home node after calibration, wherein the first calibration formula is as follows: ; Wherein, the The importance of the family node p after calibration is represented, N represents the set of all family nodes in the family relationship community, and t represents the temperature parameter.
  9. 9. A system for identifying a key person of a silvered user, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method of any one of claims 1 to 8 when executing the computer program.

Description

Method and system for identifying key people of silver-colored user Technical Field The invention relates to the technical field of intelligent endowment, in particular to a method and a system for identifying key people of a silver-colored user. Background Currently, the pension industry has formed an industry ecology that encompasses multiple areas of healthcare, smart home, emotion companion, and the like. However, the problem of structural imbalance between technical applications and aging requirements is increasingly pronounced. The intelligent endowment system excessively pursues basic functions such as health monitoring and environment sensing, but has insufficient attention to the needs of the old in core scenes, and especially neglects the core appeal of the intelligent endowment system in the scenes such as emergency rescue, daily assistance and the like, so that key contacts can be rapidly and accurately positioned. With respect to the core appeal, the existing system generally adopts a mode of manually presetting a contact list to provide support, and the mode cannot meet changeable actual demands and can also potentially threaten the life safety of the old. Disclosure of Invention The invention aims to solve the technical problem that the method and the system for identifying the key people of the silver-colored user can accurately identify the key people of the silver-colored user, ensure that the identified key people are more fit with actual demands, and practically ensure the safety of the silver-colored user. In order to solve the technical problems, the invention adopts the following technical scheme: in a first aspect, the present invention provides a method for identifying key persons of a silver-haired user, comprising: Acquiring historical multi-modal data of a silver-sending user, and performing feature extraction and feature fusion on the historical multi-modal data by adopting a cross-modal feature fusion technology to obtain historical fusion data, wherein the historical multi-modal data comprises historical identity association data, historical communication interaction data, historical image social data and historical environment perception data; Constructing a knowledge graph by taking the silver-colored user as a central node and the contact of the silver-colored user as an association node based on the history fusion data, taking the multi-factor weighted association degree of the central node and the association node as an association side, inputting the knowledge graph into a family relationship community division model constructed based on a weighted label propagation algorithm, and automatically identifying and dividing a family relationship community containing family nodes by taking the silver-colored user as a family core according to a weighted label propagation rule; collecting real-time environment sensing data and real-time communication interaction data of the silver-colored user, inputting the real-time environment sensing data and the real-time communication interaction data into a pre-trained scene classification model to classify a current scene, and obtaining a current scene classification result; combining the current scene classification result with the constructed multidimensional dynamic weight system, dynamically calculating the importance of each home node in the family relation community to the home core, and obtaining key people containing priority sequence according to the importance. The method has the advantages that the historical identity association data, the historical communication interaction data, the historical image social data and the historical environment perception data of the silver-colored users are subjected to feature extraction and feature fusion, the limitation of single-mode data is broken, a richer and comprehensive data basis is provided for the construction of the knowledge graph, in the construction of the knowledge graph, the multi-factor weighted association degree of the central node and the association node is used as the association side, the problem of scattered association information is solved, and the accuracy of division of family relationship communities is improved. The current scene classification is carried out according to the real-time environment perception data and the real-time communication interaction data of the silver-colored users, so that the current scene classification result is combined with the constructed multidimensional dynamic weight system, the importance of each home node to the home core is dynamically calculated, the accuracy of key person identification is improved, the generated key person is more fit with the actual demand, and the safety of the silver-colored users is practically ensured. Optionally, the performing feature extraction and feature fusion on the historical multi-modal data by using a cross-modal feature fusion technology, and obtaining the fused data includes: Respective