CN112765489B - Social network link prediction method and system - Google Patents
Social network link prediction method and system Download PDFInfo
- Publication number
- CN112765489B CN112765489B CN202110101233.0A CN202110101233A CN112765489B CN 112765489 B CN112765489 B CN 112765489B CN 202110101233 A CN202110101233 A CN 202110101233A CN 112765489 B CN112765489 B CN 112765489B
- Authority
- CN
- China
- Prior art keywords
- social
- user
- social influence
- data set
- under
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 51
- 238000009826 distribution Methods 0.000 claims abstract description 101
- 238000013528 artificial neural network Methods 0.000 claims abstract description 75
- 230000006870 function Effects 0.000 claims description 39
- 238000012546 transfer Methods 0.000 claims description 28
- 238000010586 diagram Methods 0.000 claims description 13
- 230000005540 biological transmission Effects 0.000 claims description 12
- 239000011159 matrix material Substances 0.000 claims description 12
- 239000013598 vector Substances 0.000 claims description 9
- 238000012549 training Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 6
- 230000008447 perception Effects 0.000 claims description 6
- 230000006855 networking Effects 0.000 claims 1
- 230000000694 effects Effects 0.000 description 12
- 238000004458 analytical method Methods 0.000 description 4
- 238000012217 deletion Methods 0.000 description 4
- 230000037430 deletion Effects 0.000 description 4
- 230000003068 static effect Effects 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 238000004140 cleaning Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 230000002776 aggregation Effects 0.000 description 2
- 238000004220 aggregation Methods 0.000 description 2
- 230000001419 dependent effect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000004931 aggregating effect Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000007621 cluster analysis Methods 0.000 description 1
- 230000019771 cognition Effects 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000003631 expected effect Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000005295 random walk Methods 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/01—Social networking
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Business, Economics & Management (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Primary Health Care (AREA)
- Marketing (AREA)
- Human Resources & Organizations (AREA)
- Probability & Statistics with Applications (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to a social network link prediction method and a social network link prediction system. The method comprises the following steps: acquiring a social network data set to be predicted; respectively inputting the social network data sets to be predicted into topic-aware social influence models under different topics to obtain social influence probability distribution of the social network data sets to be predicted under each topic; clustering social influence probability distributions under all topics of the social network data set to be predicted to obtain social influence probability distributions under each topic class; inputting the social network data set to be predicted and the social influence probability distribution under each subject category into a trained graph neural network to obtain the social influence propagation modes of each user to be predicted under different identities; and inputting the corresponding social influence propagation modes of the two users to be predicted under different identities into a similarity function to obtain similarity, and determining a link prediction result between the two users to be predicted according to the similarity. The invention improves the accuracy of the link prediction.
Description
Technical Field
The invention relates to the field of data mining, in particular to a social network link prediction method and a social network link prediction system.
Background
Early researchers defined the link prediction problem as the likelihood of speculating future connections between unconnected member nodes. In recent years, with the development of various types of social media, the problem of link prediction in a social network is receiving more and more attention from researchers. The link prediction problem is a Non-deterministic (Non-DETERMINISTIC POLYNOMIAL, NP) problem of polynomial complexity, and an excellent link prediction method should be able to accurately predict potential link objects in a social network in as short a time as possible. This is a very complex problem, and too high a time complexity may lead to a very poor user experience, while too low an accuracy would make model prediction meaningless. Although researchers have attempted to solve the link prediction problem from different angles, most model approaches focus only on static information in the social network—node attributes, and social network topology.
Node attributes are the most readily available information in a social network, which is provided in advance by a user. The node attribute is label information preset by a user or a system, so that the node attribute contains rich and concentrated user characteristics, the node attribute is main supporting information of a link prediction method in the early research stage of link prediction, and a plurality of link prediction algorithms based on the node attribute have excellent performance. In addition, the consistent performance of node attributes effectively helps the system infer social groups in the social network, which is a very compact and efficient method that has been widely used until now. Social network topologies differ from node attributes in that they focus on only local sub-graph structures of user nodes. The topology is not a kind of display information that is artificially formulated, but an implicit relation information. In contrast to node attributes, the topology does not require user preset nor system configuration of corresponding tags, and is a relatively dynamically variable message that can be adjusted to the changes in the social network. Thus, although the efficiency of link prediction using topology is significantly lower than the method using node attributes, topology-based methods still have great utility.
Most of the conventional link prediction techniques are based on node properties or topology, and some research effort is directed to fusing information of two different structures. Although some progress has been made in these research works, there are certain limitations in practical application scenarios. Because conventional link prediction techniques always assume that node properties and topology are perfectly correct, node properties and topology are not truly trustworthy due to the complexity of social networks and uncertainty in the real world. In real life, node attribute marking errors or link deletions can be caused by some objective factors; there are also some subjective factors that the user may begin to misjudge the node properties or deliberately hide the existing link relationships. In summary, the node properties and topology are not fully trusted in the real world, and the authenticity directly affects the model judgment. The topology of either node attributes or node neighborhood can be manually controlled, modified and hidden, and represents an explicit relationship between nodes, although the topology is a dynamic structure. Therefore, the link prediction method simply considering static explicit information (node attribute, topology) has unsatisfactory effect in practical application.
Disclosure of Invention
The invention aims to provide a social network link prediction method and a social network link prediction system, which improve the accuracy of link prediction.
In order to achieve the above object, the present invention provides the following solutions:
A social network link prediction method, comprising:
Acquiring a social network data set to be predicted, wherein the social network data set to be predicted is a social network topological structure diagram where a user to be predicted is located;
Respectively inputting the social network data set to be predicted into topic-aware social influence models under different topics to obtain social influence probability distribution of the social network data set to be predicted under each topic;
Clustering social influence probability distributions of all topics of the social network data set to be predicted to obtain social influence probability distributions of all topic categories of the social network data set to be predicted;
Inputting the social influence probability distribution of the social network data set to be predicted and the social influence probability distribution of the social network data set to be predicted under each subject category into a trained graph neural network to obtain social influence propagation modes of each user to be predicted under different identities; the identity comprises an influencer and an influencer;
And inputting the social influence propagation modes of the two corresponding users to be predicted under different identities into a similarity function to obtain similarity for any two users to be predicted, and determining a link prediction result between the two users to be predicted according to the similarity.
Optionally, the method for determining the trained graph neural network includes:
Acquiring a social network data set to be trained; the to-be-trained social network data set is a social network topological structure diagram among all training users;
Respectively inputting the social network data set to be trained into topic perception social influence models under different topics to obtain social influence probability distribution of the social network data set to be trained under each topic;
Clustering social influence probability distributions of the to-be-trained social network data set under all topics to obtain social influence probability distributions of the to-be-trained social network data set under each topic category;
Inputting the social influence probability distribution of the social network data set to be trained and the social network data set to be trained under each subject category into a graph neural network to train a propagation layer of the graph neural network until a loss function reaches a set condition, and determining that the graph neural network corresponding to the current propagation layer is a trained graph neural network; the loss function is constructed according to social influence propagation modes of users to be trained under different identities, which are output by the propagation layer of the graph neural network.
Optionally, the clustering is performed on social influence probability distributions of the to-be-trained social network data set under all topics, so as to obtain social influence probability distributions of the to-be-trained social network data set under each topic category, which specifically includes:
for social influence probability distribution under any several topics in the social network data set to be trained, calculating the difference value between the social influence probability distribution under the several topics;
and when the difference value is in the set range, dividing a plurality of topics corresponding to the difference value into the same topic category, and adding social influence probability distribution of a plurality of topics corresponding to the difference value to obtain the social influence probability distribution under the corresponding topic category.
Optionally, the propagation layer in the trained neural network includes an information transfer module and an update module:
The information transfer module is
Wherein ms v→u,c is the transfer information from the affected party v to the affected party u in the subject category, f MLP is a multi-layer perceptron,/>Influence propagation mode matrix for user u as influencer,/>For the social influence propagation mode under the topic category c with the user u as an influencer at the layer t of the graph neural network,/>In order to take the user u as the social influence propagation mode of the affected person under the subject category at the layer t of the graphic neural network, alpha v→u,c is attention, exp () is an exponential function based on a natural number e, and/>For the social influence probability of user u to user v in topic category c,/>For the social influence probability of the user u to the user v' in the topic category c, mt v→u,c is the transfer information of the affected user v to the user u,/>For user u as the influencer's influence propagation mode matrix, β v→u,c is the user u's attention to user v,/>For the social influence probability of user v to user u in topic category c,/>Social influence probability of the subject class c for the user v' to the user u;
The updating module is
Wherein/>In order to take a user u as an influencer in a social influence propagation mode of a topic class c at a layer t+1 of the graphic neural network, f LSTM is an LSTM function, theta 1 is a first parameter,/>In order to take the user u as the social influence propagation mode of the affected person in the topic category c in the layer t+1 of the graphic neural network, ms v'→v,c is the transmission information of the affected person v 'to the user v, θ 2 is a second parameter, and mt v'→v,c is the transmission information of the affected person v' to the user v.
Optionally, the similarity function is specifically:
Wherein S (u, v) is the similarity between user u and user v, K is the hyper-parameter, d (·) is the Euclidean distance between the vectors, S u,c is the social influence propagation mode of user u under the influencer identity in topic class c, S v,c is the social influence propagation mode of user v under the influencer identity in topic class c, T u,c is the social influence propagation mode of user u under the influencer identity in topic class c, T v,c is the social influence propagation mode of user v under the influencer identity in topic class c.
A social network link predicting system, comprising:
The acquisition module is used for acquiring a social network data set to be predicted, wherein the social network data set to be predicted is a social network topological structure diagram where a user to be predicted is located;
The input module is used for respectively inputting the social network data set to be predicted into topic-aware social influence models under different topics to obtain social influence probability distribution of the social network data set to be predicted under each topic;
The topic category probability determining module is used for clustering social influence probability distribution of the to-be-predicted social network data set under all topics to obtain social influence probability distribution of the to-be-predicted social network data set under each topic category;
The social influence propagation mode determining module is used for inputting the social influence probability distribution of the social network data set to be predicted and the social influence probability distribution of the social network data set to be predicted under each subject category into the trained graph neural network to obtain the social influence propagation modes of each user to be predicted under different identities; the identity comprises an influencer and an influencer;
And the link prediction result determining module is used for inputting the social influence propagation modes of the two corresponding users to be predicted under different identities into a similarity function to obtain similarity for any two users to be predicted, and determining the link prediction result between the two users to be predicted according to the similarity.
Optionally, the social influence propagation mode determining module includes:
The acquisition unit is used for acquiring a social network data set to be trained; the to-be-trained social network data set is a social network topological structure diagram among all training users;
The input unit is used for respectively inputting the social network data set to be trained into topic-aware social influence models under different topics to obtain social influence probability distribution of the social network data set to be trained under each topic;
The topic category probability determining unit is used for clustering social influence probability distribution of the to-be-trained social network data set under all topics to obtain social influence probability distribution of the to-be-trained social network data set under each topic category;
The model determining unit is used for inputting the social network data set to be trained and the social influence probability distribution of the social network data set to be trained under each subject category into the graph neural network to train the propagation layer of the graph neural network until the loss function reaches a set condition, and determining that the graph neural network corresponding to the current propagation layer is a trained graph neural network; the loss function is constructed according to social influence propagation modes of users to be trained under different identities, which are output by the propagation layer of the graph neural network.
Optionally, the topic category probability determining unit includes:
The difference value calculation subunit is used for calculating the difference value among social influence probability distributions under any several topics for the social influence probability distribution under any several topics in the social network data set to be trained;
And the theme category probability determination subunit is used for dividing a plurality of themes corresponding to the difference value into the same theme category when the difference value is in a set range, and adding social influence probability distribution of the themes corresponding to the difference value to obtain the social influence probability distribution under the corresponding theme category.
Optionally, the propagation layer in the trained neural network includes an information transfer module and an update module:
The information transfer module is
Wherein ms v→u,c is the transfer information from the affected party v to the affected party u in the subject category, f MLP is a multi-layer perceptron,/>Influence propagation mode matrix for user u as influencer,/>For the social influence propagation mode under the topic category c with the user u as an influencer at the layer t of the graph neural network,/>In order to take the user u as the social influence propagation mode of the affected person under the subject category at the layer t of the graphic neural network, alpha v→u,c is attention, exp () is an exponential function based on a natural number e, and/>For the social influence probability of user u to user v in topic category c,/>For the social influence probability of the user u to the user v' in the topic category c, mt v→u,c is the transfer information of the affected user v to the user u,/>For user u as the influencer's influence propagation mode matrix, β v→u,c is the user u's attention to user v,/>For the social influence probability of user v to user u in topic category c,/>Social influence probability of the subject class c for the user v' to the user u;
The updating module is
Wherein/>In order to take a user u as an influencer in a social influence propagation mode of a topic class c at a layer t+1 of the graphic neural network, f LSTM is an LSTM function, theta 1 is a first parameter,/>In order to take the user u as the social influence propagation mode of the affected person in the topic category c in the layer t+1 of the graphic neural network, ms v'→v,c is the transmission information of the affected person v 'to the user v, θ 2 is a second parameter, and mt v'→v,c is the transmission information of the affected person v' to the user v.
Optionally, the similarity function is specifically:
Wherein S (u, v) is the similarity between user u and user v, K is the hyper-parameter, d (·) is the Euclidean distance between the vectors, S u,c is the social influence propagation mode of user u under the influencer identity in topic class c, S v,c is the social influence propagation mode of user v under the influencer identity in topic class c, T u,c is the social influence propagation mode of user u under the influencer identity in topic class c, T v,c is the social influence propagation mode of user v under the influencer identity in topic class c.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects: according to the method, the social influence probability among the users is obtained by processing the social network data to be predicted by adopting the subject perception social influence model, the more accurate similarity among the users to be predicted is obtained by utilizing the more reliable social influence probability distribution input graph neural network, the potential relation in the social network can be captured more accurately, and the accuracy of link prediction is improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the drawings that are needed in the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flowchart of a social network link prediction method provided by an embodiment of the present invention;
FIG. 2 is a schematic flow chart of a social network link prediction method according to an embodiment of the present invention;
FIG. 3 is a system block diagram of a social network link prediction system provided by an embodiment of the present invention;
Fig. 4 is a graph comparing a predicted result obtained by using the social network link predicting method provided by the embodiment of the invention on a plurality of data sets with a predicted result obtained by using the existing node attribute or topology structure based method.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
In order that the above-recited objects, features and advantages of the present invention will become more readily apparent, a more particular description of the invention will be rendered by reference to the appended drawings and appended detailed description.
With the rapid development of modern internet life, social networks have become an integral part of the life of most people, and modern life is more and more separated from social networks. Each individual user node in the social network is producing data from time to time, and each social action will have an effect on the overall social network. The link prediction is used as a data processing task for mining potential relations in the social network, so that social media use experience of people can be greatly improved. At present, researchers mainly use static explicit information-node attributes and topology structures in a social network to conduct link prediction. However, in the actual application scenario of link prediction, these methods cannot achieve the expected effect. Because neither the node properties nor the topology is reliable information, this information can be hidden by human tampering. The link prediction method based on the node attribute and the topological structure is completely dependent on the reliability degree of the information, and the confidence degree of the information directly influences the final stability of the method. Therefore, the invention supports the link prediction task by utilizing the social influence propagation mode information based on dynamic reliability, and improves the context level perception to improve the link prediction efficiency. The significance and importance of the invention focus on the improvement of the traditional link prediction model based on node attribute or topological structure information, and the nodes are represented by using more reliable social influence propagation information, so that the problem of poor practical application effect of the traditional link prediction is fundamentally solved.
In order to solve the above technical problem, this embodiment provides a social network link prediction method, as shown in fig. 1 and fig. 2, which is as follows:
101: and acquiring a social network data set to be predicted. The social network data set to be predicted is a social network topological structure diagram where the user to be predicted is located.
102: And respectively inputting the social network data set to be predicted into topic-aware social influence models under different topics to obtain social influence probability distribution of the social network data set to be predicted under each topic. The model was proposed by Nicola Barbieri et al in 2012, intended to model social impact from historical social activities of a social network.
103: Clustering social influence probability distribution of the social network data set to be predicted under all topics to obtain social influence probability distribution of the social network data set to be predicted under each topic category.
104: And inputting the social influence probability distribution of the social network data set to be predicted and the social influence probability distribution of the social network data set to be predicted under each subject category into a trained graph neural network to obtain the social influence propagation modes of each user to be predicted under different identities. The identity includes an influencer and an influencer.
105: For any two users to be predicted, inputting the corresponding social influence propagation modes of the two users to be predicted under different identities into a similarity function to obtain similarity, determining a link prediction result between the two users to be predicted according to the similarity, and determining whether the two users are connected according to the link prediction result.
105 Is specifically: and (3) carrying out similarity comparison on influencers of the nodes and influencers' identities in the whole social activities (social influencer propagation modes) by utilizing final node representation (social influencer propagation modes of users to be predicted under different identities) of the graph neural network, and selecting Top-K candidate nodes as link prediction results.
In practical application, the method for determining the trained graph neural network comprises the following steps:
Acquiring a social network data set to be trained; the to-be-trained social network data set is a social network topological structure diagram among all training users.
And respectively inputting the social network data set to be trained into topic-aware social influence models under different topics to obtain social influence probability distribution of the social network data set to be trained under each topic.
Clustering social influence probability distributions of all topics of the social network data set to be trained to obtain social influence probability distributions of all topic categories of the social network data set to be trained,
Inputting the social influence probability distribution of the social network data set to be trained and the social network data set to be trained under each subject category into a graph neural network to train a propagation layer of the graph neural network until a loss function reaches a set condition, and determining that the graph neural network corresponding to the current propagation layer is a trained graph neural network; the loss function is constructed according to social influence propagation modes of users to be trained under different identities, which are output by the propagation layer of the graph neural network. The loss function is defined as: Gamma is the edge parameter.
In practical application, in order to improve training accuracy, various social network data sets to be trained can be obtained, and data cleaning works such as complementation, deletion and modification of nonsensical data or data noise in the social network data sets to be trained (deletion of values which are obviously wrong or significant and deletion of values which are obviously wrong or significant) are needed, and substantial time complementation is presumed by modifying values which are wrong but significant, such as time stamping, particularly, independent nodes (nodes which are not connected with other arbitrary nodes in the network and do not generate any social activities) which are ubiquitous in the social network are needed to be concerned, and removal of such nodes does not have any influence on a model link predictor, because influence propagation characteristics of such nodes at any moment are static and only calculation storage pressure of a model is increased. And considering that such users also belong to nonsensical users in the real world, they are a noise point for the whole system, which negatively affects the optimization of the model.
In practical application, the social influence probability distribution under each topic can be obtained as follows: taking a social network propagation log file (to-be-trained social network data set) for completing data cleaning work as input, firstly constructing the influence strength of each side in the social network under a specific topic by referring to a traditional subject perceived social influence model. The social network propagation log file with data cleaning is used as the input of a subject perception social influence model, and influence probability among users is obtained and used as the side influence intensity of the usersRepresenting the ith edge, two users are connected to form one edge. Thus, each edge and corresponding topic model gives a corresponding social impact probability. The whole influence propagation occurrence mode is as follows: when node v first gets active on item i, there is a chance to affect each inactive neighbor u, independent of the history so far. In short, when a user v generates a social action of a certain topic z at a certain time node t, and in a later time stamp, its neighboring node u also generates a social action of a corresponding topic, then the model considers that the influence intensity of the user node v on the node u under the topic z increases. The process uses a traditional topic-aware social influence propagation model (the influence intensity of each side is obtained according to the social influence probability of v on u under the topic z).
And obtaining the topic-level social influence probability distribution according to the edge influence intensity among the user nodes. After (similar to the side influence intensity being a scalar quantity, the social influence probability distribution is a vector), a topic-level social influence probability distribution is obtained(The influence strengths of each edge are combined together to form a vector to obtain a topic-level social influence probability distribution).
In practical application, the clustering is performed on social influence probability distributions of the to-be-trained social network data set under all topics to obtain social influence probability distributions of the to-be-trained social network data set under each topic category, specifically:
And calculating the difference value between the social influence probability distribution under any few topics for the social influence probability distribution under any few topics in the social network data set to be trained. And when the difference value is in the set range, dividing a plurality of topics corresponding to the difference value into the same topic category, and adding social influence probability distribution of a plurality of topics corresponding to the difference value to obtain the social influence probability distribution under the corresponding topic category. Topics (topics 1 to topics z) with similar social influence probability distribution can be clustered according to influence intensity distribution of each side under different topics to obtain corresponding context (context 1 to context K) level social influence propagation identifications (social influence probability distribution of topic categories). Defining a context-level social influence propagation identifier, wherein the expression is as follows:
wherein, Mu c refers to the social influence probability distribution of topic category c,/>Refers to the social influence probability of the first edge in the topic class c, and two users form one edge,/>Representing social influence probability of the |E| side in the topic class c,/>Representing social influence probability of the e-th edge in the topic category c,/>And (3) representing social influence probability of the e-th side on the topic z, wherein the topic category C is formed by clustering the topic z and other topics, and C represents a set formed by all topic categories.
Clustering is to minimize the difference in the individual topic impact probability distributions in the same context (topic category) from the context-level social impact propagation identities they constitute. The model utilizes KL divergence to define the difference between the topic-level social influence probability distribution and the context-level social influence propagation identification as:
The model ensures that topics in the same context have the same way of influence propagation by minimizing inter-topic differences in the context.
In practical applications, the multi-output graph neural network utilizes a context influence propagation mode to embed user nodes in the social network into corresponding context dependent spaces, and embeds the influencers or influencers roles in the social activities into different representation spaces according to the nodes. The multi-output graph neural network is mainly divided into an encoder and a propagation layer. The encoder is responsible for mapping the node and edge features (here edge features refer specifically to the impact strength distribution of the edge) into the initial node and edge vectors by separate mapping functions. The propagation layer aggregates the feature information around the node (in this embodiment, impact strength is specified), capturing the contextual social impact propagation modes of the node (user). The encoder adopts a multi-layer perceptron pre-training method to learn the initialization of nodes and edges. It should be noted that there are two roles (influencers and influencers) for nodes in a social network, and therefore each node will be initialized to two separate representations (social influencer propagation modes) separately. The propagation layer is mainly divided into information transfer and node update. The model improves the information transfer module of each propagation layer according to the influence propagation characteristics, and enlarges the range of local information aggregation of the node updating module. The main function of the information transfer step is to aggregate and transfer node information around a node to the node and update the node representation as a new input parameter. In this process, the node influence propagation mode changes with the role of the node, that is, when the node influence neighboring node and the node are influenced by the neighboring node, the content of the information transfer should also change. In addition, each node has a different degree of attention to the information conveyed by the different edges. The propagation layer in the trained graph neural network comprises an information transfer module and an updating module:
And an information transfer module:
Wherein ms v→u,c is the transfer information from the affected party v to the affected party u in the subject category, f MLP is a multi-layer perceptron,/> Influence propagation mode matrix for user u as influencer,/>For the social influence propagation mode under the topic category c with the user u as an influencer at the layer t of the graph neural network,/>In order to take the user u as the social influence propagation mode of the affected person under the subject category at the layer t of the graphic neural network, alpha v→u,c is attention, exp () is an exponential function based on a natural number e, and/>For the social influence probability of user u to user v in topic category c,/>For the social influence probability of the user u to the user v' in the topic category c, mt v→u,c is the transfer information of the affected user v to the user u,/>For user u as the influencer's influence propagation mode matrix, β v→u,c is the user u's attention to user v,/>For the social influence probability of user v to user u in topic category c,/>Social influence probability in topic category c for user v' to user u.
Node updates require local information around the node to be aggregated and updated as new input parameters. Conventional graph neural networks only focus on information of neighboring nodes, and other nodes except the neighboring nodes are ignored. This is not consistent with real life cognition and friends of friends still have an impact. Thus, the model modifies the node update module to increase the scope of local information aggregation:
the update module is as follows:
wherein/> In order to take a user u as an influencer in a social influence propagation mode of a topic class c at a layer t+1 of the graphic neural network, f LSTM is an LSTM function, theta 1 is a first parameter,/>In order to take the user u as the social influence propagation mode of the affected person in the topic category c in the layer t+1 of the graphic neural network, ms v'→v,c is the transmission information of the affected person v 'to the user v, θ 2 is a second parameter, and mt v'→v,c is the transmission information of the affected person v' to the user v.
In practical application, the similarity function is specifically:
Wherein S (u, v) is the similarity between user u and user v, K is the hyper-parameter, d (·) is the Euclidean distance between the vectors, S u,c is the social influence propagation mode of user u under the influencer identity in topic class c, S v,c is the social influence propagation mode of user v under the influencer identity in topic class c, T u,c is the social influence propagation mode of user u under the influencer identity in topic class c, T v,c is the social influence propagation mode of user v under the influencer identity in topic class c.
The embodiment also provides a social network link prediction system corresponding to the method, as shown in fig. 3, where the system includes:
the acquisition module A1 is used for acquiring a social network data set to be predicted, wherein the social network data set to be predicted is a social network topological structure diagram where a user to be predicted is located.
And the input module A2 is used for respectively inputting the social network data set to be predicted into topic-aware social influence models under different topics to obtain the social influence probability distribution of the social network data set to be predicted under each topic.
The topic category probability determining module A3 is used for clustering social influence probability distributions of the to-be-predicted social network data set under all topics to obtain the social influence probability distribution of the to-be-predicted social network data set under each topic category.
The social influence propagation mode determining module A4 is used for inputting the social influence probability distribution of the social network data set to be predicted and the social influence probability distribution of the social network data set to be predicted under each theme category into a trained graphic neural network to obtain the social influence propagation modes of each user to be predicted under different identities; the identity includes an influencer and an influencer.
And the link prediction result determining module A5 is used for inputting the social influence propagation modes of the two corresponding users to be predicted under different identities into a similarity function to obtain similarity for any two users to be predicted, and determining the link prediction result between the two users to be predicted according to the similarity.
As an alternative embodiment, the social influence propagation mode determining module includes:
The acquisition unit is used for acquiring a social network data set to be trained; the to-be-trained social network data set is a social network topological structure diagram among all training users.
The input unit is used for respectively inputting the social network data set to be trained into topic-aware social influence models under different topics to obtain social influence probability distribution of the social network data set to be trained under each topic.
The topic category probability determining unit is used for clustering social influence probability distribution of the to-be-trained social network data set under all topics to obtain the social influence probability distribution of the to-be-trained social network data set under each topic category.
The model determining unit is used for inputting the social network data set to be trained and the social influence probability distribution of the social network data set to be trained under each subject category into the graph neural network to train the propagation layer of the graph neural network until the loss function reaches a set condition, and determining that the graph neural network corresponding to the current propagation layer is a trained graph neural network; the loss function is constructed according to social influence propagation modes of users to be trained under different identities, which are output by the propagation layer of the graph neural network.
As an alternative embodiment, the topic category probability determining unit includes:
The difference value calculation subunit is used for calculating the difference value among social influence probability distributions under any several topics for the social influence probability distribution under any several topics in the social network data set to be trained;
And the theme category probability determination subunit is used for dividing a plurality of themes corresponding to the difference value into the same theme category when the difference value is in a set range, and adding social influence probability distribution of the themes corresponding to the difference value to obtain the social influence probability distribution under the corresponding theme category.
As an optional implementation manner, the propagation layer in the trained graph neural network includes an information transfer module and an update module:
And an information transfer module:
Wherein ms v→u,c is the transfer information from the affected party v to the affected party u in the subject category, f MLP is a multi-layer perceptron,/> Influence propagation mode matrix for user u as influencer,/>For the social influence propagation mode under the topic category c with the user u as an influencer at the layer t of the graph neural network,/>In order to take the user u as the social influence propagation mode of the affected person under the subject category at the layer t of the graphic neural network, alpha v→u,c is attention, exp () is an exponential function based on a natural number e, and/>For the social influence probability of user u to user v in topic category c,/>For the social influence probability of the user u to the user v' in the topic category c, mt v→u,c is the transfer information of the affected user v to the user u,/>For user u as the influencer's influence propagation mode matrix, β v→u,c is the user u's attention to user v,/>For the social influence probability of user v to user u in topic category c,/>Social influence probability in topic category c for user v' to user u.
And an updating module:
wherein/> In order to take a user u as an influencer in a social influence propagation mode of a topic class c at a layer t+1 of the graphic neural network, f LSTM is an LSTM function, theta 1 is a first parameter,/>In order to take the user u as the social influence propagation mode of the affected person in the topic category c in the layer t+1 of the graphic neural network, ms v'→v,c is the transmission information of the affected person v 'to the user v, θ 2 is a second parameter, and mt v'→v,c is the transmission information of the affected person v' to the user v.
As an alternative embodiment, the similarity function is specifically:
Wherein S (u, v) is the similarity between user u and user v, parameter K is the super-parameter, d (·) is the Euclidean distance between the vectors, S u,c is the social influence propagation mode of user u under the influencer identity in topic class c, S v,c is the social influence propagation mode of user v under the influencer identity in topic class c, T u,c is the social influence propagation mode of user u under the influencer identity in topic class c, and T v,c is the social influence propagation mode of user v under the influencer identity in topic class c.
The following describes the prediction method provided in this embodiment in further detail with reference to a more specific example, where the configuration environment of this example is as follows: CPU 8700K main frequency 3.7GHz, ROM 16G, graphic computing card NVIDIA GTX2080Ti, linux Ubuntu 18.04 system, programming language Python 3, based on Pytorch deep learning framework.
Firstly, testing the context cluster analysis of a model, wherein in a context topic perception social influence model, the clustering quantity |K| of contexts is a group of super-parameters, and the meaning is that the original topic set is divided into a coarse-granularity context level set. Aggregating topics with similar impact propagation identities into one context can effectively reduce unnecessary computational loss. However, too large a granularity of division will cause the model to lose a lot of topic information and thus not capture the propagation characteristics of social influence. Selecting a proper granularity of context partitioning can ensure computational efficiency while ensuring accuracy of predictions. Thus, the present invention performed adjustment test analysis on multiple sets of |k| parameters. Analysis finds that when |K| <5, context granularity is large and model performance is poor. Model performance is also continuously improved as |k| increases. When the I K is larger, the topic influence propagation modes in each context have high similarity, and the difference between different contexts is large. Otherwise context will lose implicit information. When |k| > =5, the model performance tends to stabilize and additional context calculations do not improve the performance of the model. On all three data sets, |k|=5 best balances efficiency and performance.
The four conditions of |k|=4, 6, 8 and 10 are selected for specific analysis, and the analysis shows that when |k|=4, infant care, catering, politics, science and technology and education are divided into the same context, the topics have a certain similarity in terms of life experience, but still have great difference from each other. When |k|=6, infant, education, and science are divided into unified contexts, and subjects that are not well related to these are divided into other contexts. When |k|=10, the infant and education are divided into different contexts, but in real life, these two and similar topics should be divided into the same context.
As shown in fig. 4, we evaluate the performance of the method on three datasets Digg, lastFM and Flixster, following the complete operational flow of the method. In addition, we also compare the performance differences of the present invention with five other classical link prediction algorithms, common Neighbours (CN), PREFERENTIAL ATTACHMENT (PA), random WALK WITH RESTART (RWR), FRIENDLINK (FL) and Path and Node Combined (PNC), respectively. Firstly, 1000 existing edges in the original data set are extracted to be used as a test set, and 1000 corresponding corrosion edges are generated by using the 1000 existing edges and added into the test set together. The F1 score for this method was calculated on different data sets and compared to other baseline methods. Experimental results show that the method achieves the optimal link prediction effect, the F1 score is 0.877, and the F1 score is improved by 21.7% compared with the F1 score of other best baseline method path node joint (PNC) models.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
1. According to the method, the propagation logs generated by social activities of users in the social network are utilized, the influence intensity of each side in the social network under different topics is generated based on the topic-level influence propagation model, influence propagation identifiers for representing the topics are constructed in a distributed mode according to the influence probability under the same topic, and the discrete topics in the social network are quantitatively considered.
2. The invention provides a method for clustering topics similar to influence propagation into a unified context-level social influence optimization model. The social influence of the context level greatly reduces redundant calculation on the premise of ensuring that the social influence information is not lost, and the prediction efficiency of the model is effectively improved. Compared with the traditional link prediction method, the social influence propagation mode can be utilized to mine more hidden relations and improve the accuracy of link prediction.
3. According to the invention, through analyzing the behaviors of the users in the social network, two different social roles, namely an influencer and an influencer, are designed for all the users, and a multi-output graph neural network model is provided for modeling the two identities respectively. Because the modeling method is very fit with the real world, the final prediction effect is obviously improved, and the model method using single identity operation is far superior.
4. The invention provides a method for enhancing link prediction by utilizing a social influence propagation mode, wherein the social influence propagation mode is dynamic and reliable information, has good authenticity and safety, can help a model to capture potential relations in a social network more accurately, and greatly improves link prediction performance.
In the present specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different point from other embodiments, and identical and similar parts between the embodiments are all enough to refer to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant points refer to the description of the method section.
The principles and embodiments of the present invention have been described herein with reference to specific examples, the description of which is intended only to assist in understanding the methods of the present invention and the core ideas thereof; also, it is within the scope of the present invention to be modified by those of ordinary skill in the art in light of the present teachings. In view of the foregoing, this description should not be construed as limiting the invention.
Claims (8)
1. A method for predicting a social network link, comprising:
Acquiring a social network data set to be predicted, wherein the social network data set to be predicted is a social network topological structure diagram where a user to be predicted is located;
Respectively inputting the social network data set to be predicted into topic-aware social influence models under different topics to obtain social influence probability distribution of the social network data set to be predicted under each topic;
Clustering social influence probability distributions of all topics of the social network data set to be predicted to obtain social influence probability distributions of all topic categories of the social network data set to be predicted;
Inputting the social influence probability distribution of the social network data set to be predicted and the social influence probability distribution of the social network data set to be predicted under each subject category into a trained graph neural network to obtain social influence propagation modes of each user to be predicted under different identities; the identity comprises an influencer and an influencer;
For any two users to be predicted, inputting social influence propagation modes of the corresponding two users to be predicted under different identities into a similarity function to obtain similarity, and determining a link prediction result between the two users to be predicted according to the similarity; the similarity function is specifically:
Wherein S (u, v) is the similarity between user u and user v, K is the hyper-parameter, d (·) is the Euclidean distance between the vectors, S u,c is the social influence propagation mode of user u under the influencer identity in topic class c, S v,c is the social influence propagation mode of user v under the influencer identity in topic class c, T u,c is the social influence propagation mode of user u under the influencer identity in topic class c, T v,c is the social influence propagation mode of user v under the influencer identity in topic class c.
2. The social network link predicting method according to claim 1, wherein the trained graph neural network determining method is as follows:
Acquiring a social network data set to be trained; the to-be-trained social network data set is a social network topological structure diagram among all training users;
Respectively inputting the social network data set to be trained into topic perception social influence models under different topics to obtain social influence probability distribution of the social network data set to be trained under each topic;
Clustering social influence probability distributions of the to-be-trained social network data set under all topics to obtain social influence probability distributions of the to-be-trained social network data set under each topic category;
Inputting the social influence probability distribution of the social network data set to be trained and the social network data set to be trained under each subject category into a graph neural network to train a propagation layer of the graph neural network until a loss function reaches a set condition, and determining that the graph neural network corresponding to the current propagation layer is a trained graph neural network; the loss function is constructed according to social influence propagation modes of users to be trained under different identities, which are output by the propagation layer of the graph neural network.
3. The social network link prediction method according to claim 2, wherein the clustering of the social influence probability distributions of the to-be-trained social network data set under all topics is performed to obtain the social influence probability distribution of the to-be-trained social network data set under each topic category, specifically:
for social influence probability distribution under any several topics in the social network data set to be trained, calculating the difference value between the social influence probability distribution under the several topics;
and when the difference value is in the set range, dividing a plurality of topics corresponding to the difference value into the same topic category, and adding social influence probability distribution of a plurality of topics corresponding to the difference value to obtain the social influence probability distribution under the corresponding topic category.
4. The social network link predicting method according to claim 1, wherein the propagation layer in the trained graph neural network comprises an information transfer module and an updating module;
The information transfer module is
Wherein ms v→u,c is the transfer information from the affected party v to the affected party u in the subject category, f MLP is a multi-layer perceptron,/>Influence propagation mode matrix for user u as influencer,/>For the social influence propagation mode under the topic category c with the user u as an influencer at the layer t of the graph neural network,/>In order to take the user u as the social influence propagation mode of the affected person under the subject category at the layer t of the graphic neural network, alpha v→u,c is attention, exp () is an exponential function based on a natural number e, and/>For the social influence probability of user u to user v in topic category c,/>For the social influence probability of the user u to the user v' in the topic category c, mt v→u,c is the transfer information of the affected user v to the user u,/>For user u as the influencer's influence propagation mode matrix, β v→u,c is the user u's attention to user v,/>For the social influence probability of user v to user u in topic category c,/>Social influence probability of the subject class c for the user v' to the user u;
The updating module is
Wherein/>In order to take a user u as an influencer in a social influence propagation mode of a topic class c at a layer t+1 of the graphic neural network, f LSTM is an LSTM function, theta 1 is a first parameter,/>In order to take the user u as the social influence propagation mode of the affected person in the topic category c in the layer t+1 of the graphic neural network, ms v'→v,c is the transmission information of the affected person v 'to the user v, θ 2 is a second parameter, and mt v'→v,c is the transmission information of the affected person v' to the user v.
5. A social networking link prediction system, comprising:
The acquisition module is used for acquiring a social network data set to be predicted, wherein the social network data set to be predicted is a social network topological structure diagram where a user to be predicted is located;
The input module is used for respectively inputting the social network data set to be predicted into topic-aware social influence models under different topics to obtain social influence probability distribution of the social network data set to be predicted under each topic;
The topic category probability determining module is used for clustering social influence probability distribution of the to-be-predicted social network data set under all topics to obtain social influence probability distribution of the to-be-predicted social network data set under each topic category;
The social influence propagation mode determining module is used for inputting the social influence probability distribution of the social network data set to be predicted and the social influence probability distribution of the social network data set to be predicted under each subject category into the trained graph neural network to obtain the social influence propagation modes of each user to be predicted under different identities; the identity comprises an influencer and an influencer;
The link prediction result determining module is used for inputting the social influence propagation modes of the two corresponding users to be predicted under different identities into a similarity function to obtain similarity for any two users to be predicted, and determining a link prediction result between the two users to be predicted according to the similarity; the similarity function is specifically:
Wherein S (u, v) is the similarity between user u and user v, K is the hyper-parameter, d (·) is the Euclidean distance between the vectors, S u,c is the social influence propagation mode of user u under the influencer identity in topic class c, S v,c is the social influence propagation mode of user v under the influencer identity in topic class c, T u,c is the social influence propagation mode of user u under the influencer identity in topic class c, T v,c is the social influence propagation mode of user v under the influencer identity in topic class c.
6. The social network link predicting system of claim 5, wherein the social influence propagation mode determining module comprises:
The acquisition unit is used for acquiring a social network data set to be trained; the to-be-trained social network data set is a social network topological structure diagram among all training users;
The input unit is used for respectively inputting the social network data set to be trained into topic-aware social influence models under different topics to obtain social influence probability distribution of the social network data set to be trained under each topic;
The topic category probability determining unit is used for clustering social influence probability distribution of the to-be-trained social network data set under all topics to obtain social influence probability distribution of the to-be-trained social network data set under each topic category;
The model determining unit is used for inputting the social network data set to be trained and the social influence probability distribution of the social network data set to be trained under each subject category into the graph neural network to train the propagation layer of the graph neural network until the loss function reaches a set condition, and determining that the graph neural network corresponding to the current propagation layer is a trained graph neural network; the loss function is constructed according to social influence propagation modes of users to be trained under different identities, which are output by the propagation layer of the graph neural network.
7. The social network link predicting system according to claim 6, wherein said topic category probability determining unit includes:
The difference value calculation subunit is used for calculating the difference value among social influence probability distributions under any several topics for the social influence probability distribution under any several topics in the social network data set to be trained;
And the theme category probability determination subunit is used for dividing a plurality of themes corresponding to the difference value into the same theme category when the difference value is in a set range, and adding social influence probability distribution of the themes corresponding to the difference value to obtain the social influence probability distribution under the corresponding theme category.
8. The social network link predicting system according to claim 5, wherein the propagation layer in the trained graph neural network comprises an information delivery module and an update module:
The information transfer module is
Wherein ms v→u,c is the transfer information from the affected party v to the affected party u in the subject category, f MLP is a multi-layer perceptron,/>Influence propagation mode matrix for user u as influencer,/>For the social influence propagation mode under the topic category c with the user u as an influencer at the layer t of the graph neural network,/>In order to take the user u as the social influence propagation mode of the affected person under the subject category at the layer t of the graphic neural network, alpha v→u,c is attention, exp () is an exponential function based on a natural number e, and/>For the social influence probability of user u to user v in topic category c,/>For the social influence probability of the user u to the user v' in the topic category c, mt v→u,c is the transfer information of the affected user v to the user u,/>For user u as the influencer's influence propagation mode matrix, β v→u,c is the user u's attention to user v,/>For the social influence probability of user v to user u in topic category c,/>Social influence probability of the subject class c for the user v' to the user u;
The updating module is
Wherein/>In order to take a user u as an influencer in a social influence propagation mode of a topic class c at a layer t+1 of the graphic neural network, f LSTM is an LSTM function, theta 1 is a first parameter,/>In order to take the user u as the social influence propagation mode of the affected person in the topic category c in the layer t+1 of the graphic neural network, ms v'→v,c is the transmission information of the affected person v 'to the user v, θ 2 is a second parameter, and mt v'→v,c is the transmission information of the affected person v' to the user v.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110101233.0A CN112765489B (en) | 2021-01-26 | 2021-01-26 | Social network link prediction method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110101233.0A CN112765489B (en) | 2021-01-26 | 2021-01-26 | Social network link prediction method and system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112765489A CN112765489A (en) | 2021-05-07 |
CN112765489B true CN112765489B (en) | 2024-04-19 |
Family
ID=75707399
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110101233.0A Active CN112765489B (en) | 2021-01-26 | 2021-01-26 | Social network link prediction method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112765489B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113705075B (en) * | 2021-07-07 | 2024-01-19 | 西北大学 | Social relationship analysis method based on graph neural network |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103678669A (en) * | 2013-12-25 | 2014-03-26 | 福州大学 | Evaluating system and method for community influence in social network |
US20170185908A1 (en) * | 2014-09-18 | 2017-06-29 | Huawei Technologies Co.,Ltd. | Method and device for predicting information propagation in social network |
CN112115377A (en) * | 2020-09-11 | 2020-12-22 | 安徽农业大学 | Graph neural network link prediction recommendation method based on social relationship |
-
2021
- 2021-01-26 CN CN202110101233.0A patent/CN112765489B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103678669A (en) * | 2013-12-25 | 2014-03-26 | 福州大学 | Evaluating system and method for community influence in social network |
US20170185908A1 (en) * | 2014-09-18 | 2017-06-29 | Huawei Technologies Co.,Ltd. | Method and device for predicting information propagation in social network |
CN112115377A (en) * | 2020-09-11 | 2020-12-22 | 安徽农业大学 | Graph neural network link prediction recommendation method based on social relationship |
Also Published As
Publication number | Publication date |
---|---|
CN112765489A (en) | 2021-05-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2023065545A1 (en) | Risk prediction method and apparatus, and device and storage medium | |
CN107292097A (en) | The feature selection approach of feature based group and traditional Chinese medical science primary symptom system of selection | |
WO2023207013A1 (en) | Graph embedding-based relational graph key personnel analysis method and system | |
CN114743037A (en) | Deep medical image clustering method based on multi-scale structure learning | |
CN112765489B (en) | Social network link prediction method and system | |
Zhao et al. | Personalized federated few-shot learning | |
Ezequiel et al. | Federated learning for privacy-aware human mobility modeling | |
Wang et al. | Context-aware mobile service adaptation via a Co-evolution eXtended Classifier System in mobile network environments | |
CN112668633B (en) | Adaptive graph migration learning method based on fine granularity field | |
Li et al. | Privacy protection in prosumer energy management based on federated learning | |
CN116307078A (en) | Account label prediction method and device, storage medium and electronic equipment | |
CN114266352B (en) | Model training result optimization method, device, storage medium and equipment | |
US11726862B1 (en) | Over-the-air machine learning | |
CN113360772B (en) | Interpretable recommendation model training method and device | |
CN115577797A (en) | Local noise perception-based federated learning optimization method and system | |
US11923074B2 (en) | Professional network-based identification of influential thought leaders and measurement of their influence via deep learning | |
Liu et al. | Towards dynamic reconfiguration of composite services via failure estimation of general and domain quality of services | |
CN114898184A (en) | Model training method, data processing method and device and electronic equipment | |
Liu et al. | srvpa: A multi-domain conversational service recommendation approach | |
CN113010759A (en) | Processing method and device of cluster set, computer readable medium and electronic equipment | |
CN112396237A (en) | Link prediction method in social network | |
Zhou et al. | Improved extension neural network and its applications | |
CN117573904B (en) | Multimedia teaching resource knowledge graph generation method and system based on recognition analysis | |
Zhong et al. | Lightweight Federated Graph Learning for Accelerating Classification Inference in UAV-assisted MEC Systems | |
Gao et al. | A contextual reasoning research based on the uncertain context |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |