CN110147911B - Social influence prediction model and prediction method based on content perception - Google Patents
Social influence prediction model and prediction method based on content perception Download PDFInfo
- Publication number
- CN110147911B CN110147911B CN201910452492.0A CN201910452492A CN110147911B CN 110147911 B CN110147911 B CN 110147911B CN 201910452492 A CN201910452492 A CN 201910452492A CN 110147911 B CN110147911 B CN 110147911B
- Authority
- CN
- China
- Prior art keywords
- network
- user
- layer
- social media
- media content
- 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 22
- 230000008447 perception Effects 0.000 title claims abstract description 12
- 230000006870 function Effects 0.000 claims abstract description 9
- 239000011159 matrix material Substances 0.000 claims description 29
- 230000003068 static effect Effects 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 3
- 238000011160 research Methods 0.000 abstract description 3
- 238000003012 network analysis Methods 0.000 abstract description 2
- 238000004364 calculation method Methods 0.000 abstract 1
- 230000006399 behavior Effects 0.000 description 11
- 238000011156 evaluation Methods 0.000 description 9
- 238000002474 experimental method Methods 0.000 description 8
- 230000003542 behavioural effect Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000011273 social behavior Effects 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
Images
Classifications
-
- 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
- G06N3/045—Combinations of 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- 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
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- 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)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Economics (AREA)
- General Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Molecular Biology (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Primary Health Care (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a social influence prediction model and a prediction method based on content perception, which relate to the field of social network analysis and comprise the following steps: and performing feature representation on network nodes by using a graph convolution network, representing social media content corresponding to each node by using a long-short term memory network (LSTM), fusing feature vector representations of the nodes and the social media content by using an RN model combined with the social media content, performing inference calculation, and comparing the obtained output with a real label so as to optimize a target function. The invention uses the graph convolution network to carry out feature representation on the network node, on one hand, the attribute information of the node can be better integrated, and on the other hand, the invention can also predict the new node in the network; meanwhile, content information of social media is considered during modeling, positive and negative samples are divided according to social media content, and the problem of deep research caused by sparse observation data is effectively solved.
Description
Technical Field
The invention relates to the field of social network analysis, in particular to a social influence prediction model and a prediction method based on content perception.
Background
The social influence is embodied by social activities among users and is represented by the phenomenon that the behaviors, ideas and the like of the users are influenced by others and changed. Online social network influence analysis mainly involves three aspects of content: 1. the identification of the influence is realized, and how to distinguish and link the influence and related elements from complicated factors is identified; 2. measuring social influence, and designing and selecting a measuring method which has certain universality and can fully discover characteristics of a social network; 3. dynamic propagation of social influence.
At present, when network nodes are represented, node attribute information cannot be integrated, new nodes in a network cannot be predicted, meanwhile, the existing modeling prediction on social influence is generally based on a neural network, content information of social media is not considered, the existing observation data is generally divided into positive and negative samples by taking a user as a main body, the observation data is sparse, and effective analysis is difficult to perform.
Disclosure of Invention
The embodiment of the invention provides a social influence prediction model and a prediction method based on content perception, which are used for solving the problems in the prior art.
A social influence prediction model and a prediction method based on content perception comprise the following steps:
an input layer for inputting information, the input information comprising: user network structure information, user network attribute information and user social media content information;
a GCN layer for extracting the user network structure information and the user network attribute information inputted from the input layer;
the Embedding layer is used for storing the D-dimensional Embedding representation of the user extracted from the GCN layer;
an LSTM layer for representing the user social media content input by the input layer using an LSTM algorithm;
the RN layer is used for putting the Embedding representation output by the Embedding layer and the social media content Embedding representation output by the LSTM layer into an RN model, and deducing whether the current node is influenced by the neighbor nodes under the given social media content by utilizing the RN model;
the output layer is used for representing the result of model prediction in a two-dimensional mode when the content of the given social media is influenced by the neighbor nodes of the given social media;
further, still include: and the Grountreth layer is used for comparing the result predicted by the model with the Grountreth layer so as to optimize the objective function.
Further, in the input layer, each instance information in the input layer is an r-proxy network composed of n users.
Further, in the GCN layer, the GCN layer receives as input m user network configuration information and m n × f user network attribute information.
Further, in the GCN layer, the user network structure information is calculated from an adjacency matrix a of the network, and the GCN layer instantiates the adjacency matrix a of the network and a user network attribute information matrix H as static matrices closely related to a standardized graph laplacian:
wherein,
a is the adjacency matrix of the network graph G, I is the identity matrix,
the diagonal matrix of the network node value is obtained, and after A (G) is obtained, the matrix and a user network attribute information matrix H are used as the input of each GCN layer together for extracting the network node characteristics.
Further, in the LSTM layer, the LSTM layer receives as input m p × q user social media content information, p referring to the number of words in the social media content information, and q referring to the initial dimension of each word.
A prediction method based on the social influence prediction model based on content perception specifically comprises the following steps:
firstly, sampling to obtain a sub-network with a fixed size, and taking the sub-network as a proxy network of each r-ego network, wherein the proxy network comprises user network structure information, user network attribute information and user social media content information;
and step two, the agent network in the step one is put into a social influence prediction model based on content perception for prediction, and a corresponding prediction result is obtained.
Further, still include:
and step three, comparing the output result in the step two with a real result, calculating loss and minimizing a negative log-likelihood function.
The invention has the beneficial effects that: the invention uses the graph convolution network to carry out feature representation on the network node, on one hand, the attribute information of the node can be better integrated, and on the other hand, the invention can also predict a new node in the network; meanwhile, content information of social media is considered during modeling, positive and negative samples are divided according to social media content, and the problem of deep research caused by sparse observation data is effectively solved.
Drawings
Fig. 1 is a schematic diagram of a social influence prediction algorithm model framework based on content awareness for a social influence prediction model and a prediction method based on content awareness according to an embodiment of the present invention;
fig. 2 is a schematic diagram of an RN model framework of a social influence prediction model and a prediction method based on content awareness according to an embodiment of the present invention;
fig. 3 is a schematic diagram of 1-neighbor/2-neighbor of a social influence prediction model and a prediction method based on content awareness according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention are clearly and completely described below with reference to the drawings in the embodiments of the present invention, but it should be understood that the scope of the present invention is not limited by the specific embodiments.
Referring to fig. 1-3, the present invention provides a social influence prediction model based on content perception, which is first defined as follows:
definition 1: r-neighbors (r-neighbors) and r-ego network
Given a social network G ═ (V, E), where G is a static and undirected network, V denotes the set of users in the social network, and E denotes the set of relationships of the users. For a given central user v, its r-neighbors are defined as
Where d (u, v) refers to the distance between user u and user vThe shortest distance in terms of hop count. As shown in FIG. 1, the dark node is taken as the center node, and the 1-neighbor and 2-neighbor of the dark node are shown respectively. The r-ego network of user v is composed of
A network of nodes of
And (4) showing.
Definition 2: r-proxy network
Given a central user v, the present invention extracts its r-ego network using a Breadth-First-Search (BFS) method. Since different users have different neighbor sizes, and the social network data with different sizes are not suitable for most deep learning models, N nodes are sampled from the r-neighbor network of the user v as a representative, which is called an r-proxy network.
Definition 3: social behavior (Social Action)
Social behavior refers to social actions performed by users in a social network on social media content, such as, for example, actions of following, like, forwarding, voting, etc. Given a behavior log A, A contains a series of triples
Indicating that user u is at time
I-actions have been done on a certain message or book. Observing the state of user v based on time can characterize it as a binary variable,
wherein, t represents the time of day,
indicating that user v produced some behavior on the ith social media content before time t,
it indicates that the user did not act on the ith social media content.
Definition 4: social media content (Social content)
Social media content refers in the present invention to objects in a social network that a user generates behavior, such as a log, a tweet, a story, and the like. Given social media content C and user v, the present invention is concerned with the impact that nodes in the r-proxy network of user v have on social media content C.
Problem definition: local Social Influence (Social affinity Locality)
Given the r-ego network of node v, the behavioral state of users in the r-ego network, and the specific social media content, the local social impact models the probability of node v behavioral state under these conditions. I.e. given an r-ego network
r-ego behavioral states of users in a network
Social media content C and time interval Δ t, local social impact aims to quantify the probability that node v produces C behavior:
assume that there are N instances, each instance is a triple (v, a, t), where v refers to the user, a refers to social behavior, and t refers to a timestamp. For such a triplet (v, a, t), we know the r-ego network-where v is located
r-ego behavioral states of users in a network
And social media content —C, this problem can be expressed as a binary graph classification problem, which can be solved by minimizing the following negative log-likelihood objective (model parameters Θ):
it should be noted that in this work it is assumed that Δ t is sufficiently large.
A social influence prediction model based on content awareness comprising:
an input layer: for inputting information, the inputted information comprising: user network structure information, user network attribute information and user social media content information;
each instance in the input is an r-proxy network consisting of n users, wherein n refers to the number of nodes in the r-proxy network, and the r-proxy network comprises three aspects of information: network structure information, network attribute information (specific content-see attribute information collection section), and social media content information.
The GCN receives m n × n network structure information and m n × f network attribute information as input, where m refers to the size of batch training;
the long-short term memory network LSTM accepts as input m p × q social media content information, where p refers to the number of words in the social media content information and q refers to the initial dimension of each word.
A GCN layer: for extracting user network structure information and user network attribute information inputted from the input layer;
given a user instance, the goal of the GCN layer is to learn a feature mapping function on the instance graph. For the algorithm presented here, a 3-layer cascaded GCN model is used in this section, with two parts for each layer input: example graph structure information and example graph network attribute information. The structure information of the example graph is calculated from the adjacency matrix of the network.
Given the adjacency matrix a and the user-network-attribute-information matrix H of the example graph, the GCN instantiates this as a static matrix closely related to the normalized graph laplacian:
herein, the
A is the adjacency matrix of the network graph G, I is the identity matrix,
is a diagonal matrix of network node values. And after A (G) is obtained, the matrix and the attribute information matrix H are used as the input of each GCN layer for extracting the network node characteristics.
Embedding layer: d dimension Embedding representation for storing n users in the instance extracted from GCN layer, which is D value is 128.
LSTM layer: the system comprises a display module, a display module and a display module, wherein the display module is used for representing user social media content corresponding to each node input by an input layer by using an LSTM algorithm;
given a piece of social media content information, the content information is first searched for the word Embedding representation in the content information (300-dimensional word Embedding representation is used herein), and then the Embedding matrix of the words is used as the input of the LSTM algorithm. For social media content herein, its LSTM output dimension is 128.
RN layer: after GCN, each node has its own Embedding representation, after LSTM represents, each social media content also has its own Embedding representation, the node Embedding representation output by the Embedding layer and the social media content Embedding representation output by the LSTM layer are put into an RN model, and the RN model is used for deducing whether the current node is influenced by the neighbor node under the given social media content;
as shown in fig. 2, the first column of objects is set as the neighborhood of the central user in the proxy network, the second column of objects is the central user itself, and the third column represents social media content, and the mathematical expression of the problem is as follows:
where o refers to the user's Embedding representation, c i Refers to a time series representation of the ith social media content,
it is referred to whether the neighbors of the central user have acted on the social media content i before time t. Since it is assumed herein that the behavior of the central user is only influenced by active neighbor users, the state (node, social media content) matrix and the historical behavior state of the user are subjected to inner product operation in the RN model, i.e. it is determined whether each user plays a role in the following reasoning process according to its state. When in use
It means that node u has generated behavior before time t for the ith social media content, otherwise it has not generated behavior.
The specific description of the RN model algorithm in conjunction with social media content is as follows:
input, wherein the Embedding of nodes in the agent network represents O, the time sequence of social media content represents C and the historical state S of users in the agent network
Output the outcome of the prediction of social influence
An output layer: outputting the result predicted by the two-dimensional representation model;
group treuth: and comparing the result of model prediction with the GrountTruth so as to optimize the objective function.
The content-aware social influence model algorithm is described in detail as follows:
input: an adjacency matrix A of a proxy network where self-users are located, an attribute information matrix H of the users, social media contents C and historical behavior states S of self-user neighbors in the proxy network
Output: results of nodes being affected
A prediction method of a social influence prediction model based on content perception specifically comprises the following steps:
firstly, sampling to obtain a sub-network with a fixed size, and using the sub-network as a proxy network of each r-ego network, wherein the proxy network comprises user network structure information, user network attribute information and user social media content information;
given a central user v, the text extracts its r-ego network using a Breadth-First-Search (BFS) method.
Since different users have different neighbor sizes, such different size network data is not suitable for most deep learning models. In order to solve this problem,
a sub-network of fixed size is sampled from the r-ego network of user v as a proxy network for the r-ego network.
There are studies that show that people are more likely to be affected by active neighbors than inactive neighbors, and therefore, more attention is paid to the active r-neighbors of user v in this text in choosing the neighbor portion.
The method comprises the following steps: all active r-neighbors of user v are taken out, and then M, which is the number of users of the proxy network-the number of active users, is randomly taken out of the remaining inactive r-neighbors. By using
Indicating ego the proxy network where the user is located,
representing the state of behaviour of the user v neighbours in the proxy network, the purpose of this equationThe objective function may be redefined as:
and step two, the agent network in the step one is put into a social influence prediction model based on content perception for prediction, and a corresponding prediction result is obtained.
And step three, comparing the output result in the step two with a real result, calculating loss and minimizing a negative log-likelihood function.
Evaluation indexes are as follows: to test the effectiveness of the social influence prediction method presented herein, the values of precision, recall, and F1 were used in the experimental section to evaluate the experimental results, respectively.
The precision ratio is as follows: the precision ratio is a measurement standard of the proportion of the data correctly predicted as the positive sample by the evaluation classifier to the data correctly predicted as the positive sample, and the mathematical expression of the precision ratio is as follows:
the TP is the condition that the positive sample is divided into the positive samples by the classifier and the positive samples are correctly divided by the corresponding classifier; and the FP is the condition that the classifier divides the negative samples into the negative samples and the corresponding classifier divides the negative samples correctly.
And (4) recall rate: a scale criterion for evaluating the positive samples divided by the classifier into positive samples is mathematically expressed as follows:
FN here is the case where the classifier divides positive samples into negative samples.
F1: an index for measuring the accuracy of the two classification models gives consideration to both the accuracy rate and the recall rate in the evaluation standard, and is a comprehensive evaluation index. The mathematical expression of the F1 evaluation index is as follows:
and (3) comparison algorithm: to verify the validity of the model proposed herein, the DeepInf-GCN algorithm, an algorithm published in 2018 on KDD that also uses deep neural networks for social impact prediction, was used herein as a comparison algorithm.
Experimental results and analysis: in order to check the effectiveness of the social influence prediction method proposed herein, data of 10 social media contents are found out on the DIGG dataset, respectively, so as to verify the content-aware social influence prediction algorithm proposed herein, and a deepnf-GCN algorithm is selected as a comparison algorithm. The evaluation criteria of the experiment were accuracy, recall and F1 values, respectively. In order to verify the effectiveness of the model comprehensively and accurately, the results are verified in two groups of experiments.
The first set of experiments: content data for 10 different social media were extracted from the Digg dataset and, assuming that their propagation patterns were the same, trained together and predicted separately.
Table 1 comparison of data sets experimental results for different contents in Digg (assuming the same propagation pattern) (%)
The test results of the content-aware social influence prediction method and the deep-GCN algorithm on 10 data sets are shown in table 1, and the experimental results are analyzed with respect to the performances of the two algorithms on each data set in terms of three evaluation indexes, namely precision, recall and F1 values. Given the same propagation pattern of these data, these data sets are trained together and predicted separately herein. From the aspect of the performance of 10 data sets on the precision rate, the algorithm provided by the invention only has no good performance of a comparison algorithm on two data sets, namely Content4 and Content10, and is superior to or equal to the comparison algorithm on the other eight data sets; from the aspect of recall rate, the algorithm provided by the invention is weaker than the comparison algorithm only on the Content4 data set, and the performance on other nine data sets is better than or equal to the comparison algorithm; from the value of F1, the algorithm presented herein is weaker than the comparison algorithm only on two data sets, Content4 and Content10, and performs better than or equal to the comparison algorithm on the other eight data sets.
Table 2 is a comparison of the relative gains of the two algorithms on each evaluation index.
Table 2 relative gain comparison (%)% of the two algorithms on each evaluation index for the first set of experiments
The second set of experiments: the data sets are still the data sets with 10 different contents, and the data sets are trained and predicted respectively on the assumption that the propagation modes of the data sets are different.
Table 3 comparison of the results of the experiments on the data sets of different contents of Digg (assuming different propagation patterns) (%)
The results of the second set of experiments are shown in table 3. Given the different propagation modes of these data sets, the force-of-influence parameters to be captured for different data sets will also be different. From this point of view, the above 10 data sets were also compared with their respective predicted results after training. The results of 10 data sets were analyzed: firstly, on the precision rate, the results of the algorithm proposed herein on 7 data sets are superior to the comparison algorithm; secondly, the performance of the algorithm provided by the method is superior to or equal to that of a comparison algorithm on the recall ratio except for results on two data sets, namely Content6 and Content 10; finally, for the composite index value of F1, the algorithm presented herein outperforms the comparison algorithm in all 10 data sets, and the highest relative gain value is 16.67%.
Table 4 is a comparison of the relative gains of the two algorithms on each index.
TABLE 4 relative gain comparison (%)
The effectiveness of the algorithm provided by the invention can be verified by combining the results of the two groups of experiments, and in addition, the algorithm provided by the invention also solves the problem that the prediction result cannot be explained.
In summary, the invention uses the graph convolution network to perform feature representation on the network node, so that on one hand, the attribute information of the node can be better integrated, and on the other hand, the invention can predict a new node in the network; meanwhile, content information of social media is considered during modeling, positive and negative samples are divided according to social media content, and the problem of deep research caused by sparse observation data is effectively solved.
The above disclosure is only one specific embodiment of the present invention, however, the present invention is not limited thereto, and any modifications that can be made by those skilled in the art should fall within the protection scope of the present invention.
Claims (8)
1. A social influence prediction model based on content awareness, comprising:
an input layer for inputting information, the input information comprising: user network structure information, user network attribute information and user social media content information;
a GCN layer for extracting the user network structure information and user network attribute information inputted from the input layer;
the Embedding layer is used for storing the D-dimensional Embedding representation of the user extracted from the GCN layer;
the LSTM layer is used for outputting the user social media content input by the input layer into an Embedding representation of the social media content by using an LSTM algorithm;
the RN layer is used for putting the Embedding representation output by the Embedding layer and the social media content Embedding representation output by the LSTM layer into the RN model, and deducing whether the current node is influenced by the neighbor node under the given social media content by utilizing the RN model;
the mathematical expression of the RN model is as follows:
where o refers to the user's Embedding representation, c i Refers to a time series representation of the ith social media content,
refers to whether a neighbor of the central user has made an overaction on social media content i before time t, v refers to the central user, u refers to the neighbor of v, Σ refers to the sum, g θ A three-layer perceptron network is shown,
representing a two-tiered perceptron network;
and the output layer is used for representing the result of model prediction in a two-dimensional form under the influence of the neighbor nodes of the social media content.
2. The content-awareness-based social influence prediction model of claim 1, further comprising: and the Grountreth layer is used for comparing the result predicted by the model with the Grountreth layer so as to optimize the objective function.
3. The content-awareness-based social influence prediction model of claim 1, wherein in the input layer, each instance of information in the input layer is an r-agent network consisting of n users;
the r-proxy network definition is that N nodes are sampled from an r-neighbor network of a user v as a representative;
the r-neighbor is defined as
Where d (u, v) refers to the shortest distance between user u and user v in terms of hop count.
4. The content-aware-based social influence prediction model of claim 1, wherein the GCN layer receives as input m user network structure information and m n x f user network attribute information.
5. The content-awareness-based social influence prediction model according to claim 1, wherein the user network structure information is calculated from an adjacency matrix a of the network in the GCN layer, and the GCN layer instantiates the adjacency matrix a of the network and the user network attribute information matrix H as static matrices closely related to a normalized graph laplacian:
wherein,
a is the adjacency matrix of the network graph G, I is the identity matrix,
is a diagonal matrix of network node values, after obtaining A (G), the matrix is summedThe user network attribute information matrix H together serves as an input to each GCN layer for extracting network node characteristics.
6. The content-aware-based social influence prediction model of claim 1, wherein in the LSTM layer, the LSTM layer receives m p x q of user social media content information as input, p referring to the number of words in the social media content information, and q referring to an initial dimension of each word.
7. A content-awareness-based social influence prediction model prediction method according to claim 3, specifically comprising the following steps:
firstly, sampling to obtain a sub-network with a fixed size, and using the sub-network as a proxy network of each r-ego network, wherein the proxy network comprises user network structure information, user network attribute information and user social media content information;
the r-ego network is defined as being composed of
A network of nodes in (1);
and step two, the agent network in the step one is put into a social influence prediction model based on content perception for prediction, and a corresponding prediction result is obtained.
8. The prediction method of claim 7, further comprising:
and step three, comparing the output result in the step two with a real result, calculating loss and minimizing a negative log-likelihood function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910452492.0A CN110147911B (en) | 2019-05-28 | 2019-05-28 | Social influence prediction model and prediction method based on content perception |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910452492.0A CN110147911B (en) | 2019-05-28 | 2019-05-28 | Social influence prediction model and prediction method based on content perception |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110147911A CN110147911A (en) | 2019-08-20 |
| CN110147911B true CN110147911B (en) | 2022-09-09 |
Family
ID=67593571
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910452492.0A Active CN110147911B (en) | 2019-05-28 | 2019-05-28 | Social influence prediction model and prediction method based on content perception |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110147911B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110969187B (en) * | 2019-10-30 | 2020-12-11 | 创意信息技术股份有限公司 | Semantic analysis method for map migration |
| CN111310975A (en) * | 2020-01-19 | 2020-06-19 | 浙江工业大学 | Multi-task message propagation prediction method based on depth model |
| CN111339449A (en) * | 2020-03-24 | 2020-06-26 | 青岛大学 | User motion trajectory prediction method, device, equipment and storage medium |
| CN112199608B (en) * | 2020-11-03 | 2021-05-28 | 北京中科研究院 | Social media rumor detection method based on network information propagation graph modeling |
| CN112508085B (en) * | 2020-12-05 | 2023-04-07 | 西安电子科技大学 | Social network link prediction method based on perceptual neural network |
| CN112667920A (en) * | 2020-12-29 | 2021-04-16 | 复旦大学 | Text perception-based social influence prediction method, device and equipment |
| CN116029601B (en) * | 2023-01-17 | 2024-06-18 | 浙江警察学院 | RF-RFECV-based traffic safety video propaganda effect evaluation method |
| CN117764233A (en) * | 2023-12-25 | 2024-03-26 | 北京智谱华章科技有限公司 | Social behavior prediction method, device, equipment and medium based on content perception |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3073433A1 (en) * | 2015-03-24 | 2016-09-28 | Xerox Corporation | Language identification on social media |
| CN107895215A (en) * | 2017-12-21 | 2018-04-10 | 北京理工大学 | The prediction of community network influence power and maximization System and method for based on neutral net |
| CN108038240A (en) * | 2017-12-26 | 2018-05-15 | 武汉大学 | Based on content, the social networks rumour detection method of user's multiplicity |
| CN109767301A (en) * | 2019-01-14 | 2019-05-17 | 北京大学 | Recommended method and system, computer installation, computer readable storage medium |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9412092B2 (en) * | 2013-01-14 | 2016-08-09 | Google Inc. | Generating a filtered view of a content stream |
-
2019
- 2019-05-28 CN CN201910452492.0A patent/CN110147911B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3073433A1 (en) * | 2015-03-24 | 2016-09-28 | Xerox Corporation | Language identification on social media |
| CN107895215A (en) * | 2017-12-21 | 2018-04-10 | 北京理工大学 | The prediction of community network influence power and maximization System and method for based on neutral net |
| CN108038240A (en) * | 2017-12-26 | 2018-05-15 | 武汉大学 | Based on content, the social networks rumour detection method of user's multiplicity |
| CN109767301A (en) * | 2019-01-14 | 2019-05-17 | 北京大学 | Recommended method and system, computer installation, computer readable storage medium |
Non-Patent Citations (1)
| Title |
|---|
| DeepInf:Social Influence Prediction with Deep Learning;Jiezhong Qiu,etal;《 Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining》;20180719;2110-2119 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110147911A (en) | 2019-08-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110147911B (en) | Social influence prediction model and prediction method based on content perception | |
| Zhou et al. | Multiobjective biogeography based optimization algorithm with decomposition for community detection in dynamic networks | |
| Tran et al. | Selection of significant input variables for time series forecasting | |
| CN112711953A (en) | Text multi-label classification method and system based on attention mechanism and GCN | |
| CN110321493B (en) | Abnormity detection and optimization method and system of social network and computer equipment | |
| CN105956798A (en) | Sparse random forest-based method for assessing running state of distribution network device | |
| Selvarajah et al. | Dynamic network link prediction by learning effective subgraphs using CNN-LSTM | |
| CN117036060A (en) | Vehicle insurance fraud recognition method, device and storage medium | |
| Abinaya et al. | Spam detection on social media platforms | |
| Agana et al. | Short-term load forecasting based on a hybrid deep learning model | |
| CN116545679A (en) | Industrial situation security basic framework and network attack behavior feature analysis method | |
| Ho et al. | Discovering communities of users on social networks based on topic model combined with Kohonen network | |
| CN115329834A (en) | Cable state determination method, device and system | |
| Handayani et al. | Sentiment Analysis Of Electric Cars Using Recurrent Neural Network Method In Indonesian Tweets | |
| CN114239750A (en) | Alarm data processing method, device, storage medium and equipment | |
| Nanduri et al. | Modified Fuzzy Approach to Automatic Classification of Cyber Hate Speech from the Online Social Networks (OSN's). | |
| CN109308565B (en) | Crowd performance grade identification method and device, storage medium and computer equipment | |
| Felix et al. | Fuzzy cognitive DEMATEL technique for modeling cause and effect of climate change | |
| Ramya et al. | Integrating Quantum Computing and NLP for Advanced Cyber Threat Detection. | |
| Mirhosseini et al. | Metaheuristic search algorithms in solving the n-similarity problem | |
| Shen et al. | Recursive filtration method for detecting community structure in networks | |
| Gata et al. | The Feasibility of Credit Using C4. 5 Algorithm Based on Particle Swarm Optimization Prediction | |
| CN115062300B (en) | Intrusion detection method and system based on multi-channel deep convolution | |
| Abirami et al. | Water Quality Analysis and Prediction using Machine Learning | |
| Naik et al. | An improved harmony search-based functional link higher order ANN for nonlinear data classification |
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 |