CN109300057B - Network group forming mechanism discovery method based on social user hidden feature representation - Google Patents
Network group forming mechanism discovery method based on social user hidden feature representation Download PDFInfo
- Publication number
- CN109300057B CN109300057B CN201811011747.1A CN201811011747A CN109300057B CN 109300057 B CN109300057 B CN 109300057B CN 201811011747 A CN201811011747 A CN 201811011747A CN 109300057 B CN109300057 B CN 109300057B
- Authority
- CN
- China
- Prior art keywords
- user
- network
- social
- connection
- formula
- 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
Images
Classifications
-
- 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—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/01—Social networking
Abstract
The invention provides a network group forming mechanism discovery method based on social user hidden feature representation, and relates to the technical field of networks. The method comprises the following steps: constructing a social network based on the user connection relation; learning by utilizing a node2vec method to obtain a hidden feature representation of the social network; generating networks under different influence mechanisms based on the hidden feature representation of the social network; and analyzing the network characteristics under the different influence mechanisms to determine the formation mechanism of the real network. The invention combines the selection mechanism and the influence mechanism, dynamically generates the network based on the hidden preference of the user, more vividly depicts the change of the real social network, and thus more accurately determines the formation mechanism of the real network.
Description
Technical Field
The invention relates to the technical field of social networks, in particular to a network group forming mechanism discovery method based on social user hidden feature representation.
Background
With the development of web2.0 applications and other various types of social media, Online Social Networks (OSNs) have become the most important platforms for people's network life, and on these platforms, users do not exist independently, and they may be influenced by social choices or social influences and other mechanisms to form social connections or join in groups, the social choices represent connection relationships such as attention generated by users due to interests, and the social influences represent connection relationships such as attention generated by users influenced by important influence in the network.
Most of the existing researches on network generation mechanisms are that users are more prone to obtain information by connecting high-number nodes, and a BA network generation model is formed under the influence mechanism, but other important characteristics in a social network, such as clustering and community structure, are not considered in the model. Some studies show that not only the high degree nodes are connected by users, but also the high degree nodes are connected by the users because of personal interests and a few low degree nodes, so that the social selection of relationship formation because of similar preference is also an important mechanism for the network evolution formation. Such social choices can be understood as both explicit homogeneity, which represents the similarity of user preferences in explicit properties, such as age, location, etc., and implicit homogeneity, which represents the similarity of user preferences in some implicit preferences. Under the selection mechanism, such as a spatial stochastic graph model, general characteristics of the social network, such as "small world", power law distribution and high clustering performance, can be described, however, when the social network generation and evolution mechanism is researched, the model needs to define the network size. Recent research considering selection and influence mechanisms, such as a kinship model (kinship model) using the same color to represent the intimacy degree between nodes, and the addition of a new node can judge whether the new node selects an existing color or a new color according to the degree; for example, the popularity similarity model (PS) is regarded as the node generation time, the similarity is used to measure the angular distance between two nodes, and the new node is added by selecting the nearest m nodes to connect based on the hyperbolic distance on the polar coordinates. However, these models only express some characteristics of the real network, and there is no further study on how the selection and influence mechanism influences the node connection, the edge change, and the like in the network formation process.
Disclosure of Invention
Technical problem to be solved
Aiming at the defects of the prior art, the invention provides a network group forming mechanism discovery method based on the hidden feature representation of social users, which can solve the technical problem that a selection and influence mechanism influences in the network forming process.
(II) technical scheme
In order to achieve the purpose, the invention is realized by the following technical scheme:
a network group formation mechanism discovery method based on social user hidden feature representation comprises the following steps,
constructing a social network based on the user connection relation;
learning by utilizing a node2vec method to obtain a hidden feature representation of the social network;
generating networks under different influence mechanisms based on the hidden feature representation of the social network;
and analyzing the network characteristics under the different influence mechanisms to determine the formation mechanism of the real network.
Further, learning by using a node2vec method to obtain a hidden feature representation of the social network; the method comprises the following steps:
constructing a neighbor user set through a random walk process;
constructing an objective function aiming at a social network;
obtaining a likelihood function of a neighbor user set by the target function;
unitizing the characteristic dot product represented by the likelihood function;
and continuously optimizing by a random gradient rising method, and learning to obtain a better user hidden feature matrix.
Further, the generating of the network under different influence mechanisms based on the hidden feature representation of the social network; the method comprises the following steps:
at an initial time t0Random selection of m0Each initial user constructs a full-connection network;
calculating the number of new connections generated at time t;
calculating the connection probability between the user and the user at the time t;
and dynamically generating a simulation social network aiming at the full-connection network and the connection probability matrix.
Further, the social network is constructed based on the user connection relation; the method comprises the following steps:
defining a user set U in the user connection relationship as: u ═ U1,...,ui,...,uj,...unN represents the number of users; with EijRepresenting user uiAnd user ujThe connection relation between the users is 1 < i < n, 1 < j < n, if the user uiAnd user ujIf a connection exists, then Eij1 denotes two usersForm an edge in between, otherwise Eij0; e represents a connection relation set; at omegaijRepresenting user uiAnd user ujConnection relation between EijW represents a set of weights;
definition Eii＝1；
And connecting the users with connection relations in sequence to construct a social network R ═ U, E and W.
Further, the learning by using the node2vec method to obtain the latent feature representation of the social network includes:
in the constructed social network R, an initial user u is given0Simulating random walk of fixed length l, user uvWandering to user uxIs characterized by the formula (1):
in the formula (1), uvAnd uxRespectively representing a user v and a user x in the wandering process; z is a normalization constant; pivxIs u characterized by formula (2)vTo uxNon-standard transition probability of (2):
πvx＝αpq(t,x)·ωvxwherein, in the step (A),
in the formula (2), dtxThe shortest distance between the user t and the user x is represented, and the shortest distance in random walk is set to be not more than 3; alpha is alphapq(t, x) represents the probability that user t has walked to user x, p and q are control parameters; omegavxRepresenting the relationship weight of the user v and the user x;
constructing a neighbor user set through a random walk process, and defining
A set of neighbor users representing users;
for social network R ═ (U, E, W), an objective function is constructed as characterized by equation (3):
in the formula (3), f is a user feature matching function; f (u) ═ f (u)1),...,f(ui),...f(un)]Representing a user implicit feature matrix, wherein the user implicit feature matrix f (u) is an n x d matrix; f (u)i)＝(ai1,...,aik,…aid) Representing user uiThe user uiHidden feature vector f (u)i) Is a 1 × d vector, aikRepresenting user uiThe k-th dimension feature preference value of (1); d represents the number of hidden features of the user; likelihood function P (N) of neighbor user setS(u) | f (u)) is characterized by formula (4):
in the formula (4), user uiIs a characteristic dot product unitization characterized by equation (5):
combining equations (4) and (5), the objective function characterized by equation (3) is reduced as characterized by equation (6):
aiming at the target function represented by the formula (6), continuously optimizing by a random gradient ascending method, and learning to obtain a better user hidden feature matrix f (u).
Further, the generating a network under different influence mechanisms based on the hidden feature representation of the social network includes:
at an initial time t0Random selection of m0Each initial user constructs a full-connection network;
the number of new connections generated when time t is calculated is characterized by equation (7):
Δmt＝Nt k-(Nt-1)k (7)
in the formula (7), NtThe value range of the change rate k is set to [1.1,1.7 ] and represents the number of users at time t]；
User u at calculated time tiWith user ujConnection probability of
As characterized by formula (8):
in the formula (8), betai(0≤βi≦ 1) representing user uiThe interest selection weights are sampled from the probability density function;
is the time t user u characterized by equation (9)jDegree of influence of (c):
in the formula (9), the reaction mixture is,
representing user ujDegrees at time t;
in formula (8), sim (u)i,uj) Representing user u at time tiWith user ujIs a latent feature vector f (u) as characterized by equation (10)i)＝(ai1,...,aik,…aid) And f (u)j)＝(aj1,...,ajk,…ajd) Cosine similarity of (c):
the connection probability among all users forms a connection probability matrix P when defining time tt；
And dynamically generating a simulation social network aiming at the full-connection network and the connection probability matrix.
Further, the simulation social network comprises an influence type generation network, an interest type generation network and a neutralization type generation network.
Further, the analyzing the network characteristics under the different influence mechanisms to determine the formation mechanism of the real network includes:
aiming at the influence type generation network, the interest type generation network, the neutral type generation network and the real network, the distribution condition of the four networks is analyzed on the degree distribution, the clustering coefficient, the KNN and the community size, the generation mechanism which is more accordant with the real network distribution is judged, and the fact that the real network is more biased to which generation mechanism can be determined.
(III) advantageous effects
The invention discloses a network group forming mechanism discovery method based on-line social user hidden feature representation, which uses a two-dimensional matrix to represent a social network relation based on a user; learning a hidden feature representation of the user's social relationship based on node2 vec; generating networks under three different influence mechanisms of a selection type network, an influence type network and a neutral type network based on the hidden feature representation of the user social relationship; and analyzing the network characteristics to determine the forming mechanism of the real network. The invention combines the selection mechanism and the influence mechanism, dynamically generates the network based on the hidden preference of the user, more vividly depicts the change of the real social network, and thus more accurately determines the formation mechanism of the real network.
Drawings
FIG. 1 is a schematic flow diagram of the present invention;
FIG. 2a is a visual diagram of Eu-email and Youtube clustering coefficients of two real social network data sets and their comparisons in a selection type, an influence type and a neutral type in the present invention;
FIG. 2b is a visualization diagram of Eu-email and Youtube real social network data sets and their degree distribution in selection type, influence type and neutral type;
FIG. 2c is a visualization diagram of the Eu-email and Youtube real social network data sets and their KNN for comparison in the selection type, the influence type and the neutral type.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention.
As shown in fig. 1, a method for discovering a network group formation mechanism based on a social user implicit feature representation according to an embodiment of the present invention includes the following steps:
constructing a social network based on the user connection relation;
learning by utilizing a node2vec method to obtain a hidden feature representation of the social network;
generating networks under different influence mechanisms based on the hidden feature representation of the social network;
and analyzing the network characteristics under the different influence mechanisms to determine the formation mechanism of the real network.
Each step is described in detail below:
step 1, establishing a social network R ═ (U, E, W) aiming at the connection relation of users
Defining: in the user connection relationship, the user set U is as follows: u ═ U1,...,ui,...,uj,...unN represents the number of users; with EijRepresenting user uiAnd user ujThe connection relation between the users is 1 < i < n, 1 < j < n, if the user uiAnd user ujIf a connection exists, then Eij1, indicates that an edge is formed between two users, otherwise Eij0; e represents a connection relation set; at omegaijRepresenting user uiAnd user ujConnection relation between EijWeight of, W tableShowing a set of weights;
defining: eii＝1；
And connecting the users with connection relations in sequence to construct a social network R ═ U, E and W.
Step 2, learning the hidden characteristics of the social relationship of the user by using the node2vec method
The node2vec method performs feature extraction by using the relation between adjacent nodes, adopts a random walk strategy, and combines a Breadth-first Sampling (BFS) search strategy and a Depth-first Sampling (DFS) search strategy to select the adjacent nodes.
Step 2.1, in the social network R constructed in step 1, an initial user u is given0Simulating random walk of fixed length l, user uvWandering to user uxIs characterized by the formula (1):
in the formula (1), uvAnd uxRespectively representing a user v and a user x in the wandering process; z is a normalization constant; pivxIs u characterized by formula (2)vTo uxNon-standard transition probability of (2):
πvx＝αpq(t,x)·ωvxwherein, in the step (A),
in the formula (2), dtxThe shortest distance between the user t and the user x is represented, and the shortest distance in random walk is set to be not more than 3; alpha is alphapq(t, x) represents the probability that the user t walks to the user x, p and q are control parameters, the parameters p and q are used for controlling the walking speed, and p controls the probability of revisiting a certain node, and generally a larger value is set to prevent the revisited node from being sampled; q allows two different searches, inward and outward, if q > 1, random walk is more likely to sample the immediate neighbors, similar to the BFS strategy, belonging to inward search, otherwise more likely to sample awayRemote nodes, similar to the DFS policy, belong to outward search; omegavxRepresenting the weight of the relationship of user v to user x.
Constructing a neighbor user set through a random walk process, and defining
Representing a set of neighbor users of a user.
Step 2.2, constructing an objective function characterized by the formula (3) for the social network R ═ U, E, W:
in the formula (3), f is a user feature matching function; f (u) ═ f (u)1),...,f(ui),...f(un)]Representing a user implicit feature matrix, wherein the user implicit feature matrix f (u) is an n x d matrix; f (u)i)＝(ai1,...,aik,…aid) Representing user uiThe user uiHidden feature vector f (u)i) Is a 1 × d vector, aikRepresenting user uiThe k-th dimension feature preference value of (1); d represents the number of hidden features of the user; likelihood function P (N) of neighbor user set based on condition independence assumption of mutual independence between neighbor nodesS(u) | f (u)) is characterized by formula (4):
P(NS(u)|f(u))＝∑P(ui|f(u)) (4)
in equation (4), user u is based on the assumption that the target node and the neighboring node influence each other symmetrically in the feature spaceiIs a characteristic dot product unitization characterized by equation (5):
combining equations (4) and (5), the objective function characterized by equation (3) can be reduced to that characterized by equation (6):
aiming at the target function represented by the formula (6), a better user hidden feature matrix f (u) can be obtained through learning by continuously optimizing by a random gradient ascending method.
Step 3, generating a network under different influence mechanisms by utilizing the user hidden feature matrix f (u), and step 3.1, at the initial time t0Random selection of m0Each initial user constructs a full-connection network;
step 3.2, calculating the new connection number generated in time t, as represented by formula (7):
Δmt＝Nt k-(Nt-1)k···(7)
in the formula (7), NtThe value range of the change rate k is set to [1.1,1.7 ] and represents the number of users at time t]。
Step 3.3, calculating time t user uiWith user ujConnection probability of
As characterized by formula (8):
in equation (8), the connection probability is a linear function of the selection factor and the influence factor; beta is ai(0≤βi≦ 1) representing user uiIs the user uiWith user ujConsidering the trade-off factors of influence degree and selectivity when establishing connection, the interest selection weight is sampled from probability density function, considering that the user may select connection because of interest, or establish connection because of influence of connected node user, or the combined influence of two factors, then it is assumed here that preference distribution obeyed by β is three probability density functions respectively: [0,1]Interval monotonically increasing, [0,1]The intervals are monotonically decreased and uniformly distributed; when the probability density function is monotonically increasing, the average ofThe value is more than 0.5, which indicates that the selection factor has stronger influence than the influence factor, and the generation of the network is indicated by 'high'; when the probability density function is monotonically decreased, the mean value of beta is less than 0.5, which means that the influence factor is stronger than the influence of the selection factor, and the generation of the network is represented by low;
is the time t user u characterized by equation (9)jDegree of influence of (c):
in the formula (9), the reaction mixture is,
representing user ujDegrees at time t;
in formula (8), sim (u)i,uj) Representing user u at time tiWith user ujIs a latent feature vector f (u) as characterized by equation (10)i)＝(ai1,...,aik,…aid) And f (u)j)＝(aj1,...,ajk,…ajd) Cosine similarity of (c):
defining: the connection probability among all users at time t forms a connection probability matrix Pt。
Step 3.4, aiming at the full-connection network and the connection probability matrix, dynamically generating a simulation social network
In consideration of social network dynamics in a real network environment, the phenomena of connection establishment and connection cancellation exist among existing node users;
setting a current in-connection threshold value to be P for an existing full-connection network at time ttMedian T of all probability values1 tOnly if the probability of connection is greater than the thresholdWhen the value is positive, the connection is formed, otherwise, the connection is disconnected; the number of new connections generated from equation (7) should be Δ mt+mt′，mt' represents the number of connections broken at time t;
when the time is t, other new users in the U are sequentially added into the full-connection network at the time t-1, and a new user U is settExisting user u in a fully connected network with said time t-1iIs characterized by the normalized connection probability of equation (11):
according to the connection probability matrix P in the step 3.3tSelecting from unconnected edges
Connection with high connection probability
The connection probability of the connection is sequentially
Connection EiWill be given a probability
Is selected. Wherein the content of the first and second substances,
the above process is repeated until all users are joined to the network.
And 3.5, sampling the interest selection weights from different probability density functions in the step 3.3, and repeating the steps 3.1-3.4, so that three types of simulation social networks of an influence type, an interest type and a neutralization type can be obtained.
Step 4, comparing the three types of simulated social networks with the social network R so as to determine the formation mechanism of the social network R
Aiming at the influence type generation network, the interest type generation network, the neutral type generation network and the real network, the distribution condition of the four networks is analyzed on the degree distribution, the clustering coefficient, the KNN and the community size, the generation mechanism which is more accordant with the real network distribution is judged, and the fact that the real network is more biased to which generation mechanism can be determined.
And 5, performing an experiment by using the standard data set, comparing and analyzing the social network performance indexes such as the clustering coefficient, the KNN and the degree distribution, and comparing the social network performance indexes with the real network so as to determine the forming mechanism of the real network.
The experimental demonstration aiming at the method comprises the following steps:
1) preparing a standard data set
The effectiveness of the method is verified by using two real social network data sets of Eu-email and Youtube as a standard data set, wherein the data set is a large-scale social network data set collected and arranged by Stanford university. The Eu-email data set is real email network data of a research institution in Europe, comprising 25571 side data of 1005 independent users from 42 departments, wherein members of the departments can send mails to each other, and the data set does not record data sent or received from the outside of the institution. The Youtube data set is social data on a video sharing website, 66164 edge data of 12382 independent users from 100 groups are sampled, and the users can establish friend relationships with others and can create or join the groups.
2) Evaluation index
In a comparison test between a real social network and three types of generation mechanism networks, the following indexes are adopted: degree distribution, measuring the distribution of the number of connecting edges of the nodes; average clustering coefficient, which represents the overall signs of node clustering or clustering; KNN, which represents the neighbor average degree distribution of the node with k degree; cosine similarity, which represents the degree of coherence of the two networks.
3) Experiments were performed on standard data sets
To verify the effectiveness of the proposed method, modeling and prediction were performed on the data sets of Eu-email and Youtube, respectively. Firstly, obtaining the hidden features of a user through Node2vec by social connection relation data, wherein the dimensions of the hidden features are all 30; secondly, respectively simulating three types of networks of a selection type (SI, high), an influence type (SI, low) and a neutral type (SI, uniform); finally, the simulation network and the real social network are compared on the evaluation index, the experimental result is shown in table 1, the method of the invention finds that the Eu-email real social network belongs to a neutral type, namely, the Eu-email real social network represents that the connection established by the user in the network is influenced by the common action of an interest type and an influence type, and the YouTube real social network belongs to an influence type, namely, the Eu-email real social network represents that the connection established by the user in the network is influenced by more influenced users and is consistent with the real research result.
TABLE 1
In conclusion, compared with the prior art, the beneficial effects of the embodiment of the invention are as follows:
1. the invention provides a group formation mechanism discovery method based on-line social user implicit feature representation, which is characterized in that a selection mechanism and an influence mechanism are combined, wherein the selection mechanism is constructed based on user implicit preference, and compared with sampling subject to distribution, the method provided by the invention is more consistent with the real social situation of a user.
2. The invention provides that the user connection probability is a balance between a selection mechanism and an influence mechanism, and in the dynamic generation process of the network, the generation and cancellation of the connection obey the probability under a certain generation mechanism, thereby more vividly depicting the change of the real social network.
3. In the real social environment, influence factors influencing the generation of the connection relation of the user may be different, and the invention can find that the mechanism formed by the real network is an influence type, a selection type or a neutralization type according to different user hidden characteristics aiming at different types of data and social environments.
4. The method can be used for scenes such as display feedback of user purchase and the like, implicit feedback of user browsing and the like, social network relationship and the like, and the discovery of the network generation mechanism can be used for other research fields such as recommendation, social network and the like, so that the application range is wide.
It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (7)
1. A network group forming mechanism discovery method based on social user hidden feature representation is characterized by comprising the following steps,
constructing a social network based on the user connection relation;
learning by utilizing a node2vec method to obtain a hidden feature representation of the social network;
generating networks under different influence mechanisms based on the hidden feature representation of the social network;
analyzing the network characteristics under the different influence mechanisms, and determining the formation mechanism of the real network;
the learning and obtaining of the hidden feature representation of the social network by using the node2vec method comprises the following steps:
in the constructed social network R, an initial user u is given0Simulating random walk of fixed length l, user uvWandering to user uxIs characterized by the formula (1):
in the formula (1), uvAnd uxRespectively representing a user v and a user x in the wandering process; z is a normalization constant; pivxIs u characterized by formula (2)vTo uxNon-standard transition probability of (2):
πvx＝αpq(t,x)·ωvxwherein, in the step (A),
in the formula (2), dtxThe shortest distance between the user t and the user x is represented, and the shortest distance in random walk is set to be not more than 3; alpha is alphapq(t, x) represents the probability that user t has walked to user x, p and q are control parameters; omegavxRepresenting the relationship weight of the user v and the user x;
constructing a neighbor user set through a random walk process, and defining
A set of neighbor users representing users;
for social network R ═ (U, E, W), an objective function is constructed as characterized by equation (3):
in the formula (3), f is a user feature matching function; f (u) ═ f (u)1),...,f(ui),...f(un)]Representing a user implicit feature matrix, wherein the user implicit feature matrix f (u) is an n x d matrix; f (u)i)＝(ai1,...,aik,…aid) Representing user uiThe user uiHidden feature vector f (u)i) Is a 1 × d vector, aikRepresenting user uiThe k-th dimension feature preference value of (1); d represents the number of hidden features of the user; likelihood function P (N) of neighbor user setS(u) | f (u)) is characterized by formula (4):
in the formula (4), user uiIs a characteristic dot product unitization characterized by equation (5):
combining equations (4) and (5), the objective function characterized by equation (3) is reduced as characterized by equation (6):
aiming at the target function represented by the formula (6), continuously optimizing by a random gradient ascending method, and learning to obtain a better user hidden feature matrix f (u).
2. The method for discovering network group formation mechanism based on social user implicit feature representation according to claim 1, wherein the implicit feature representation of the social network is obtained by learning through a node2vec method; the method comprises the following steps:
constructing a neighbor user set through a random walk process;
constructing an objective function aiming at a social network;
obtaining a likelihood function of a neighbor user set by the target function;
unitizing the characteristic dot product represented by the likelihood function;
and continuously optimizing by a random gradient rising method, and learning to obtain a better user hidden feature matrix.
3. The method for discovering mechanism of forming network group based on implicit feature representation of social users according to claim 1, wherein the implicit feature representation based on social network generates network under different influence mechanism; the method comprises the following steps:
at an initial time t0Random selection of m0Each initial user constructs a full-connection network;
calculating the number of new connections generated at time t;
calculating the connection probability between the user and the user at the time t;
and dynamically generating a simulation social network aiming at the full-connection network and the connection probability matrix.
4. The method for discovering network group formation mechanism based on social user implicit feature representation according to claim 1, wherein the social network is constructed based on user connection relation; the method comprises the following steps:
defining a user set U in the user connection relationship as: u ═ U1,...,ui,...,uj,...unN represents the number of users; with EijRepresenting user uiAnd user ujThe connection relation between the users is 1 < i < n, 1 < j < n, if the user uiAnd user ujIf a connection exists, then Eij1, indicates that an edge is formed between two users, otherwise Eij0; e represents a connection relation set; at omegaijRepresenting user uiAnd user ujConnection relation between EijW represents a set of weights;
definition Eii＝1；
And connecting the users with connection relations in sequence to construct a social network R ═ U, E and W.
5. The method for discovering mechanism of forming network group based on implicit feature representation of social users according to claim 1, wherein the generating of the network under different influence mechanisms based on implicit feature representation of social network comprises:
at an initial time t0Random selection of m0Each initial user constructs a full-connection network;
the number of new connections generated when time t is calculated is characterized by equation (7):
Δmt＝Nt k-(Nt-1)k (7)
in the formula (7), NtThe value range of the change rate k is set to [1.1,1.7 ] and represents the number of users at time t]；
User u at calculated time tiWith user ujConnection probability of
As characterized by formula (8):
in the formula (8), betai(0≤βi≦ 1) representing user uiThe interest selection weights are sampled from the probability density function;
is the time t user u characterized by equation (9)jDegree of influence of (c):
in the formula (9), the reaction mixture is,
representing user ujDegrees at time t;
in formula (8), sim (u)i,uj) Representing user u at time tiWith user ujIs a latent feature vector f (u) as characterized by equation (10)i)＝(ai1,...,aik,…aid) And f (u)j)＝(aj1,...,ajk,…ajd) Cosine similarity of (c):
the connection probability among all users forms a connection probability matrix P when defining time tt；
And dynamically generating a simulation social network aiming at the full-connection network and the connection probability matrix.
6. The method for discovering network group formation mechanism based on social user implicit feature representation according to claim 5, wherein the simulated social network is an influence type generation network, an interest type generation network and a neutralization type generation network.
7. The method for discovering mechanism of forming network group based on hidden feature representation of social users according to claim 6, wherein the analyzing the network characteristics under the different influence mechanisms to determine the mechanism of forming real network comprises:
aiming at the influence type generation network, the interest type generation network, the neutral type generation network and the real network, the distribution condition of the four networks is analyzed on the degree distribution, the clustering coefficient, the KNN and the community size, the generation mechanism which is more accordant with the real network distribution is judged, and the fact that the real network is more biased to which generation mechanism can be determined.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811011747.1A CN109300057B (en) | 2018-08-31 | 2018-08-31 | Network group forming mechanism discovery method based on social user hidden feature representation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811011747.1A CN109300057B (en) | 2018-08-31 | 2018-08-31 | Network group forming mechanism discovery method based on social user hidden feature representation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109300057A CN109300057A (en) | 2019-02-01 |
| CN109300057B true CN109300057B (en) | 2021-08-10 |
Family
ID=65165807
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811011747.1A Active CN109300057B (en) | 2018-08-31 | 2018-08-31 | Network group forming mechanism discovery method based on social user hidden feature representation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109300057B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111881366A (en) * | 2020-07-31 | 2020-11-03 | 深圳市万物云科技有限公司 | User community discovery method based on graph machine learning and related components |
| CN112581299A (en) * | 2020-12-30 | 2021-03-30 | 中科院计算技术研究所大数据研究院 | Detection method of network group in online social network based on topic and information interaction behavior |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103595623A (en) * | 2013-11-28 | 2014-02-19 | 中国科学技术大学 | Opportunistic routing behavioral modeling method based on mobile social network node social characteristics |
| CN105868447A (en) * | 2016-03-24 | 2016-08-17 | 南京邮电大学 | User communication behavior analysis and model simulation system based on double-layer network |
| CN106682991A (en) * | 2016-12-21 | 2017-05-17 | 重庆邮电大学 | Information propagation model based on online social network and propagation method thereof |
| CN107330115A (en) * | 2017-07-12 | 2017-11-07 | 广东工业大学 | A kind of information recommendation method and device |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CA2924889A1 (en) * | 2015-03-24 | 2016-09-24 | Mxm Nation Inc. | System an method for facilitating credit redemptions |
-
2018
- 2018-08-31 CN CN201811011747.1A patent/CN109300057B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103595623A (en) * | 2013-11-28 | 2014-02-19 | 中国科学技术大学 | Opportunistic routing behavioral modeling method based on mobile social network node social characteristics |
| CN105868447A (en) * | 2016-03-24 | 2016-08-17 | 南京邮电大学 | User communication behavior analysis and model simulation system based on double-layer network |
| CN106682991A (en) * | 2016-12-21 | 2017-05-17 | 重庆邮电大学 | Information propagation model based on online social network and propagation method thereof |
| CN107330115A (en) * | 2017-07-12 | 2017-11-07 | 广东工业大学 | A kind of information recommendation method and device |
Non-Patent Citations (1)
| Title |
|---|
| 同质性和社会影响对混合型社交网络形成的仿真分析;何军等;《现代情报》;20170430;第37卷(第4期);第87-94页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109300057A (en) | 2019-02-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111428147B (en) | Social recommendation method of heterogeneous graph volume network combining social and interest information | |
| CN108596774B (en) | Social information recommendation algorithm and system based on deep network embedded features | |
| CN109190030B (en) | Implicit feedback recommendation method fusing node2vec and deep neural network | |
| CN113705772A (en) | Model training method, device and equipment and readable storage medium | |
| CN107657043B (en) | Content-based mixed graph model image recommendation method | |
| CN106708953A (en) | Discrete particle swarm optimization based local community detection collaborative filtering recommendation method | |
| CN109325875B (en) | Implicit group discovery method based on hidden features of online social users | |
| CN110209946B (en) | Social and community-based product recommendation method, system and storage medium | |
| KR101356948B1 (en) | User interest inference method and system in sns using topics on social activities with neighbors | |
| CN108470075A (en) | A kind of socialization recommendation method of sequencing-oriented prediction | |
| CN107145541B (en) | Social network recommendation model construction method based on hypergraph structure | |
| CN109300057B (en) | Network group forming mechanism discovery method based on social user hidden feature representation | |
| Agarwal et al. | Trust-enhanced recommendation of friends in web based social networks using genetic algorithms to learn user preferences | |
| CN109960755B (en) | User privacy protection method based on dynamic iteration fast gradient | |
| CN111723285A (en) | Depth spectrum convolution collaborative filtering recommendation method based on scores | |
| CN115270007B (en) | POI recommendation method and system based on mixed graph neural network | |
| CN113255895A (en) | Graph neural network representation learning-based structure graph alignment method and multi-graph joint data mining method | |
| Su et al. | Multi-view heterogeneous molecular network representation learning for protein–protein interaction prediction | |
| CN113239266B (en) | Personalized recommendation method and system based on local matrix decomposition | |
| CN109857928A (en) | User preference prediction technique based on polynary credit evaluation | |
| CN112364245B (en) | Top-K movie recommendation method based on heterogeneous information network embedding | |
| CN111008334B (en) | Top-K recommendation method and system based on local pairwise ordering and global decision fusion | |
| CN111143704B (en) | Online community friend recommendation method and system integrating user influence relationship | |
| CN110825978B (en) | Multitask collaborative filtering method based on neighbor user feature sharing | |
| Cheng et al. | Collaborative filtering algorithm based on data mixing and filtering |
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 |