CN106372239A - Social network event correlation analysis method based on heterogeneous network - Google Patents

Abstract

The invention discloses a social network event correlation analysis method based on a heterogeneous network. The method specifically comprises the following steps: constructing a figure-event heterogeneous network in a social network space; extracting information characteristics of nodes and edges in the figure-event heterogeneous network; performing community detection under the edge-connection constraint of the figure-event heterogeneous network, so as to obtain a figure community and an event community; performing heterogeneous community detection on the figure-event heterogeneous network by virtue of figure-event correlation constraints, and researching the community detection results, so as to achieve the event correlation analysis aim. According to the method disclosed by the invention, the heterogeneous network is utilized, structures of three graphs such as users, events and user events, in the social network are increased, the events in the social network are further analyzed, and the aim of mining event correlation in the social network is achieved.

Description

Social network event correlation analysis method based on heterogeneous network

Technical Field

The invention belongs to the technical field of networks, and particularly relates to a social network event correlation analysis method based on a heterogeneous network.

Background

A social network is the creation of an online community for a group of people with the same interests and activities. Such services are often based on the internet, and provide various interaction paths for users to contact and exchange. Microblogs (microblogs) are representative of social networks. Is a social form that allows users to update short text instantly and to publish it publicly. It allows anyone to read or only by a group selected by the user. Compared with the event publishing of the traditional media, the short text and the instant characteristic of the microblog enable the event to be published and spread more rapidly, and the influence of the social network is wider. Event analysis research for social networks has therefore attracted a great deal of research.

In the traditional social network analysis, a homogeneous network analysis method is used, and analysis is performed on a single type of object. For example, unilaterally analyzing the content of the distribution for the users in the network, and the analysis result emphasizes the correlation between similar objects. In reality, the content and form of the social network tend to be complicated and diversified, such as different users communicating with the same content, the interest level of the same user in different content, and so on. The association between different types of objects is becoming more and more compact. The original homogeneous network analysis cannot effectively reflect the connection among different types of objects in the social network, so that a heterogeneous network analysis method is introduced to further analyze the social network.

Since the current social network tends to be complicated and diversified, and various potential relations exist among various objects of different types, the introduction of heterogeneous networks provides an important means for the analysis of complex networks.

For the community discovery of heterogeneous networks, Commor P M et al propose a method combining the community discovery of heterogeneous networks and graph classification. By studying multi-task learning (multi-task learning) of heterogeneous networks, two sub-networks of homogeneous networks are derived from one heterogeneous network, one for classification and one for community division. Through the relevance of the two sub-networks in the heterogeneous network, the two sub-networks are classified and socially divided at the same time. Grady D et al propose a link classification and significant link mining method for heterogeneous networks. Link characteristics of heterogeneous networks are described by link saliency (salient links), a parameter of a new classification of link characteristics of complex networks. The network can be divided into a dominant link structure and a non-dominant link structure by utilizing significance. For relevance analysis in social networks, Liu L et al propose heterogeneous multi-network based node impact analysis. And researching the influence and the propagation relationship among friends in the social network, and combining the link star systems in the heterogeneous network with the text information to calculate the direct influence of the nodes of the topic level. On the basis of the influence calculation, the article studies the propagation and aggregation of influences to analyze indirect influences between nodes.

However, the analysis of the traditional homogeneous network in the social network has the following problems: all objects of a social network cannot be reflected in the same network; the characteristics of different objects of a social network cannot be reflected and analyzed in the same network; communities found in the homogeneous network cannot accurately reflect the conditions of the social network; the interrelationship between different objects in a social network cannot be discovered.

Disclosure of Invention

In order to solve the problem of homogeneous networks in social network analysis, the invention provides a social network event correlation analysis method based on heterogeneous networks.

The specific technical scheme of the invention is as follows: a social network event correlation analysis method based on heterogeneous networks specifically comprises the following steps:

step 1: constructing a figure-event heterogeneous network in a social network space;

step 2: extracting information characteristics of nodes and edges in the person-event heterogeneous network;

and step 3: under the connection limit of a character-event heterogeneous network, carrying out community division to obtain a character community and an event community;

and 4, step 4: and by means of the character-event association constraint, heterogeneous community division is performed on the character-event heterogeneous network, and the purpose of event association analysis is achieved by means of research on division results.

The invention has the beneficial effects that: the method for analyzing the event association of the social network based on the heterogeneous network increases the structures of the user, the event and the user event in the social network by utilizing the heterogeneous network, and achieves the purpose of mining the event association in the social network by further analyzing the event in the social network.

Drawings

Fig. 1 is a structural diagram of a social network, and there are mainly three types of heterogeneous objects: event, author, forwarder. The author and the forwarder can be classified as a character, meanwhile, mutual connection exists between every two heterogeneous objects, and due to the fact that the real social network tends to be complex, the heterogeneous network has good expansibility and intuitiveness, and a powerful method is provided for analysis of the social network.

FIG. 2 is an application of a heterogeneous network to a social network, in which different types of objects of the social network are corresponding to different types of nodes in the heterogeneous network, and connections between the social network objects are corresponding to connection edges of the nodes.

FIG. 3 is a division result of a homogeneous community of a community discovery of a person-event heterogeneous network without considering the influence of heterogeneous communities. The method can analyze some characteristics of the social network, such as the interrelationship between people and events, and the like.

Fig. 4 is a community discovery result of a person-event heterogeneous network under a heterogeneous association constraint condition, and it can be seen that, on the basis of homogeneous community division, the heterogeneous association constraint can provide a way to further divide homogeneous objects, and can more effectively study the association between events (persons). An effective method is provided for further analysis of the social network.

Detailed Description

The embodiments of the present invention will be further described with reference to the accompanying drawings.

A heterogeneous network may consist of octaves (H ═ V, E, ∑ V, ∑ E, l_V,l_ES, d) as shown in fig. 1. Wherein, the node set in the network is V ═ (V)_ij) In the network, the edge set is E ═ E (E)_ij) The node classes in the network are collected asThe edge classes in the network are collected intoThe mapping of nodes in the network to their feature vectors isThe mapping of edges to their feature vectors in the network iss is a source node set, and d is a destination node set.

The method provided by the embodiment of the invention mainly achieves the following aims through the construction and analysis of the heterogeneous network of the social network: aiming at the characteristics of the multi-type objects of the social network, a heterogeneous network is constructed and accurately described; accurately reflecting the characteristics of different types of objects of the social network through the characteristic extraction and analysis of the heterogeneous network; in the heterogeneous network, a community discovery algorithm is used for extracting communities existing in the social network, and on the basis of community discovery, association constraint of the heterogeneous network is used for further association analysis of the communities, so that connection behind the communities is mined.

The method comprises the following specific steps:

step 1: constructing a figure-event heterogeneous network in a social network space;

step 2: extracting information characteristics of nodes and edges in the person-event heterogeneous network;

and step 3: under the connection limit of a character-event heterogeneous network, carrying out community division to obtain a character community and an event community;

and 4, step 4: and by means of the character-event association constraint, heterogeneous community division is performed on the character-event heterogeneous network, and the purpose of event association analysis is achieved by means of research on division results.

The specific method for constructing the person-event heterogeneous network in the social network space in the step 1 is as follows:

social networks mainly contain two main classes of different types: people and events. On the social network, people can achieve the purpose of a media source by publishing event related information. Other people can propagate the event through functions of attention, forwarding, comment and the like. By constructing a heterogeneous network of people and events, the social connection of the social network can be roughly described, and the effect of better describing the events is achieved.

Taking a representative Twitter of a social network and a microblog as an example, the steps for constructing the person-event heterogeneous network are as follows:

(1) analyzing information issued by a user in the social network by using a cluster analysis method to obtain a cluster result, and screening and optimizing the cluster result to obtain a summary of events, wherein the summary is used as an event node of a character-event heterogeneous network;

(2) collecting user information on the social network, wherein the user information comprises information such as user ID, user name, creation time, geographic coordinates and the like, and the user name ID is used as a unique identifier to construct a character node of the character-event heterogeneous network;

(3) and constructing the node connecting edges of the character-event heterogeneous network according to the relationship between the events, the relationship between the users and the events.

As shown in fig. 2, the Twitter person-event heterogeneous network is represented as H ═ V (V, E, ∑ V, ∑ E, l_V,l_ES, d), the set of network nodes is V ═ V (V)₁₁,v₁₂,v₂₁,v₂₂,v₂₃,v₂₄,v₃₁,v₃₂) Wherein v is₁₁Refer to a person 1, v in a social network₁₂Person 2 in the social network; v. of₂₁Refer to events 1, v in a social network₂₂Refer to event 2, v₂₃Refer to event 3, v₂₄Refers to event 4; v. of₃₁To a forwarder 1, v in a social network₃₂Referred to as forwarder 2 in the social network.

The network edge set is shown in formula (2-1), wherein e₁₁Refer to other people in the social network who have social contact with the author 1, e₁₂Refer to other people in the social network who have social contact with author 2; e.g. of the type₂₁Refer to the authoring relationship associated with event 1, e₂₂Refer to the authoring relationship associated with event 2, e₂₃Refer to the authoring relationship associated with event 3; e.g. of the type₃₁Relating to event 1, e₃₂Relating to event 2, e₃₃Refers to the incidence relation of event 3; e.g. of the type₄₁Refers to the forwarding relation associated with event 1, e₄₂Refers to the forwarding relationship associated with event 2. The network node classes are collected asThe network edge class set is shown in formula (2-2), and the source node set is s ═ v₁₁,..), and the destination node set is d ═ v ·₁₂,......)。

E＝(e₁₁,e₁₂,e₂₁,e₂₂,e₂₃,e₃₁,e₃₂,e₃₃,e₄₁,e₄₂) (2-1)

By constructing the character-event heterogeneous network, the social network is abstracted into a heterogeneous network model, and the purpose of accurately describing the complex social network is achieved.

The feature extraction of the person-event heterogeneous network in the step 2 is specifically as follows:

aiming at different types of objects and different relations among the objects in the social network, a series of characteristics can be used for describing, and the purpose of perfecting the characteristics of the heterogeneous network is achieved. On the basis of the heterogeneous network constructed in the above, the feature information of each node and edge in the network is further collected, and the construction of the heterogeneous network is perfected.

Taking the above person-event heterogeneous network as an example, the node feature extraction of the person object is shown in table 1:

TABLE 1

The node feature extraction of the event object is shown in table 2:

TABLE 2

Event node characteristics
	Time, place of occurrence of event
Event keywords (e.g. hashtags, etc.)
	The type of event (e.g. sports, entertainment, politics, etc.)
Number of authors, forwarders, etc. included in the event
	Influence value of event

Node-edge feature extraction between various objects is shown in table 3:

TABLE 3

Character-event	Event-event	User-user
			Forward (authoring) time	Node feature cosine similarity	Node feature cosine similarity
Time of interest	Interval of occurrence	Attention or attention-being
			Number of events (clusters)

The features of the above heterogeneous networks are abstracted into a set of vectors, as shown in table 4. In the event feature vector,refer to the feature vector of event 3, which includes: time, location, class to which the tag belongs, keywords, hashtag, author authors and forwarders. In the context of the Twitter feature vector,refer to the feature vector of person 1 in the social network, which includes: ID, profile, geographic location, friends, original events writers, and forward events rewriter. In the event-event feature vector,what represents is a feature vector of the relationship between event 1 and other events, including: correlation matrix between eventsThe event is associated with a time. In the author-event feature vector,refer to the feature vector between the character and the original event, including: the time a person posted the event, the event that the event persisted, and the person posted the event. In the forwarder-event feature vector,refers to a feature vector between a person and a forwarding event, including: the time of the person forwarding the event, the event that the event persists, and the person forwarding the event. In the user-user feature vector,refer to the relationship between people in a social network, including: correlation matrix between charactersInter-concern relationships between authors.

TABLE 4

Obtaining an event-event incidence matrix:wherein, the element of the ith row and the jth column refers to the relationship between the event i and the event j.

The user-user association matrix is shown in formula (2-3), wherein the element in the ith row and the jth column refers to the relationship between the character i and the character j.

User-event correlation matrix: a. the_U-EWhere the element in the ith row and jth column refers to the relationship between person i and event j (e.g., forwarding, originality, etc.).

$A_{U-U}=\left\{S(l_{v_{ij}},l_{v_{kt}})\right\}---(2-3)$

As shown in fig. 3, feature extraction in a Twitter persona-event heterogeneous network can be represented by the following set: the set of mappings of nodes to their feature vectors is shown in equation (2-4), where,refers to the characteristic relationship between users, wherein the first item is the ID number of the user, the second item is the friend characteristic of the user,refer to the characteristic relation between events, wherein the first term is the similarity between events, and the second term is the characteristic value set of events. Set of mappings of edges to their feature vectors in a network

$l_V=\left(\begin{array}{c}{l}_{v_{11}}=(376573526,(5,2,1\left)\right),\\ l_{v_{31}}=(0,(\left(battery,bgr,cambridge\right),\left(school\right)\left)\right)\end{array}\right)---(2-4)$

The method comprises the following steps of carrying out community division under the connection limit of a character-event heterogeneous network to obtain a character community and an event community as follows:

through the construction and feature extraction of the character-event heterogeneous network, the community discovery is further carried out on the character-event heterogeneous network, and the relevance of the character-character, the event-event and the character-event is obtained.

Community discovery may be performed in a person-event heterogeneous network according to the following criteria:

the relationship between the characters can be described by the behavior of whether the two are related to each other. The closeness of the relationship between the two can be described in terms of the time span of the contact, and the longer the contact time is, the closer the relationship between the two is. The importance of the relationship between the two can be measured according to the importance of the people (in the twitter social network, for example, the number of friends and fans of the people, the number of sent tweets, and the like).

The event-event relationship can be described by whether there are the same features between the two (in a twitter social network, such as hashtags, people, content, places, etc. of the event). The degree of closeness of the relationship between the two events can be described according to the similarity of the two events, and if the similarity of the features between the events is high, the closer the relationship between the two events is, the higher the degree of connection between the events is. Meanwhile, the importance degree of similar events is described according to the characteristics of the events (in the twitter social network, such as the forwarding number and the comment number of the events).

The relationship between the characters and the events can be described by the behaviors between the two (in the twitter social network, such as the publishing, forwarding, commenting and the like of the characters about the events), and the closeness degree between the two can be described according to the quantity and quality of the information related to the event published by the characters (in the twitter social network, such as the tweets, the commenting and the like of the characters about the events). If the quantity and quality of the published information are higher, the closer the relationship between the two is, the higher the attention of the person to the event is. Meanwhile, the importance degree of the relation between the character and the event is described according to the characteristics of the importance degree of the character (in the twitter social network, such as the number of friends of the character, the number of fans, the number of sent tweets and the like) and the importance degree of the event (in the twitter social network, such as the characteristics of the number of forwarded events, the number of comments and the like).

According to the three standards, community discovery is carried out on the twitter character-event heterogeneous network, and the community discovery is expressed by a mathematical set as follows:

character-character incidence matrix A_U-UObtaining a character-task community division result C through user social relation and continuous edge community division_U-U：

$A_{U-U}=\left\{S(l_{v_{ij}},l_{v_{kt}})\right\}\→C_{U-U}$

Event-event correlation matrix A_E-EEvent-event community division result C is obtained through event content association degree and non-connected subgraph mining_E-E：

$A_{E-E}=\left\{S(l_{v_{3i}},l_{v_{3j}})\right\}\→C_{E-E}$

People-event correlation matrix A_U-EConstraining community division results C by social relationship between users and events_U-UAnd C_U-E。

In the step 4, heterogeneous communities are divided for the person-event heterogeneous network through the person-event association constraint, and the specific process of achieving the purpose of event association analysis through the research on the division results is as follows:

through the community discovery of the person-event heterogeneous network, the original loose network structure is divided into communities with more compact structures. Through the result analysis of the communities, the relevance among events in the network can be obtained, and even the relevance hidden under the surface can be obtained. Therefore, on the basis of network community division, people and event communities are further restricted and divided, and the purpose of mining event relevance is achieved.

Based on the above analysis of the person-event heterogeneous network, an optimization objective function F is proposed:

$F=min\left|\right|A_{U-E}-C_{U-U}{A}_{U-E}{C}_{E-E}|{|}_F^2$

wherein,representing a two-norm.

The objective function is to divide the event-event community by adjusting the result C_E-EAnd the division result closest to the actual situation of the heterogeneous network is achieved, and the event relevance of the social network is obtained on the basis of comparing the division results.

For example, in FIG. 3, the community division result can be obtained for community discovery of a person-event heterogeneous network

$C_{U-U}=\left[\begin{array}{ccccc}1& 1& 1& 0& 0\\ 1& 1& 1& 0& 0\\ 1& 1& 1& 0& 0\\ 0& 0& 0& 1& 1\\ 0& 0& 0& 1& 1\end{array}\right],$

Event community division result

Person-event constraint

$C_{U-U}{A}_{U-E}{C}_{E-E}=\left[\begin{array}{cccc}0& 0& 0& 0\\ 0& 0& 0& 0\\ 0& 0& 0& 0\\ 2& 2& 2& 0\\ 2& 2& 2& 0\end{array}\right],$

The objective function F is 4.2426.

By adjusting community division, as shown in fig. 4, event community repartitioning results are shown in formula (2-5), repartitioning results are shown in formula (2-6), and an objective function F is 1.4142, so that a better objective node can be obtainedFruit, so C'_E-EIs a better event-event community division result.

$C_{E-E}^{\′}=\left[\begin{array}{cccc}1& 0& 0& 0\\ 0& 1& 1& 0\\ 0& 1& 1& 0\\ 0& 0& 0& 1\end{array}\right]---(2-5)$

$C_{U-U}{A}_{U-E}{C}_{E-E}^{\′}=\left[\begin{array}{cccc}0& 0& 0& 0\\ 0& 0& 0& 0\\ 0& 0& 0& 0\\ 2& 0& 0& 0\\ 2& 0& 0& 0\end{array}\right]---(2-6)$

In summary, the social network is analyzed through the steps of heterogeneous network construction, feature extraction, community discovery, relationship constraint and the like. On the basis of a person-event heterogeneous network, the target function F can further divide the originally divided event-event communities through a constraint matrix, so that the means and the method for event correlation analysis are improved, and the accurate analysis of event correlation is facilitated.

The invention introduces a heterogeneous network on the basis of the original social network analysis, extracts the characteristics of the heterogeneous network by constructing a character-event heterogeneous network, and further researches the social network by community discovery and association analysis of the heterogeneous network, and brings the following effects by introducing the heterogeneous network:

through the construction of the heterogeneous network, the mutual relation between heterogeneous objects in the social network is reflected, the current trend of complicated and multi-element social network development trend is adapted, and the expansibility and the innovation are good;

selecting and extracting features of a heterogeneous network, imaging abstract social network objects, and providing a method and a means for further analyzing the social network by establishing a feature vector, a feature set, an incidence matrix and the like;

according to the further analysis of the heterogeneous network, on the basis of homogeneous community division, the community is further divided by utilizing heterogeneous association constraint, so that the accuracy of community division is improved, and a new method is provided for mining the relation behind the community.

The above description is only exemplary of the present invention and should not be taken as limiting the scope of the present invention, and any modifications, equivalents, improvements and the like that are within the spirit and principle of the present invention should be included in the present invention.

Claims (3)

1. A social network event correlation analysis method based on heterogeneous networks specifically comprises the following steps:

step 1: constructing a figure-event heterogeneous network in a social network space;

step 2: extracting information characteristics of nodes and edges in the person-event heterogeneous network;

and step 3: under the connection limit of a character-event heterogeneous network, carrying out community division to obtain a character community and an event community;

and 4, step 4: and by means of the character-event association constraint, heterogeneous community division is performed on the character-event heterogeneous network, and the purpose of event association analysis is achieved by means of research on division results.

2. The method for analyzing the correlation between the events in the social network based on the heterogeneous network as claimed in claim 1, wherein the specific process for constructing the person-event heterogeneous network in the social network space in step 1 is as follows:

analyzing information issued by a user in the social network by using a cluster analysis method to obtain a cluster result, and screening and optimizing the cluster result to obtain a summary of events, wherein the summary is used as an event node of a character-event heterogeneous network;

collecting user information on the social network, wherein the user information comprises information such as user ID, user name, creation time, geographic coordinates and the like, and the user name ID is used as a unique identifier to construct a character node of the character-event heterogeneous network;

and constructing the node connecting edges of the character-event heterogeneous network according to the relationship between the events, the relationship between the users and the events.

3. The method of claim 2, wherein the representation of the Twitter person-event heterogeneous network is H ═ (V, E, ∑ V, ∑ E, l)_V,l_ES, d), the set of network nodes is V ═ V (V)₁₁,v₁₂,v₂₁,v₂₂,v₂₃,v₂₄,v₃₁,v₃₂) Wherein v is₁₁Refer to a person 1, v in a social network₁₂Person 2 in the social network; v. of₂₁Refer to events 1, v in a social network₂₂Refer to event 2, v₂₃Refer to event 3, v₂₄Refers to event 4; v. of₃₁To a forwarder 1, v in a social network₃₂Referred to as forwarder 2 in the social network.

The network edge set is shown in formula (2-1), wherein e₁₁Refers to social networking with author 1Other persons in contact, e₁₂Refer to other people in the social network who have social contact with author 2; e.g. of the type₂₁Refer to the authoring relationship associated with event 1, e₂₂Refer to the authoring relationship associated with event 2, e₂₃Refer to the authoring relationship associated with event 3; e.g. of the type₃₁Relating to event 1, e₃₂Relating to event 2, e₃₃Refers to the incidence relation of event 3; e.g. of the type₄₁Refers to the forwarding relation associated with event 1, e₄₂Refers to the forwarding relationship associated with event 2. The network node classes are collected asThe network edge class set is shown in formula (2-2), and the source node set is s ═ v₁₁,..), and the destination node set is d ═ v ·₁₂,......)。

E＝(e₁₁,e₁₂,e₂₁,e₂₂,e₂₃,e₃₁,e₃₂,e₃₃,e₄₁,e₄₂) (2-1)

By constructing the character-event heterogeneous network, the social network is abstracted into a heterogeneous network model, and the purpose of accurately describing the complex social network is achieved.