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

A kind of social networkies event relation analyzing method based on heterogeneous network

Technical field

The invention belongs to networking technology area is and in particular to a kind of social networkies event correlation based on heterogeneous network is analyzed Method.

Background technology

Social networkies are to have same interest for a group to create on-line communities with the people of activity.This kind of service is often based on The Internet, provides the user various contacts, the interactive path of exchange.Microblogging (microblog) is as the representative of social networkies.It is A kind of social form allowing user's immediate updating brief text and can publishing.It allows anyone to read or can only Read by the group that user selects.Issue with respect to the event of traditional media, the short text of microblogging and instant feature make event More rapid in issuing propagation, the power of influence of social networkies is also more extensive.Therefore for the event analysis of social networkies Research has attracted substantial amounts of research.

In traditional social network analysis, employ the method that homogenous network is analyzed, the object for single type enters Row analysis.For example it is directed to the user in network, issue content etc. and be unilaterally analyzed, the result of analysis lays particular emphasis on homogeneous object Between association situation.And in reality, the content and form of social networkies trends towards complicating and diversification, for example not Exchanged for same content with user, same user is to level of interest of different content etc..Between different object Association more and more tightr.And the analysis of original homogenous network can not effectively reflect the different object in social networkies Between contact, so introducing heterogeneous network analysis method that social networkies are further analyzed.

Because current social networkies trend towards complicating and diversification, exist various between the object of number of different types Potential relation, so the introducing of heterogeneous network, is that the analysis of complex network provides a kind of important means.

For the community discovery of heterogeneous network, comar p m et al. proposes community discovery and the figure point of heterogeneous Multi net voting The method that class combines.By studying the multi-task learning (multi-task learning) of heterogeneous network, from a heterogeneous network The middle sub-network deriving two homogenous network, a sub-network is used for classifying, and a sub-network is used for corporations and divides.By two Relatedness in heterogeneous network for the sub-network, is classified to the two simultaneously and corporations divide.Grady d et al. proposes different Link classification and the notable link method for digging of matter network.By a kind of parameter of the new classification of complex network link characteristics The significance (salient links) of link is describing the link characteristics of heterogeneous network.Network can be divided into using significance Notable link structure and non-significant link structure.For the correlation analysis in social networkies, liu l et al. proposes based on different The node power of influence analysis of matter Multi net voting.Power of influence between friend and its propagation relation in research social networkies, by heterogeneous network Link galaxy in network is combined with text message, to calculate the node direct influence of topic level.Calculate in power of influence On the basis of, article have studied the propagation of power of influence and is polymerized, with the indirect influence between analysis node.

But traditional homogenous network is in the analysis in social networkies there is problems in that society can not be reflected in consolidated network Hand over all objects of network；The feature with the analysis different object of social networkies can not be reflected in consolidated network；In homogenous network The corporations finding can not accurately reflect the situation of social networkies；The mutually interconnection between different objects in social networkies cannot be found System.

Content of the invention

For solving the problems, such as homogenous network in social network analysis, the present invention proposes a kind of society based on heterogeneous network Hand over network event association analysis method.

The concrete technical scheme of the present invention is: a kind of social networkies event relation analyzing method based on heterogeneous network, tool Body includes following steps:

Step 1: build personage-event heterogeneous network in social network space；

Step 2: extract the information characteristics on personage-event heterogeneous network interior joint and side；

Step 3: under company's side constraint of personage-event heterogeneous network, carry out corporations' division, obtain personage corporations, event Corporations；

Step 4: by personage-event correlation constraint, the division of heterogeneous corporations is carried out to personage-event heterogeneous network, passes through Research to division result, reaches the purpose of event correlation analysis.

Beneficial effects of the present invention: the social networkies event relation analyzing method based on heterogeneous network of the present invention utilizes different Matter network, increases user in social networkies, the construction of event and three figures of customer incident, by entering to the event in social networkies Row is further to be analyzed, and reaches the purpose excavating the event correlation in social networkies.

Brief description

Fig. 1 is the structure chart of social networkies, mainly has the heterogeneous object of three classes: event, author, turns originator.Wherein author and Turn originator and all can be classified as personage one class, exist between heterogeneous object two-by-two simultaneously and connect each other, due to real social networkies Tend to complicated, heterogeneous network because having good expansibility and intuitive, the analysis for social networkies provides one Plant powerful method.

Fig. 2 is application on social networkies for the heterogeneous network, and the different object of social networkies is corresponded to heterogeneous network Dissimilar node in network, the contact between social networking objects is corresponded to the Lian Bianzhong of node.

Fig. 3 is the community discovery of personage-event heterogeneous network, the homogeneity society in the case of not considering the impact of heterogeneous corporations The division result of group.The method can be analyzed to some features of social networkies, the phase and event between such as between personage Mutual relation etc..

Fig. 4 is the community discovery of personage-event heterogeneous network under the conditions of heterogeneous interconnection constraint as a result, it is possible to see same On the basis of matter corporations divide, heterogeneous interconnection constraint can be to provide a kind of mode that homogeneity object is further divided, can With the relatedness between significantly more efficient research event (personage).Analysis further for social networkies provides one kind effectively Method.

Specific embodiment

Below in conjunction with the accompanying drawings embodiments of the invention are described further.

Heterogeneous network refers to different types of object as node, and the different relations between different objects are given birth to as side The network becoming.One heterogeneous network can be made up of h=(v, e, ∑ v, ∑ e, l eight tuples_v,l_e, s, d), as shown in Figure 1.Its In, nodes collection is combined into v=(v_ij), in network, line set is e=(e_ij), nodes category set isIn network, side class set is combined intoNodes being mapped as to its characteristic vectorSide being mapped as to its characteristic vector is connected in networkS saves for source Point set, d is purpose node set.

Method provided in an embodiment of the present invention mainly reaches following mesh by the heterogeneous network creation analysis of social networkies : for the feature of social networkies polymorphic type object, build heterogeneous network and accurate description is carried out to it；By the spy of heterogeneous network Levy extraction and analysis, accurately reflect the feature of social networkies different object；In heterogeneous network, calculated using community discovery Method, extracts corporations present in social networkies, on the basis of community discovery, using the interconnection constraint of heterogeneous network, to corporations Carry out further association analysiss, excavate the contact of corporations behind.

Specifically comprise the following steps that

Step 1: build personage-event heterogeneous network in social network space；

Step 2: extract the information characteristics on personage-event heterogeneous network interior joint and side；

Step 3: under company's side constraint of personage-event heterogeneous network, carry out corporations' division, obtain personage corporations, event Corporations；

Step 4: by personage-event correlation constraint, the division of heterogeneous corporations is carried out to personage-event heterogeneous network, passes through Research to division result, reaches the purpose of event correlation analysis.

The concrete grammar building personage-event heterogeneous network in social network space in above-mentioned steps 1 is as follows:

Social networkies mainly comprise main dissimilar of two classes: personage and event.On social networkies, personage passes through to issue Information concerning events, to reach the purpose of source of media.Remaining personage the function such as can forward and comment on, to event by concern Propagated.By the heterogeneous network structure to personage-event, can substantially depict the social connections of social networkies, reach To the more preferable effect describing event.

, the step building personage-event heterogeneous network is as follows taking the representative twitter of social networkies and micro-blog as a example:

(1) method using cluster analyses, the information that user in social networkies is issued is analyzed, and obtains cluster knot Really, screening and optimizing is carried out to cluster result, obtain the summary of event, in this, as the event node of personage-event heterogeneous network；

(2) user profile on social networkies is collected, including the id of user, user name, creation time, geographical seat The information such as mark, using user name id as unique mark, build personage's node of personage-event heterogeneous network；

(3) according to the relation between event and event, the pass between the relation between user and user and user and event System, the node building personage-event heterogeneous network connects side.

As Fig. 2, twitter personage-event heterogeneous network be expressed as h=(v, e, ∑ v, ∑ e, l_v,l_e, s, d), network Node set is v=(v₁₁,v₁₂,v₂₁,v₂₂,v₂₃,v₂₄,v₃₁,v₃₂), wherein, v₁₁Refer to the personage 1, v in social networkies₁₂Refer to It is the personage 2 in social networkies；v₂₁Refer to the event 1, v in social networkies₂₂Refer to event 2, v₂₃Refer to event 3, v₂₄ Refer to event 4；v₃₁Refer to social networkies in turn originator 1, v₃₂Refer to social networkies in turn originator 2.

Shown in network line set such as formula (2-1), wherein, e₁₁Refer to, in social networkies, there is social connections with author 1 Other personages, e₁₂Refer in social networkies, have other personages of social connections with author 2；e₂₁Refer to and event 1 phase The creation relation of association, e₂₂Refer to the creation relation being associated with event 2, e₂₃The creation referring to be associated with event 3 is closed System；e₃₁Refer to the incidence relation of event 1, e₃₂Refer to the incidence relation of event 2, e₃₃Refer to the incidence relation of event 3； e₄₁Refer to the forwarding relation being associated with event 1, e₄₂Refer to the forwarding relation being associated with event 2.Network node classification Collection is combined intoShown in network edge class set such as formula (2-2), source node Collection is combined into s=(v₁₁...), destination node collection is combined into d=(v₁₂,......).

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

By the structure of personage-event heterogeneous network, social networkies are abstracted in heterogeneous network model, reach to complexity The purpose of the accurate description of social networkies.

In above-mentioned steps 2, the feature extraction of personage-event heterogeneous network is specific as follows:

For the different relations between object different types of in social networkies and object, a series of feature can be used To be portrayed, to be reached the purpose improving heterogeneous network feature.On the basis of the heterogeneous network hereinbefore building, collect further The characteristic information on each node and side in network, improves the structure of heterogeneous network.

, the node diagnostic of who object extracts as shown in table 1 taking personage above-event heterogeneous network as a example:

Table 1

The node diagnostic of event object extracts as shown in table 2:

Table 2

Event node feature
	The time that event occurs, place
Event key word (such as: hashtags etc.)
	The affiliated type of event (such as: physical culture, amusement and politics etc.)
Author's quantity included by event, turns originator quantity etc.
	The impact force value of event

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

Table 3

Personage-event	Event-event	User-user
			(creation) time of forwarding	Node diagnostic cosine similarity	Node diagnostic cosine similarity
The concern time	There is interval of events	Pay close attention to or be concerned
			Event (cluster) quantity

The feature abstractionization of above heterogeneous network is arrived vector set, specifically as shown in table 4.In affair character vector,Refer to the characteristic vector of event 3, which includes: time time, position location, generic class, key word Keywords, label hastag, author authors and turn the features such as originator rewritters.In twitter characteristic vector,Refer to the characteristic vector of personage 1 in social networkies, including: id, personal information profile, geographical position Location, good friend friends, the feature such as original event writters and forwarding event rewritter.Special in event-event Levy in vector,Represent is the characteristic vector of relation between event 1 and other events, comprising: interrelated between event MatrixThe event correlation time.In author-affair character vector,Refer between personage and original event Characteristic vector, including: personage issues the time of event, and the event of event duration and personage issue event.Turning originator-thing In part characteristic vector,Refer to the characteristic vector between personage and forwarding event, including: personage forward event when Between, the event of event duration and personage forward event.In user-user characteristic vector,Refer to personage in social networkies Relation and personage between, including: the correlation matrix between personageMutual concern relation between author.

Table 4

Obtain event-event correlation matrix:Wherein, the i-th row, the element of jth row refers to thing Relation between part i and event j.

Shown in user-user incidence matrix such as formula (2-3), wherein, the i-th row, the element of jth row refer to personage i with Relation between personage j.

User-event correlation matrix: a_u-e={ write or retweet }, wherein, the i-th row, the element that jth arranges refers to It is relation between personage i and event j (as forwarded and original etc.).

$a_{u-u}=\left\{s(l_{v_{ij}},l_{v_{kt}})\right\}---(2-3)$

As shown in figure 3, the feature extraction in twitter personage-event heterogeneous network can be represented with following set: section The mapping set such as formula (2-4) of point to its characteristic vector is shown, wherein,Refer to the characteristic relation between user, wherein Section 1 is user No. id, and Section 2 is good friend's feature of user,Refer to the characteristic relation between event, wherein Section 1 For the similarity between event, Section 2 is the characteristic value collection of event.In network, side is to the mapping set of its characteristic vector

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

Above-mentioned steps, under company's side constraint of personage-event heterogeneous network, carry out corporations' division, obtain personage corporations, thing Part corporations are specific as follows:

By structure and the feature extraction of personage-event heterogeneous network, further society is carried out to personage-event heterogeneous network Group finds, obtains personage-personage, event-event, the relatedness of personage-event three.

Can carry out community discovery according to following standard in personage-event heterogeneous network:

Relation between personage-personage can be described by whether there being therebetween the behavior interkniting.Two Between person, the tightness degree of relation can describe according to the time span of contact, and the time of contact is longer, then therebetween Relation is tightr.Therebetween the importance of relation can be weighed according to the importance of personage (in twitter social networkies, As good friend's number of personage, vermicelli number, send quantity pushing away literary composition etc.).

Can be by whether there being therebetween identical feature (social in twitter in the relation between event-event In network, the hashtags of such as event, personage, content, place etc.) describing.Therebetween the tightness degree of relation can root To describe according to the similarity of the two, if the characteristic similarity between event is high, relation therebetween is tightr, event Degree of contact and event between is higher.Feature according to event is (in twitter social networkies, as the forwarding of event simultaneously Number, the feature such as comment number), to portray the importance degree of similar case.

Can by behavior therebetween (in twitter social networkies, such as in the relation between personage-event Personage with regard to the issue of event, forwards, comment etc.) describing, tightness degree therebetween can issue event according to personage The quality and quantity of relevant information (in twitter social networkies, if personage is with regard to the literary composition that pushes away of event, comment etc.) to describe. If the quality and quantity releasing news is higher, illustrate that relation therebetween is tightr, personage gets over to the degree of concern of event High.Simultaneously the significance level according to personage (in twitter social networkies, good friend's number of such as personage, vermicelli number and transmission push away literary composition Quantity etc.) and event significance level (in twitter social networkies, the such as feature such as the forwarding number of event and comment number) Etc. feature, to portray the significance level of personage-event relation.

According to three above standard, twitter personage-event heterogeneous network is carried out by community discovery, and uses mathematical set It is indicated as follows:

Personage-personage incidence matrix a_u-uObtain personage-task corporations and draw by user social contact relation and the division of Lian Bian corporations Divide result c_u-u:

$a_{u-u}=\left\{s(l_{v_{ij}},l_{v_{kt}})\right\}\→c_{u-u}$

Event-event correlation matrix a_e-eExcavated by the event content degree of association and non-interconnected subgraph and obtain event-event society Group's division result c_e-e:

$a_{e-e}=\left\{s(l_{v_{3i}},l_{v_{3j}})\right\}\→c_{e-e}$

Personage-event correlation matrix a_u-eConstrain corporations' division result c by the social networks between user and event_u-u And c_u-e.

Pass through personage-event correlation constraint in above-mentioned steps 4, the division of heterogeneous corporations carried out to personage-event heterogeneous network, By the research to division result, the detailed process reaching the purpose of event correlation analysis is as follows:

By the community discovery of personage-event heterogeneous network, originally loose network structure is divided into structure one by one Even closer corporations.By the interpretation of result to corporations, the relatedness between event in network can be obtained, or even hide Association under surface.Therefore, on the basis of network community division, personage and event corporations are carried out further about Bundle and division, reach the purpose excavating event correlation.

Based on the analysis of personage above-event heterogeneous network, optimization objective function f proposed:

$f=min\left|\right|a_{u-e}-c_{u-u}{a}_{u-e}{c}_{e-e}|{|}_f^2$

Wherein,Represent two norms.

The purpose of object function is, by adjusting result c that event-event corporations divide_e-e, to reach closest different The division result of matter network practical situation, on the basis of comparing division result, obtains the event correlation of social networkies.

Taking Fig. 3 as a example, community discovery to personage-event heterogeneous network, personage corporations division result can be obtained

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

Event corporations division result

Personage-timing constraints

$c_{u-u}{a}_{u-e}{c}_{e-e}=\left[\begin{array}{cccc}0& 0& 0& 0\\ 0& 0& 0& 0\\ 0& 0& 0& 0\\ 2& 2& 2& 0\\ 2& 2& 2& 0\end{array}\right],$

Object function f=4.2426.

Divided by adjusting corporations, as shown in figure 4, event corporations repartition shown in result such as formula (2-5), again draw Shown in the result such as formula (2-6) divided, object function f=1.4142, more excellent objective result can be obtained, so c'_e-eIt is more excellent Event-event corporations 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 sum, by the structure of heterogeneous network, feature extraction, community discovery, the step such as relation constraint, come to social activity Network is analyzed.Original division can be completed based on, on personage-event heterogeneous network, object function f passes through constraint matrix Event-event corporations further divided, improve event correlation analysis means and method, be conducive to for event The accurate analysis of relatedness.

The present invention, on original social network analysis, introduces heterogeneous network, by building personage-event heterogeneous network Network, extracts feature, the community discovery of heterogeneous network and the association analysiss of heterogeneous network, social networkies is further ground Study carefully, by the introducing of heterogeneous network, bring following effect:

By the structure of heterogeneous network, reflect the mutual relation between heterogeneous object in social networkies, adapted to current Tend to complicate, the social networkies development trend of multielement, there is good expansibility and novelty；

The feature selection of heterogeneous network and extraction, by abstract social networking objects tool as changing, by setting up characteristic vector, Characteristic set, a kind of method such as incidence matrix, there is provided ways and means analyzing social networkies further；

For the analysis further of heterogeneous network, on the basis of homogeneity corporations divide, using heterogeneous interconnection constraint, to society Group is further divided, thus lifting the accuracy of corporations' division, also provides a kind of new method to excavate corporations' back of the body Contact afterwards.

The above, only presently preferred embodiments of the present invention, and non-limiting protection scope of the present invention, all at this Any modification, equivalent substitution and improvement done within bright spirit and principle etc., should be included in protection scope of the present invention it Interior.

Claims (3)

1. a kind of social networkies event relation analyzing method based on heterogeneous network, specifically includes following steps:

Step 1: build personage-event heterogeneous network in social network space；

Step 2: extract the information characteristics on personage-event heterogeneous network interior joint and side；

Step 3: under company's side constraint of personage-event heterogeneous network, carry out corporations' division, obtain personage corporations, event corporations；

Step 4: by personage-event correlation constraint, the division of heterogeneous corporations is carried out to personage-event heterogeneous network, by drawing Divide the research of result, reach the purpose of event correlation analysis.

2. a kind of social networkies event relation analyzing method based on heterogeneous network according to claim 1, its feature exists In in the structure social network space described in step 1, personage-event heterogeneous network detailed process is as follows:

Using the method for cluster analyses, the information that user in social networkies is issued is analyzed, and obtains cluster result, to cluster Result carries out screening and optimizing, obtains the summary of event, in this, as the event node of personage-event heterogeneous network；

User profile on social networkies is collected, including the letter such as the id of user, user name, creation time, geographical coordinate Breath, using user name id as unique mark, builds personage's node of personage-event heterogeneous network；

According to the relation between event and event, the relation between the relation between user and user and user and event, build The node of personage-event heterogeneous network connects side.

3. a kind of social networkies event relation analyzing method based on heterogeneous network according to claim 2, its feature exists It is expressed as h=(v, e, ∑ v, ∑ e, l in, twitter personage-event heterogeneous network_v,l_e, s, d), set of network nodes is v =(v₁₁,v₁₂,v₂₁,v₂₂,v₂₃,v₂₄,v₃₁,v₃₂), wherein, v₁₁Refer to the personage 1, v in social networkies₁₂Finger is social networkies In personage 2；v₂₁Refer to the event 1, v in social networkies₂₂Refer to event 2, v₂₃Refer to event 3, v₂₄Refer to event 4；v₃₁Refer to social networkies in turn originator 1, v₃₂Refer to social networkies in turn originator 2.

Shown in network line set such as formula (2-1), wherein, e₁₁Refer to, in social networkies, there is its of social connections with author 1 He is personage, e₁₂Refer in social networkies, have other personages of social connections with author 2；e₂₁Refer to be associated with event 1 Creation relation, e₂₂Refer to the creation relation being associated with event 2, e₂₃Refer to the creation relation being associated with event 3； e₃₁Refer to the incidence relation of event 1, e₃₂Refer to the incidence relation of event 2, e₃₃Refer to the incidence relation of event 3；e₄₁Refer to Be the forwarding relation being associated with event 1, e₄₂Refer to the forwarding relation being associated with event 2.Network node category set ForShown in network edge class set such as formula (2-2), source node set For s=(v₁₁...), destination node collection is combined into d=(v₁₂,......).

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

By the structure of personage-event heterogeneous network, social networkies are abstracted in heterogeneous network model, reach to complicated social The purpose of the accurate description of network.