CN104573017A - Network water army group identifying method and system - Google Patents

Abstract

Disclosed is a network water army group identifying method and system. The method comprises obtaining irritable response data of multiple network groups; according to the irritable response data, marking out suspicious network water army group samples; according to the suspicious network water army group samples and existing network water arm group samples, determining the irritable response characteristics of network water army groups; according to the irritable response characteristics of the network water army groups, establishing network water army group active identification models; identifying the network water army groups inside the multiple network groups through the network water army group active identification models. The network water army group identifying method and system solves the technical problem that traditional network water army identifying method cannot identify network water armies before the network water armies cause harm, achieves active identification on the network water armies and avoids the harm caused by the network water armies. Besides, by categorizing network users with identical or similar behavioral characteristics into one network group, the network water arm identification accuracy can be enhanced.

Description

The method and system of group of recognition network waterborne troops

Technical field

The present invention relates to a kind of network navy recognition technology field, the specifically method and system of group of a kind of recognition network waterborne troops.

Background technology

Network navy refers to that those are driven by commercial interest, for reaching as affected netizen opinions, upsetting the improper objects such as network environment, manufactures, propagates the general name of the network spam suggestion producers such as false suggestion and junk information in internet.Namely network navy identification use web information service from current internet environment, determines that high discrimination feature and behavior pattern find hiding waterborne troops.Mostly the recognition methods of tradition Passive Network waterborne troops is to carry out after network navy causes certain dangerous act, to the detection of network navy, there is certain hysteresis quality, and its research is under network navy harm drives, pay close attention to the impact that network navy has produced, be subject to the impact of network navy variability deceptive practices.Such network navy recognition methods cannot stop it to occur before network navy works the mischief, poor to the preventive effect of network navy.And the research of network navy and improvement, key is from source containment the spreading unchecked of junk information, and before therefore not producing a large amount of junk information in network navy, causing serious web influence, completes and finds to be of great significance with the work tool of detection to it.Network navy also can be understood as the outlier in the whole network user, but its feature and normal users are very close, has behavior complicated and changeable, therefore identifies the very difficult of single network navy, and network navy Study of recognition is difficult to accomplish comprehensively, accurately.

Network community refers to the group be made up of more than 2 or 2 network users, the group of waterborne troops that namely network navy group is made up of more than 2 or 2 network navy.What network community and network navy group all referred to is the virtual community that the user account that network exists forms, and it is not be made up of the real-life real user of correspondence.Network navy group may be manipulated by some or certain several real user behind, same or analogous attribute or behavior pattern can be formed, such as all appear in certain several target product comment, user's name has certain parallel pattern (all adopting unordered numeral+letter) etc.And network navy group is when being subject to external irritant, after the model that member of community as group of Forum network waterborne troops delivers is commented on, the real user of manipulation network navy group, waterborne troops's user account that can manipulate in network navy group is responded, group properties in its response data is more outstanding, and it is comparatively easy to make its identification.

Summary of the invention

For this reason, technical matters to be solved by this invention is that mostly traditional network navy recognition methods is to carry out after network navy causes certain dangerous act, and cannot before network navy works the mischief, stop it to occur, very poor to the preventive effect of network navy, thus propose a kind ofly before network navy causes certain harm, the method for hiding network navy to be gone out by initiative recognition.

For solving the problems of the technologies described above, technical scheme provided by the invention is as follows:

A method for group of recognition network waterborne troops, comprises the following steps:

Obtain the irritability response data of multiple network community, network community comprises multiple doubtful network user for network navy with same or similar behavioural characteristic and/or attributive character, and irritability response data is that the network user in network community delivers to it response data that content carries out replying the laggard row response of comment and generation other network users;

According to irritability response data, mark group of suspicious network waterborne troops sample;

According to group of suspicious network waterborne troops sample and existing network navy group sample, determine that network navy group irritability responds feature;

Respond feature according to network navy group irritability, set up network navy group initiative recognition model;

Network navy group initiative recognition Model Identification is utilized to go out network navy group in multiple network community.

As optimization, obtain the process of the irritability response data of multiple network community, comprising:

Set up intelligent interaction robot, intelligent interaction robot generates software by intelligent interaction robot and sets up;

Determine that intelligent interaction robot makes an initiative sally strategy, the strategy that makes an initiative sally formulates the corresponding behavior pattern that makes an initiative sally for the behavioral characteristic according to known network navy, comprises time, content, mode;

According to making an initiative sally, strategy carries out the behavior that makes an initiative sally, and the behavior of making an initiative sally is that the mutual robot of operative intelligence comments on the content that network community is delivered or initiatively delivers subject content targetedly;

Collect the response data of multiple network community as irritability response data.

As optimization, the robot account of normal users behavior artificially simulated by intelligent interaction machine, possesses all properties of normal users.

As optimization, the irritability response data of network community is the response data that intelligent interaction robot carries out the network user in network community after the behavior that makes an initiative sally, comprise respond account, respond content, the response time, response account hour of log-on and at least one responded in account grade.

As optimization, the process marking group of suspicious network waterborne troops sample according to irritability response data comprises:

According to irritability response data, mark earliest time and time the latest of all responses of each or subnetwork group;

If the earliest time of all responses of a network community and the latest time interval are within a certain period of time, be labeled as group of suspicious network waterborne troops.

As optimization, network navy group irritability response feature comprises at least one in response tight ness rating, the response time degree of approach, response content similarity, group's response degree.

As optimization, respond tight ness rating and calculated by following formula (I) (II):

$\mathrm{GRC}\left(g\right)={\max }_{r\∈R_g}\left(\mathrm{RC}(g,r)\right)---\left(I\right)$

$\mathrm{RC}(g,r)=\left\{\begin{array}{cc}0& \mathrm{ifL}(g,r)-F(g,r)>\mathrm{\α}\\ 1-\frac{L(g,r)-F(g,r)}{\mathrm{\α}}& \mathrm{otherwise}\end{array}\right.---\left(\mathrm{II}\right)$

Wherein, L (g, r), F (g, r) time the latest that the network user in network community g responds the intelligent interaction robot for it and earliest time is represented respectively, RC (g, r) be that the network user in network community g is to the response tight ness rating for its intelligent interaction robot, GRC (g) represents the response tight ness rating of network community g, that all-network user in network community g responds the maximal value of tight ness rating to the intelligent interaction robot for it, α, for responding tightness parameter, is used for weighing the response tight ness rating of network community g.

As optimization, the response time degree of approach, is calculated by following formula (III) (IV):

$\mathrm{GRTF}\left(g\right)={\max }_{r\∈R_g}\left(\mathrm{RTF}(g,r)\right)---\left(\mathrm{III}\right)$

$\mathrm{RTF}(g,r)=\left\{\begin{array}{cc}0& \mathrm{ifL}(g,r)-A\left(r\right)>\mathrm{\β}\\ 1-\frac{L(g,r)-A(r)}{\mathrm{\β}}& \mathrm{otherwise}\end{array}\right.---\left(\mathrm{IV}\right)$

Wherein, L (g, r), A (r) represents the time the latest that a network user in network community g responds the intelligent interaction robot for it respectively, intelligent interaction robot for this network community carries out the time of the behavior that makes an initiative sally, RTF (g, r) be that the network user in network community g is to the response time degree of approach for its intelligent interaction robot, GRTF (g) represents the response time degree of approach of network community g, be in network community g all-network user to the maximal value of the response time degree of approach for its intelligent interaction robot, β is response time proximity parameters, be used for weighing the response time degree of approach of network community.

As optimization, respond content similarity and calculated by following formula (V) (VI):

$\mathrm{GRCS}\left(g\right)={\max }_{r\∈R_g}\left(\mathrm{RCS}(g,r)\right)---\left(V\right)$

$\mathrm{RCS}(g,r)={\mathrm{avg}}_{m_i,m_j\&Element;g,i<j}\left(\cos (\mathrm{rc}(m_i,r),\mathrm{rc}(m_j,r))\right)---\left(\mathrm{VI}\right)$

Wherein, rc (m _i, r), rc (m _j, r) represent the network user m in network community g respectively _iand m _jto the content that the intelligent interaction robot for it responds, cos (rc (m _i, r), rc (m _j, r)) and be used for computational grid user m _iresponse content and m _jthe similarity of response content, RCS (g, r) be the average of the response content of a network user in network community g and the response content similarity of the every other network user, GRCS (g) represents the response content similarity of network community g, is that all-network user in network community g responds the maximal value of the average of content similarity to the intelligent interaction robot for it.

As optimization, group's response degree, is calculated by following formula (VII) (VIII):

$\mathrm{GRE}\left(g\right)={\max }_{r\&Element;R_g}\left(\mathrm{RE}(g,r)\right)---\left(\mathrm{VII}\right)$

$\mathrm{RE}(g,r)=\frac{\left|R_g\right|}{g}---\left(\mathrm{VIII}\right)$

Wherein, | R _g|, g represents that member of community in network community g is to for the response number of its one of them intelligent interaction robot and the total number of persons of network community g respectively, RE (g, r) be that network community g is to the group's response degree for its one of them intelligent interaction robot, GRE (g) represents group's response degree of network community g, is the maximal value of network community g to group's response degree of all intelligent interaction robots for it.

As optimization, set up network navy group initiative recognition model according to network navy group irritability response feature and comprise:

Respond feature according to network navy group irritability, set up network navy group irritability and respond model;

Respond model according to network navy group irritability, set up network navy group initiative recognition model.

As optimization, set up network navy group irritability and respond model, comprise following concrete steps:

Respond feature according to calculated network navy group irritability, utilize formula (Ⅸ), the suspicious degree of irritability of computational grid group:

$\mathrm{GS}\left(g\right)={\&PartialD;}_1\mathrm{GRC}\left(g\right)+{\&PartialD;}_2\mathrm{GRTF}\left(g\right)+{\&PartialD;}_3\mathrm{GRCS}\left(g\right)+{\&PartialD;}_4\mathrm{GRE}\left(g\right)---\left(\mathrm{IX}\right)$

Wherein, GS (g) represents the suspicious degree of the irritability of network community g, GRC (g), GRTF (g), GRCS (g), GRE (g) represent the response tight ness rating of network community, the response time degree of approach respectively, respond content similarity and group's response degree feature for the undetermined coefficient factor, by the weight deciding group of heterogeneous networks waterborne troops irritability feature.

As optimization, network navy group initiative recognition Model Identification is utilized to go out network navy group in multiple network community:

The suspicious degree of irritability that model calculates each network community in multiple network community is responded according to network navy group irritability;

Suspicious for the irritability of each network community degree and the suspicious degree threshold value pre-set are compared, the network community being greater than the suspicious degree correspondence of irritability of suspicious degree threshold value is network navy group; Remaining is normal group.

A system for group of recognition network waterborne troops, comprising:

Irritability data acquisition module: for obtaining the irritability response data of multiple network community;

Group of suspicious network waterborne troops sample identification module: mark group of suspicious network waterborne troops sample according to the irritability response data of network community;

Network navy group irritability responds characteristic determination module: according to group of suspicious network waterborne troops sample and existing network navy group sample, determines that network navy group irritability responds feature;

Network navy group initiative recognition model building module: respond feature according to network navy group irritability, set up network navy group initiative recognition model;

Network navy group identification module: utilize network navy group initiative recognition Model Identification to go out network navy group in multiple network community.

As optimization, irritability data acquisition module comprises:

Intelligent interaction robot sets up unit: for setting up the behavior of simulation normal users, possessing the intelligent interaction robot of normal users all properties;

Make an initiative sally policy determining unit: for determining the strategy that makes an initiative sally of intelligent interaction robot;

Make an initiative sally unit: according to making an initiative sally, strategy carries out corresponding behavior;

Irritability response data collector unit: collect the response data of multiple network community as irritability response data.

As optimization, group of suspicious network waterborne troops sample identification module comprises:

Response time identify unit: according to irritability response data, marks earliest time and time the latest of all responses of each or subnetwork group;

Suspicious network group identify unit: if the earliest time of all responses of a network community and the latest time interval are within a certain period of time, be labeled as group of suspicious network waterborne troops.

As optimization, network navy group irritability is responded characteristic determination module and is comprised:

Respond tight ness rating computing unit: for calculating the response tight ness rating feature of each network community;

Response time proximity computation unit: for calculating the response time degree of approach feature of each network community;

Respond content similarity computing unit: for calculating the response content similarity feature of each network community;

Group's response degree computing unit: for calculating group's response degree feature of each network community.

As optimization, network navy group initiative recognition model building module comprises:

Network navy group irritability is responded model and is set up unit: first respond feature according to network navy group irritability and set up network navy group irritability response model, and then respond model according to network navy group irritability, set up network navy group initiative recognition model.

As optimization, network navy group irritability response model is set up unit and is comprised:

Irritability suspicious degree computation subunit: respond feature according to calculated network navy group irritability, calculate the suspicious degree of irritability of each network community.

As optimization, network navy group identification module comprises:

Irritability suspicious degree comparing unit: suspicious for the irritability of each network community degree and the suspicious degree threshold value pre-set are compared, the network community being greater than the suspicious degree correspondence of irritability of suspicious degree threshold value is network navy group; Remaining is normal group.

Technique scheme of the present invention has the following advantages compared to existing technology:

The invention provides the method for group of a kind of recognition network waterborne troops, carry out initiative recognition and go out a large amount of network navy groups hidden in network, can before network navy causes certain harm just can initiative recognition out, realize detection and the prevention of network navy.

In addition, network navy is also the discrete network user, and its attribute is very close with normal user with behavioural characteristic, identifies the very difficult of single network navy.But the real user handling network navy generally can handle multiple network navy account simultaneously, the attribute of this all or part of network navy just causing it to handle and behavioural characteristic have very high similarity, group properties is obvious, has better operability and accuracy to the identification of such network navy group compared with the identification of single network waterborne troops.First the present invention is marked by artificial identification expert, mark off multiple network community, ensure that in group, member is the network user that dubiety is higher as far as possible, its member comprises network user's account of the individual most likely network navy of 5-6, to ensure that network navy group irritability responds the accuracy of feature.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of the method for group of a kind of recognition network waterborne troops according to the embodiment of the present invention;

Fig. 2 is the process flow diagram of the method identifying group of waterborne troops of forum in accordance with another embodiment of the present invention;

Fig. 3 is the system architecture diagram of a kind of recognition network waterborne troops according to the embodiment of the present invention.

Embodiment

In order to make content of the present invention more easily be clearly understood, below according to a particular embodiment of the invention and by reference to the accompanying drawings, the present invention is further detailed explanation.

Embodiment 1

Network navy group is when being subject to external irritant, after the model that member of community as group of Forum network waterborne troops delivers is commented on, the real user of manipulation network navy group, waterborne troops's user account that can manipulate in network navy group is responded, group properties in its response data is comparatively outstanding, and it is comparatively easy to make its identification.So the present embodiment is on the basis of existing network waterborne troops recognition technology, utilize behavioural characteristic and/or the attributive character of the network navy group had been found that, analyze the network community irritability response data collected, group of suspicious network waterborne troops sample is gone out via handmarking, determine that network navy group irritability responds feature, set up network navy group initiative recognition model.

As shown in Figure 1, present embodiments provide the method for group of a kind of recognition network waterborne troops, comprise the following steps:

Step S1. obtains the irritability response data of multiple network community.Network community marks rear division by artificial identification expert, ensure that in group, member is the network user of the doubtful waterborne troops that possibility is higher as far as possible, group's scale can not be very large, its member comprises the network user of the individual most likely network navy of 5-6, and be the network navy with identical or similar behavioural characteristic and attributive character handled by one or more real user with identical object as far as possible, to ensure that network navy group irritability responds the accuracy of feature, later stage can utilize the mode of frequent-item to improve the precision of group's division in irritability response data.The irritability response data obtaining the plurality of network community comprises following process particularly:

First, based on different target platforms, as forum, Renren Network, microblogging etc., set up the robot account of corresponding simulation normal users behavior, Ye Ji intelligent interaction robot, it possesses all properties of normal users.User behavior pattern and the attributive character difference of different target platforms are larger, in order to all properties making intelligent interaction robot can possess normal users, need the intelligent interaction robot correspondingly adjusting foundation according to the difference of target platform.Intelligent interaction robot generates software by specific intelligent interaction robot and sets up, and according to oneself requirement, selects specific intelligent interaction robot, and is generated by intelligent interaction robot generation software.

Then, after determining the strategy that makes an initiative sally of intelligent interaction robot, the mutual robot of operative intelligence carries out the behavior that makes an initiative sally.According to a large amount of behavior data analysis network navy group behavior patterns having confirmed as network navy at present, determine the corresponding strategy that makes an initiative sally.The strategy that makes an initiative sally is the behavior pattern that makes an initiative sally comprising the time made an initiative sally, the content made an initiative sally, the mode that makes an initiative sally etc.Behavioural characteristic based on the network navy of different platform is different, different from the strategy that makes an initiative sally of its corresponding intelligent interaction robot.For forum, analyze the pattern of posting of waterborne troops of forum, as issued the temporal mode of model according to waterborne troops of forum, respectively the model of the member of community being divided into forum user group is commented on when it is delivered peak and delivers low ebb; Or deliver corresponding model content for the content that waterborne troops of multiple forum user delivers or reply the extreme property criticized content etc.

Finally, the irritability response data of above-mentioned all-network group is collected.After the network user during intelligent interaction robot is to network community makes an initiative sally behavior, the network user responds the behavior of making an initiative sally for intelligent interaction robot, produce one or more the response data comprising and respond account, response content, response time, response account hour of log-on and respond in account grade, this response data is the irritability response data of network community.

The irritability response data that step S2. collects according to step S1, mark group of suspicious network waterborne troops sample, detailed process comprises: mark network members in each or subnetwork group according to above-mentioned irritability response data and make an initiative sally the earliest time of all responses of behavior and time the latest to intelligent interaction robot.If the earliest time of all responses of a network community and the latest time interval are within a certain period of time, be labeled as group of suspicious network waterborne troops, as being then labeled as group of suspicious network waterborne troops in 2 weeks.Mark group of network consisting waterborne troops of group of multiple suspicious network waterborne troops sample.Also group of suspicious network waterborne troops sample can be obtained in conjunction with frequent-item technology, to improve the accuracy of group of suspicious network waterborne troops sample in this step.Store this network navy group sample, as next step input data.

Step S3., according to group of the suspicious network waterborne troops sample marked in step 2 and existing network navy group sample, determines that network navy group irritability responds feature.Particularly, analyze the irritability response data of group of above-mentioned suspicious network waterborne troops sample, in conjunction with attribute and the behavioural characteristic of network navy group well known in the prior art sample, determine that network navy group irritability responds feature.The network navy group irritability determined is responded feature and is comprised response tight ness rating, the response time degree of approach, responds content similarity and group's response degree.

Step S3 also comprises the response tight ness rating, the response time degree of approach, response content similarity and the group's response degree that calculate each network community, specifically comprises:

1. utilize formula (I) (II), the response tight ness rating of computational grid group:

$\mathrm{GRC}\left(g\right)={\max }_{r\&Element;R_g}\left(\mathrm{RC}(g,r)\right)---\left(I\right)$

$\mathrm{RC}(g,r)=\left\{\begin{array}{cc}0& \mathrm{ifL}(g,r)-F(g,r)>\mathrm{\&alpha;}\\ 1-\frac{L(g,r)-F(g,r)}{\mathrm{\&alpha;}}& \mathrm{otherwise}\end{array}\right.---\left(\mathrm{II}\right)$

Wherein, L (g, r), F (g, r) time the latest that the network user in network community g responds the intelligent interaction robot for it and earliest time is represented respectively, RC (g, r) be that the network user in network community g is to the response tight ness rating for its intelligent interaction robot, GRC (g) represents the response tight ness rating of network community g, formula (II) is first utilized to calculate each user in network community g to the response tight ness rating for its intelligent interaction robot, then formula (I) is utilized to calculate all-network user in network community g to the maximal value of the response tight ness rating for its intelligent interaction robot, as the response tight ness rating of network community g.It is respond the highest network navy group of tight ness rating to obtain that the response tight ness rating of network community g gets maximal value, and namely contact network navy group the most closely, its influence degree is larger.α is for responding tightness parameter, and be used for weighing the response tight ness rating of network community g, its value can be set to 30 days.

If a network community in multiple network community (a, b, c, d, e) to be made up of 5 network users, always have 5 intelligent interaction robots to initiate to make an initiative sally behavior to this network community, marking the time earliest time that network user a responds these 5 intelligent interaction robots is F _a, the time is L the latest _a, i.e. F (g, r)=F _a, L (g, r)=L _a, according to formula II, if L _a-F _a> α, so RC (g, r)=0 of network user a, otherwise, be 1-(L _a-F _a)/α; So analogize, obtain the response tight ness rating RC (g, r) of the network user b, c, d, e successively.Finally, according to formula I, comparing the tight angle value of response of these 5 network users, which is maximum, and maximum that responds tight angle value namely as the tight angle value of response of this network community.

2. utilize formula (III) (IV), calculate the response time degree of approach:

$\mathrm{GRTF}\left(g\right)={\max }_{r\&Element;R_g}\left(\mathrm{RTF}(g,r)\right)---\left(\mathrm{III}\right)$

$\mathrm{RTF}(g,r)=\left\{\begin{array}{cc}0& \mathrm{ifL}(g,r)-A\left(r\right)>\mathrm{\&beta;}\\ 1-\frac{L(g,r)-A(r)}{\mathrm{\&beta;}}& \mathrm{otherwise}\end{array}\right.---\left(\mathrm{IV}\right)$

Wherein, L (g, r), A (r) to represent in network community g time the latest that a network user responds the intelligent interaction robot for it respectively, the time of the behavior that makes an initiative sally for the intelligent interaction robot of this network community, RTF (g, r) be that the network user in network community g is to the response time degree of approach for its intelligent interaction robot, GRTF (g) represents the response time degree of approach of network community g, be in network community g all-network user to the maximal value of the response time degree of approach for its intelligent interaction robot.First utilize each network user in formula (IV) computational grid group g to the response time degree of approach for its intelligent interaction robot, then all-network user in network community g is to the maximal value of the response time degree of approach for its intelligent interaction robot, as the response time degree of approach of network community g to utilize formula (III) to draw.Getting maximal value is to obtain the minimum network navy group of response time difference, namely after intelligent interaction robot carries out the behavior that makes an initiative sally, network members in this network navy group has carried out response behavior to intelligent interaction robot within the shortest time, this network navy group is the network navy group enlivened the most, and its influence degree is larger.β is response time proximity parameters, is used for weighing the response time degree of approach of network community, and its value can be set to 20 days.

3. utilize formula (V) (VI), calculate and respond content similarity:

$\mathrm{GRCS}\left(g\right)={\max }_{r\&Element;R_g}\left(\mathrm{RCS}(g,r)\right)---\left(V\right)$

$\mathrm{RCS}(g,r)={\mathrm{avg}}_{m_i,m_j\&Element;g,i<j}\left(\cos (\mathrm{rc}(m_i,r),\mathrm{rc}(m_j,r))\right)---\left(\mathrm{VI}\right)$

Wherein, rc (m _i, r), rc (m _j, r) represent the network user m in network community g respectively _iand m _jto the content that the intelligent interaction robot for it responds, cos (rc (m _i, r), rc (m _j, r)) and be used for computational grid user m _iresponse content and m _jthe similarity of response content, RCS (g, r) be the average of similarity of response content of the every other network user in the response content of the network user in network community g and this network community, GRCS (g) represents the response content similarity of network community g, is that all-network user in network community g responds the maximal value of the average of content similarity to the intelligent interaction robot for it.Getting maximal value is respond the maximum network navy group of content similarities to obtain, and namely affect network navy group the most severe, its extent of injury is larger.

Now illustrate the response content similarity computation process of network community: establish in a network community and have 5 member (a, b, c, d, e), the all intelligent interaction robots for this network community are represented with F, the first step, first formula (VI) is utilized to calculate member a to the response content of F and member b to the similarity Xab of the response content of F, then the similarity Xac of the response content of member a and the response content of member c is calculated respectively, the similarity Xad of the response content of member a and the response content of member d, the similarity Xae of the response content of member a and the response content of member e, finally obtain Xab, Xac, Xad, the mean value of Xae, be response content and the member b of member a, member c, member d, the average Ja of the response content similarity of member e, in like manner, formula (VI) is utilized to calculate the average Jb of response content similarity of the response content of member b and member a, member c, member d, member e, the average Jc of the response content similarity of the response content of member c and member a, member b, member d, member e, the average Jd of the response content similarity of the response content of member d and member a, member b, member c, member e, the average Je of the response content similarity of the response content of member e and member a, member b, member c, member d, second step, utilizes the maximal value in formula (V) computation of mean values Ja, average Jb, average Jc, average Jd, Je, is the response content similarity of network community.

4. utilize formula (VII) (VIII), calculate group's response degree:

$\mathrm{GRE}\left(g\right)={\max }_{r\&Element;R_g}\left(\mathrm{RE}(g,r)\right)---\left(\mathrm{VII}\right)$

$\mathrm{RE}(g,r)=\frac{\left|R_g\right|}{g}---\left(\mathrm{VIII}\right)$

Wherein, | R _g|, g represents that member of community in network community g is to for the response number of its one of them intelligent interaction robot and the total number of persons of network community g respectively, RE (g, r) be that network community g is to the group's response degree for its one of them intelligent interaction robot, GRE (g) represents group's response degree of network community g, is the maximal value of network community g to group's response degree of all intelligent interaction robots for it.Getting maximal value is to obtain the maximum network navy group of group's response degree, and namely active network navy group, its extent of injury is also larger.

As the embodiment that can replace, respond tight ness rating, the response time degree of approach, respond content similarity, group's response degree and can calculate a kind of or several arbitrarily combination wherein, then carry out subsequent calculations according to it.

Step S4. responds feature according to network navy group irritability, sets up network navy group initiative recognition model.Detailed process comprises:

First, respond feature according to network navy irritability, set up network navy group irritability and respond model, namely respond feature according to the network navy group irritability calculated in step S3, utilize formula $\mathrm{GS}\left(g\right)={\&PartialD;}_1\mathrm{GRC}\left(g\right)+{\&PartialD;}_2\mathrm{GRTF}\left(g\right)+{\&PartialD;}_3\mathrm{GRCS}\left(g\right)+{\&PartialD;}_4\mathrm{GRE}\left(g\right)$ The suspicious degree of irritability of computational grid group, wherein, GS (g) represents the suspicious degree of the irritability of network community g, GRC (g), GRTF (g), GRCS (g), GRE (g) represent the response tight ness rating of network community, the response time degree of approach respectively, respond content similarity and group's response degree feature, as formula (I), (III), (V) (VII). for the undetermined coefficient factor, by the weight deciding group of heterogeneous networks waterborne troops irritability feature.Herein can get 0.3,0.3,0.3,0.1 respectively, so the irritability of network community suspicious degree GS (g) is greater than 0.5 and can thinks that this network community is network navy group.

In step s3, a kind of or several arbitrarily combination wherein only can be calculated when calculating response tight ness rating, the response time degree of approach, response content similarity, group's response are spent, correspondingly, the feature that can calculate according to these in this step carrys out the suspicious degree of irritability according to above-mentioned computation model computational grid group by the mode of rational weighted sum.

Then, set up network navy initiative recognition model, network navy group irritability is responded the input data of output data as network navy initiative recognition model of model.

Step S5. utilizes the network navy group initiative recognition Model Identification set up in step S4 to go out network navy group in multiple network community.Detailed process is: first respond according to network navy group irritability the suspicious degree of irritability that model calculates each network community.Then suspicious for the irritability of each network community degree and the suspicious degree threshold value pre-set are compared, the network community that the suspicious degree of irritability is greater than suspicious degree threshold value is defined as network navy group, and remaining is defined as normal group.The result that group of formula (Ⅹ) computational grid waterborne troops irritability response model specifically can be utilized to return, determine network navy group and normal group:

$\mathrm{GS}\left(g\right)=\left\{\begin{array}{cc}{N}_g& \mathrm{ifGS}\left(g\right)<\mathrm{\&tau;}\\ S_g& \mathrm{otherwise}\end{array}\right.---\left(X\right)$

Wherein, GS (g) represents the suspicious degree of the irritability of network community g, N _grepresent normal group, S _grepresent network navy group, τ is suspicious degree threshold value, is used for group of diffServ network waterborne troops and normal group.Like this, through said process, the group of all-network waterborne troops in the network community of division just can be identified.

Finally, store the network navy group identified, be convenient to later stage inquiry comparison.

The method of the group of recognition network waterborne troops utilizing the present embodiment to provide, can go out with regard to initiative recognition the network navy group hidden in network before network navy causes certain harm, not only achieve the initiative recognition of network navy, the harm that can also prevent network navy from causing in time.And, multiple doubtful network navy with identical or similar behavioural characteristic and attributive character is formed a network community, responds feature by the irritability in this network community and come group of recognition network waterborne troops, the accuracy of network navy identification can be improved.

Embodiment 2

As shown in Figure 2, present embodiments provide a kind of method of the group of recognition network waterborne troops based on forum, specifically comprise the following steps:

Start most, need artificial division to go out multiple forum user group.The forum user of some is divided into multiple forum user group comprising 5-6 member of community by artificial identification expert, ensure that the member of community in forum user group is the waterborne troops of forum that dubiety is higher as far as possible when dividing, and be by a real user or waterborne troops of the forum account manipulated by several real user with identical object as far as possible.Certainly can not ensure that in group, member is waterborne troops of forum, the later stage can utilize the mode of frequent-item to improve the precision of group's division in irritability response data.

Step 1. sets up intelligent interaction robot.Set up the intelligent interaction robot possessing normal forum user all properties, namely simulate the robot account of normal forum user behavior.

Step 2. determines that intelligent interaction robot makes an initiative sally strategy making an initiative sally.According to the behavior pattern of the waterborne troops of forum determined at present, determine the corresponding strategy that makes an initiative sally.Such as, according to the temporal mode of posting of waterborne troops of forum, delivering peak period and delivering the low ebb phase it is oppositely commented on.

Step 3. collects the irritability response data of forum user group.After intelligent interaction robot is commented on the model that the forum user in forum user group is delivered, forum user can respond comment, produce response data, comprise and respond account, response content, response time, response account hour of log-on and respond account grade etc.Collect the irritability response data of this type of response data as forum user group.

Step 4. marks group of waterborne troops of suspicious forum sample.According to the irritability response data of collecting in step 3, what mark that the forum user in part forum user group comments on all models responded to the active of intelligent interaction robot delivers earliest time and time the latest.If the forum user in certain forum user group the active of intelligent interaction robot is commented on all models responded deliver earliest time and the latest the time interval in 2 weeks, be labeled as group of waterborne troops of suspicious forum by this forum user group, so collect group of waterborne troops of multiple suspicious forum as group of waterborne troops of suspicious forum sample.

Step 5., according to group of waterborne troops of suspicious forum sample and group of waterborne troops of existing forum sample, determines that group of waterborne troops of forum irritability responds feature.Particularly, analyze the irritability response data of group of waterborne troops of above-mentioned suspicious forum sample, in conjunction with attribute and the behavioural characteristic of group of waterborne troops of forum well known in the prior art, determine that group of waterborne troops of forum irritability responds feature.Group of waterborne troops of forum irritability is responded feature and is comprised response tight ness rating, the response time degree of approach, responds content similarity and group's response degree.

Group of waterborne troops of the forum irritability that step 6. calculates each forum user group responds feature.Specifically comprise:

1. utilize formula (I) (II), calculate the response tight ness rating of forum user group:

$\mathrm{GRC}\left(g\right)={\max }_{r\&Element;R_g}\left(\mathrm{RC}(g,r)\right)---\left(I\right)$

$\mathrm{RC}(g,r)=\left\{\begin{array}{cc}0& \mathrm{ifL}(g,r)-F(g,r)>\mathrm{\&alpha;}\\ 1-\frac{L(g,r)-F(g,r)}{\mathrm{\&alpha;}}& \mathrm{otherwise}\end{array}\right.---\left(\mathrm{II}\right)$

Wherein, L (g, r), F (g, r) respectively represent forum user community g a forum user to for its intelligent interaction robot respond time the latest and earliest time, RC (g, r) be that a forum user in forum user group g is to the response tight ness rating for its intelligent interaction robot, GRC (g) represents the response tight ness rating of forum user community g, formula (II) is first utilized to calculate each forum user in forum user group g to the response tight ness rating for its intelligent interaction robot, then formula (I) is utilized to calculate all forum users in forum user group g to the maximal value of the response tight ness rating for its intelligent interaction robot, as the response tight ness rating of forum user group g.It is respond the highest group of waterborne troops of forum of tight ness rating to obtain that the response tight ness rating of forum user group g gets maximal value, and namely contact group of waterborne troops of forum the most closely, its influence degree is larger.α is for responding tightness parameter, and be used for weighing the response tight ness rating of net forum user group g, its value is set to 30 days.

2. utilize formula (III) (IV), calculate the response time degree of approach of forum user group:

$\mathrm{GRTF}\left(g\right)={\max }_{r\&Element;R_g}\left(\mathrm{RTF}(g,r)\right)---\left(\mathrm{III}\right)$

$\mathrm{RTF}(g,r)=\left\{\begin{array}{cc}0& \mathrm{ifL}(g,r)-A\left(r\right)>\mathrm{\&beta;}\\ 1-\frac{L(g,r)-A(r)}{\mathrm{\&beta;}}& \mathrm{otherwise}\end{array}\right.---\left(\mathrm{IV}\right)$

Wherein, L (g, r), A (r) represents the time the latest that a forum user in forum user community g is responded the intelligent interaction robot for it respectively, intelligent interaction robot for this forum user group carries out the time of initiatively commenting on behavior, RTF (g, r) be that a forum user in forum user group g is to the response time degree of approach for its intelligent interaction robot, GRTF (g) represents the response time degree of approach of forum user community g, be in forum user group g all forum users to the maximal value of the response time degree of approach for its intelligent interaction robot.First formula (IV) is utilized to calculate in forum user group g each forum user to the response time degree of approach for its intelligent interaction robot, then in forum user group g, all forum users are to the maximal value of the response time degree of approach for its intelligent interaction robot, as the response time degree of approach of forum user group g to utilize formula (III) to draw.Getting maximal value is to obtain the minimum group of waterborne troops of forum of response time difference, namely after intelligent interaction robot carries out active comment behavior, all forum members of this group of waterborne troops of forum have all carried out response behavior within the shortest time, this group of waterborne troops of forum is the group of waterborne troops of forum enlivened the most, and its influence degree is larger.β is response time proximity parameters, is used for weighing the response time degree of approach of forum user group, and its value is set to 20 days.

3. utilize formula (V) (VI), calculate the response content similarity of forum user group:

$\mathrm{GRCS}\left(g\right)={\max }_{r\&Element;R_g}\left(\mathrm{RCS}(g,r)\right)---\left(V\right)$

$\mathrm{RCS}(g,r)={\mathrm{avg}}_{m_i,m_j\&Element;g,i<j}\left(\cos (\mathrm{rc}(m_i,r),\mathrm{rc}(m_j,r))\right)---\left(\mathrm{VI}\right)$

Wherein, rc (m _i, r), rc (m _j, r) represent the member of community m in forum user community g respectively _iand m _jto the content that the intelligent interaction robot for it responds, cos (rc (m _i, r), rc (m _j, r)) be used for calculating member of community m _iresponse content and m _jthe similarity of response content, RCS (g, r) be the average of similarity of response content of every other forum user in the response content of a member of community in forum user group g and forum user group g, GRCS (g) represents the response content similarity of forum user community g, is that all forum users in forum user group g respond the maximal value of the average of content similarity to the intelligent interaction robot for it.Getting maximal value is respond the maximum group of waterborne troops of forum of content similarities to obtain, and namely affect group of waterborne troops of forum the most severe, its extent of injury is larger.

4. utilize formula (VII) (VIII), calculate group's response degree of forum user group:

$\mathrm{GRE}\left(g\right)={\max }_{r\&Element;R_g}\left(\mathrm{RE}(g,r)\right)---\left(\mathrm{VII}\right)$

$\mathrm{RE}(g,r)=\frac{\left|R_g\right|}{g}---\left(\mathrm{VIII}\right)$

Wherein, | R _g|, g represents that member of community in forum user community g is to for the response number of its one of them intelligent interaction robot and the total number of persons of forum user group g respectively, RE (g, r) be that forum user group g is to the group's response degree for its one of them intelligent interaction robot, GRE (g) represents group's response degree of forum user community g, is the maximal value of forum user group g to group's response degree of all intelligent interaction robots for it.Getting maximal value is to obtain the maximum group of waterborne troops of forum of group's response degree, and namely active group of waterborne troops of forum, its extent of injury is also larger.

Step 7. is set up group of waterborne troops of forum irritability and is responded model.Namely respond feature according to group of waterborne troops of the forum irritability calculated in step 7, utilize formula $\mathrm{GS}\left(g\right)={\&PartialD;}_1\mathrm{GRC}\left(g\right)+{\&PartialD;}_2\mathrm{GRTF}\left(g\right)+{\&PartialD;}_3\mathrm{GRCS}\left(g\right)+{\&PartialD;}_4\mathrm{GRE}\left(g\right)$ Calculate the suspicious degree of irritability of forum user group, wherein, GS (g) represents the suspicious degree of the irritability of forum user community g, GRC (g), GRTF (g), GRCS (g), GRE (g) represent the response tight ness rating of forum's user community, the response time degree of approach respectively, respond content similarity and group's response degree feature, as formula (I), (III), (V) (VII). get 0.3,0.3,0.3,0.1 respectively, so the irritability of forum user group suspicious degree GS (g) is greater than 0.5 and can thinks that this forum user group is group of waterborne troops of forum.

Step 8. sets up group of waterborne troops of forum initiative recognition model, identifies group of waterborne troops of forum.The result utilizing group of waterborne troops of formula (Ⅹ) calculating forum irritability response model to return, distinguishes group of waterborne troops of forum and normal group:

$\mathrm{GS}\left(g\right)=\left\{\begin{array}{cc}{N}_g& \mathrm{ifGS}\left(g\right)<\mathrm{\&tau;}\\ S_g& \mathrm{otherwise}\end{array}\right.---\left(X\right)$

Wherein, GS (g) represents the suspicious degree of the irritability of forum user community g, N _grepresent normal group, S _grepresent group of waterborne troops of forum, τ is suspicious degree threshold value, is used for distinguishing group of waterborne troops of forum and normal group.

The method of the group of recognition network waterborne troops utilizing the present embodiment to provide is to identify the waterborne troops in forum, can go out with regard to initiative recognition the group of waterborne troops wherein hidden before waterborne troops of forum causes certain harm, not only achieve the initiative recognition of waterborne troops of forum, the harm that waterborne troops of forum causes can also be prevented in time.And, multiple waterborne troops of doubtful forum with identical or similar behavioural characteristic and attributive character is formed a forum user group, respond feature by the irritability in this forum user group and identify group of waterborne troops of forum, the accuracy that waterborne troops of forum identifies can be improved.

Embodiment 3

Present embodiments provide the system of group of a kind of recognition network waterborne troops, comprising:

Irritability data acquisition module, for obtaining the irritability response data of multiple network community.Specifically comprise: intelligent interaction robot sets up unit, for setting up the behavior of simulation normal users, possessing the intelligent interaction robot of normal users all properties; Make an initiative sally policy determining unit, for determining the strategy that makes an initiative sally of intelligent interaction robot; Make an initiative sally unit, for carrying out corresponding behavior according to the strategy that makes an initiative sally; Irritability response data collector unit, for collecting the response data of multiple network community as irritability response data.

Group of suspicious network waterborne troops sample identification module, for marking group of suspicious network waterborne troops sample according to the irritability response data of network community.Specifically comprise: response time identify unit, for according to irritability response data, mark earliest time and time the latest of all responses of each or subnetwork group; Suspicious network group identify unit, if the earliest time of all responses of a network community and the latest time interval are within a certain period of time, is labeled as group of suspicious network waterborne troops.

Network navy group irritability responds characteristic determination module, for according to group of suspicious network waterborne troops sample and existing network navy group sample, determines that network navy group irritability responds feature.Specifically comprise: respond tight ness rating computing unit, for calculating the response tight ness rating feature of each network community; Response time proximity computation unit, for calculating the response time degree of approach feature of each network community; Respond content similarity computing unit, for calculating the response content similarity feature of each network community; Group's response degree computing unit, for calculating group's response degree feature of each network community.

Network navy group initiative recognition model building module, for responding feature according to network navy group irritability, sets up network navy group initiative recognition model.Comprising: network navy group irritability is responded model and is set up unit, sets up network navy group irritability response model for first responding feature according to network navy group irritability and then sets up network navy group initiative recognition model according to network navy group irritability response model.Network navy group irritability response model is set up in unit and is comprised again the suspicious degree computation subunit of irritability, for responding feature according to calculated network navy group irritability, calculates the suspicious degree of irritability of each network community.

Network navy group identification module, goes out network navy group in multiple network community for utilizing network navy group initiative recognition Model Identification.Comprising the suspicious degree comparing unit of irritability, for suspicious for the irritability of each network community degree and the suspicious degree threshold value pre-set are compared, the network community being greater than the suspicious degree correspondence of irritability of suspicious degree threshold value is network navy group, and remaining is normal group.

The system of the group of recognition network waterborne troops utilizing the present embodiment to provide, can go out with regard to initiative recognition the network navy group hidden in network before network navy causes certain harm, not only achieve the initiative recognition of network navy, the harm that can also prevent network navy from causing in time.And, multiple doubtful network navy with identical or similar behavioural characteristic and attributive character is formed a network community, responds feature by the irritability in this network community and come group of recognition network waterborne troops, the accuracy of network navy identification can be improved.

Obviously, above-described embodiment is only for clearly example being described, and the restriction not to embodiment.For those of ordinary skill in the field, can also make other changes in different forms on the basis of the above description.Here exhaustive without the need to also giving all embodiments.And thus the apparent change of extending out or variation be still among the protection domain of the invention.

Claims (20)

1. a method for group of recognition network waterborne troops, is characterized in that, comprises the following steps:

Obtain the irritability response data of multiple network community, described network community comprises multiple doubtful network user for network navy with same or similar behavioural characteristic and/or attributive character, and described irritability response data is that the network user in network community delivers to it response data that content carries out replying the laggard row response of comment and generation other network users;

According to described irritability response data, mark group of suspicious network waterborne troops sample;

According to group of described suspicious network waterborne troops sample and existing network navy group sample, determine that network navy group irritability responds feature;

Respond feature according to described network navy group irritability, set up network navy group initiative recognition model;

Described network navy group initiative recognition Model Identification is utilized to go out network navy group in described multiple network community.

2. the method for group of recognition network waterborne troops as claimed in claim 1, it is characterized in that, the process of the irritability response data of the multiple network community of described acquisition, comprising:

Set up intelligent interaction robot, described intelligent interaction robot generates software by intelligent interaction robot and sets up;

Determine that described intelligent interaction robot makes an initiative sally strategy, described in the strategy that makes an initiative sally formulate the corresponding behavior pattern that makes an initiative sally for the behavioral characteristic according to known network navy, comprise time, content, mode;

Carry out according to the described strategy that makes an initiative sally the behavior that makes an initiative sally, described in the behavior of making an initiative sally be that the mutual robot of operative intelligence comments on the content that described network community is delivered or initiatively delivers subject content targetedly;

Collect the response data of described multiple network community as irritability response data.

3. the method for recognition network waterborne troops as claimed in claim 2, it is characterized in that, the robot account of normal users behavior artificially simulated by described intelligent interaction machine, possesses all properties of normal users.

4. the method for the group of recognition network waterborne troops according to any one of claim 1-3, it is characterized in that, the irritability response data of described network community is the response data of the network user in described network community after described intelligent interaction robot carries out the behavior that makes an initiative sally, and comprises at least one responded account, response content, response time, response account hour of log-on and respond in account grade.

5. the method for group of recognition network waterborne troops as claimed in claim 1, is characterized in that, the described process marking group of suspicious network waterborne troops sample according to described irritability response data comprises:

According to described irritability response data, mark earliest time and time the latest of all responses of each or subnetwork group;

If the earliest time of all responses of a network community and the latest time interval are within a certain period of time, be labeled as group of suspicious network waterborne troops.

6. the method for group of recognition network waterborne troops as claimed in claim 1, is characterized in that, described network navy group irritability response feature comprises at least one in response tight ness rating, the response time degree of approach, response content similarity, group's response degree.

7. the method for group of recognition network waterborne troops as claimed in claim 6, is characterized in that,

Described response tight ness rating is calculated by following formula (I) (II):

$\mathrm{GRC}\left(g\right)={\max }_{r\&Element;R_g}\left(\mathrm{RC}(g,r)\right)---\left(I\right)$

$\mathrm{RC}(g,r)=\left\{\begin{array}{cc}0& \mathrm{ifL}(g,r)-F(g,r)>\mathrm{\&alpha;}\\ 1-\frac{L(g,r)-F(g,r)}{\mathrm{\&alpha;}}& \mathrm{otherwise}\end{array}\right.---\left(\mathrm{II}\right)$

Wherein, L (g, r), F (g, r) time the latest that the network user in network community g responds the intelligent interaction robot for it and earliest time is represented respectively, RC (g, r) be that the network user in network community g is to the response tight ness rating for its intelligent interaction robot, GRC (g) represents the response tight ness rating of network community g, that all-network user in network community g responds the maximal value of tight ness rating to the intelligent interaction robot for it, α, for responding tightness parameter, is used for weighing the response tight ness rating of network community g.

8. the method for group of recognition network waterborne troops as claimed in claim 6, is characterized in that,

The described response time degree of approach, is calculated by following formula (III) (IV):

$\mathrm{GRTF}\left(g\right)={\max }_{r\&Element;R_g}\left(\mathrm{RTF}(g,r)\right)---\left(\mathrm{III}\right)$

$\mathrm{RTF}(g,r)=\left\{\begin{array}{cc}0& \mathrm{ifL}(g,r)-A\left(r\right)>\mathrm{\&beta;}\\ 1-\frac{L(g,r)-A\left(r\right)}{\mathrm{\&beta;}}& \mathrm{otherwise}\end{array}\right.---\left(\mathrm{IV}\right)$

Wherein, L (g, r), A (r) represents the time the latest that a network user in network community g responds the intelligent interaction robot for it respectively, intelligent interaction robot for this network community carries out the time of the behavior that makes an initiative sally, RTF (g, r) be that the network user in network community g is to the response time degree of approach for its intelligent interaction robot, GRTF (g) represents the response time degree of approach of network community g, be in network community g all-network user to the maximal value of the response time degree of approach for its intelligent interaction robot, β is response time proximity parameters, be used for weighing the response time degree of approach of network community.

9. the method for group of recognition network waterborne troops as claimed in claim 6, is characterized in that,

Described response content similarity is calculated by following formula (V) (VI):

$\mathrm{GRCS}\left(g\right)={\max }_{r\&Element;R_g}\left(\mathrm{RCS}(g,r)\right)---\left(V\right)$

$\mathrm{RCS}(g,r)={\mathrm{avg}}_{m_i,m_j\&Element;g,i<j}\left(\cos (\mathrm{rc}(m_i,r),\mathrm{rc}(m_j,r))\right)---\left(\mathrm{VI}\right)$

Wherein, rc (m _i, r), rc (m _j, r) represent the network user m in network community g respectively _iand m _jto the content that the intelligent interaction robot for it responds, cos (rc (m _i, r), rc (m _j, r)) and be used for computational grid user m _iresponse content and m _jthe similarity of response content, RCS (g, r) be the average of the response content of a network user in network community g and the response content similarity of the every other network user, GRCS (g) represents the response content similarity of network community g, is that all-network user in network community g responds the maximal value of the average of content similarity to the intelligent interaction robot for it.

10. the method for group of recognition network waterborne troops as claimed in claim 6, is characterized in that,

Described group response degree, is calculated by following formula (VII) (VIII):

$\mathrm{GRE}\left(g\right)={\max }_{r\&Element;R_g}\left(\mathrm{RE}(g,r)\right)---\left(\mathrm{VII}\right)$

$\mathrm{RE}(g,r)=\frac{\left|R_g\right|}{g}---\left(\mathrm{VIII}\right)$

Wherein, | R _g|, g represents that member of community in network community g is to for the response number of its one of them intelligent interaction robot and the total number of persons of network community g respectively, RE (g, r) be that network community g is to the group's response degree for its one of them intelligent interaction robot, GRE (g) represents group's response degree of network community g, is the maximal value of network community g to group's response degree of all intelligent interaction robots for it.

The method of 11. groups of recognition network waterborne troops according to any one of claim 1-10, is characterized in that, describedly responds feature according to described network navy group irritability and sets up network navy group initiative recognition model and comprise:

Respond feature according to described network navy group irritability, set up network navy group irritability and respond model;

Respond model according to described network navy group irritability, set up network navy group initiative recognition model.

The method of 12. groups of a kind of recognition network waterborne troops according to any one of claim 1-11, is characterized in that, described network navy group irritability of setting up responds model, comprises following concrete steps:

Respond feature according to calculated network navy group irritability, utilize formula (Ⅸ), the suspicious degree of irritability of computational grid group:

$\mathrm{GS}\left(g\right)={\&PartialD;}_1\mathrm{GRC}\left(g\right)+{\&PartialD;}_2\mathrm{GRTF}\left(g\right)+{\&PartialD;}_3\mathrm{GRCS}\left(g\right)+{\&PartialD;}_4\mathrm{GRE}\left(g\right)---\left(\mathrm{IX}\right)$

Wherein, GS (g) represents the suspicious degree of the irritability of network community g, GRC (g), GRTF (g), GRCS (g), GRE (g) represent the response tight ness rating of network community, the response time degree of approach respectively, respond content similarity and group's response degree feature for the undetermined coefficient factor, by the weight deciding group of heterogeneous networks waterborne troops irritability feature.

The method of 13. groups of recognition network waterborne troops according to any one of claim 1-12, is characterized in that, the described described network navy group initiative recognition Model Identification that utilizes goes out network navy group in described multiple network community:

The suspicious degree of irritability that model calculates each network community in described multiple network community is responded according to described network navy group irritability;

Suspicious for the irritability of each network community degree and the suspicious degree threshold value pre-set are compared, the network community being greater than the suspicious degree correspondence of irritability of described suspicious degree threshold value is network navy group; Remaining is normal group.

The system of 14. 1 kinds of groups of recognition network waterborne troops, is characterized in that, comprising:

Irritability data acquisition module: for obtaining the irritability response data of multiple network community;

Group of suspicious network waterborne troops sample identification module: mark group of suspicious network waterborne troops sample according to the irritability response data of described network community;

Network navy group irritability responds characteristic determination module: according to group of described suspicious network waterborne troops sample and existing network navy group sample, determines that network navy group irritability responds feature;

Network navy group initiative recognition model building module: respond feature according to described network navy group irritability, set up network navy group initiative recognition model;

Network navy group identification module: utilize described network navy group initiative recognition Model Identification to go out network navy group in described multiple network community.

The system of 15. groups of recognition network waterborne troops as claimed in claim 14, is characterized in that, described irritability data acquisition module comprises:

Intelligent interaction robot sets up unit: for setting up the behavior of simulation normal users, possessing the intelligent interaction robot of normal users all properties;

Make an initiative sally policy determining unit: for determining the strategy that makes an initiative sally of described intelligent interaction robot;

Make an initiative sally unit: carry out corresponding behavior according to the described strategy that makes an initiative sally;

Irritability response data collector unit: collect the response data of described multiple network community as irritability response data.

The system of 16. groups of recognition network waterborne troops as claimed in claim 14, is characterized in that, group of described suspicious network waterborne troops sample identification module comprises:

Response time identify unit: according to described irritability response data, marks earliest time and time the latest of all responses of each or subnetwork group;

Suspicious network group identify unit: if the earliest time of all responses of a network community and the latest time interval are within a certain period of time, be labeled as group of suspicious network waterborne troops.

The system of 17. groups of recognition network waterborne troops as claimed in claim 14, is characterized in that, described network navy group irritability is responded characteristic determination module and comprised:

Respond tight ness rating computing unit: for calculating the response tight ness rating feature of each network community;

Response time proximity computation unit: for calculating the response time degree of approach feature of each network community;

Respond content similarity computing unit: for calculating the response content similarity feature of each network community;

Group's response degree computing unit: for calculating group's response degree feature of each network community.

The system of 18. groups of recognition network waterborne troops as claimed in claim 14, is characterized in that, described network navy group initiative recognition model building module comprises:

Network navy group irritability is responded model and is set up unit: first respond feature according to described network navy group irritability and set up network navy group irritability response model, and then respond model according to described network navy group irritability, set up network navy group initiative recognition model.

The system of 19. groups of recognition network waterborne troops as claimed in claim 18, is characterized in that, described network navy group irritability response model is set up unit and comprised:

Irritability suspicious degree computation subunit: respond feature according to calculated network navy group irritability, calculate the suspicious degree of irritability of each network community.

The system of 20. groups of recognition network waterborne troops as claimed in claim 14, is characterized in that, described network navy group identification module comprises:

Irritability suspicious degree comparing unit: suspicious for the irritability of each network community degree and the suspicious degree threshold value pre-set are compared, the network community being greater than the suspicious degree correspondence of irritability of suspicious degree threshold value is network navy group; Remaining is normal group.