Movatterモバイル変換


[0]ホーム

URL:


CN103617235A - Method and system for network navy account number identification based on particle swarm optimization - Google Patents

Method and system for network navy account number identification based on particle swarm optimization
Download PDF

Info

Publication number
CN103617235A
CN103617235ACN201310611396.9ACN201310611396ACN103617235ACN 103617235 ACN103617235 ACN 103617235ACN 201310611396 ACN201310611396 ACN 201310611396ACN 103617235 ACN103617235 ACN 103617235A
Authority
CN
China
Prior art keywords
user
particle
waterborne troops
vector
module
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201310611396.9A
Other languages
Chinese (zh)
Other versions
CN103617235B (en
Inventor
黄超
牛温佳
管洋洋
李倩
刘萍
郭莉
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Institute of Information Engineering of CAS
Original Assignee
Institute of Information Engineering of CAS
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Institute of Information Engineering of CASfiledCriticalInstitute of Information Engineering of CAS
Priority to CN201310611396.9ApriorityCriticalpatent/CN103617235B/en
Publication of CN103617235ApublicationCriticalpatent/CN103617235A/en
Application grantedgrantedCritical
Publication of CN103617235BpublicationCriticalpatent/CN103617235B/en
Expired - Fee Relatedlegal-statusCriticalCurrent
Anticipated expirationlegal-statusCritical

Links

Images

Classifications

Landscapes

Abstract

Translated fromChinese

本发明涉及一种基于粒子群算法的网络水军账号识别方法,具体包括以下步骤:步骤1:收集β个用户的用户信息,从所述每个用户信息中提取λ个相关指标,得到β个指标向量;步骤2:对λ个指标根据需要分配权重,得到权重向量;步骤3:将β个指标向量按照其中每个指标大小进行排序,得到λ个序列;步骤4:选取每个用户为粒子,应用粒子群算法,将符合条件的粒子保存到水军用户列表;步骤5:选取水军用户列表中所有可疑水军用户,将符合条件的所有邻居加入水军用户列表。本发明在实时性方面,本发明提出的基于改进粒子群算法的网络水军账号识别方法更为适宜。

The present invention relates to a particle swarm algorithm-based identification method for online navy accounts, which specifically includes the following steps: Step 1: Collect user information of β users, extract λ related indicators from each user information, and obtain β Indicator vector; Step 2: Assign weights to λ indicators according to needs to obtain weight vectors; Step 3: Sort β indicator vectors according to the size of each indicator to obtain λ sequences; Step 4: Select each user as a particle , apply the particle swarm optimization algorithm, and save the eligible particles to the water army user list; Step 5: Select all suspicious water army users in the water army user list, and add all eligible neighbors to the water army user list. In terms of real-time performance of the present invention, the method for identifying network navy account numbers based on the improved particle swarm algorithm proposed by the present invention is more suitable.

Description

A kind of account recognition methods of network waterborne troops and system based on particle cluster algorithm
Technical field
The present invention relates to a kind of account recognition methods of network waterborne troops and system based on particle cluster algorithm.
Background technology
On network, information content is growing brings very large challenge to network supervision, a large amount of network pushing hands are full of in internet environment, from " Jia Junpeng " to " Miss lotus ", arrive again " superfine product is month in and month out little ", these influential network events occur time and again, have shown its black strength---network waterborne troops behind.Network waterborne troops utilizes network environment to manufacture public topic, thereby guides public opinion guiding, and the influence power of some unique individual, unit is risen violently, and therefrom seeks unlawful interests.The operator scheme of network waterborne troops is divided into two kinds, and a kind of is to make a show of power, and a kind of is advertisement.Wherein, making a show of power is to deliver particular idea for particular topic, take to make a show of power that many spread pattern is comparatively neat between 20 to 30 words as number of words is pasted by the waterborne troops of object, and viewpoint is more consistent, and content repeatability is high especially; And take waterborne troops that advertisement is object, to paste feature more obvious, wherein can significantly comprise the information such as gray product, phone or hyperlink, and the advertisement waterborne troops position of pasting is many in paste on top, occurs, especially sofa pastes position and the most often occurs.From impact, if enlivened in forum, exist the waterborne troops that makes a show of power in a large number to paste, bring wrong spin can to numerous netizens, especially, when water subsides content has instigating or poisons and bewitches character, consequence will be very severe; And discriminating is pasted comparatively intuitively easily by advertisement waterborne troops, although pasting, mass advertising water also can cause the disorderly and unsystematic of network environment, but can't cause harm significantly to public opinion, therefore, the waterborne troops's account recognition methods in this patent is mainly for the identification of the waterborne troops's account of making a show of power.
At semantic analysis and artificial intelligence field, network waterborne troops identification problem was had to correlative study.The recognition methods of the more typical waterborne troops in semantic analysis field is machine learning method: the sample note complete or collected works of forum are divided into a plurality of subsets, then utilize neural network respectively antithetical phrase training get a plurality of sorters, finally by this classifiers, each model in forum is classified, for waterborne troops, paste and delete note or shielding processing.Yet in this process, the selection of training set has very large uncertainty, in training set, the accuracy of judgement degree that proportion badly influences sorter pastes in waterborne troops; And waterborne troops is attached in different time sections for different topic objects, in order to identify the waterborne troops about certain sensitive subjects in certain time period, paste, need to carry out again the training of sorter for this specific topics, visible this mode workload is very large; In addition, the model that this mode need to will be delivered each is examined classification, and the user who has had a strong impact on forum experiences.Large for semantic analysis mode workload, to affect user's experience problem.
Summary of the invention
Technical matters to be solved by this invention is, for the deficiencies in the prior art, provide a kind of and guaranteeing that the user of network waterborne troops differentiates under the prerequisite of accuracy, improved largely the network waterborne troops account recognition methods based on particle cluster algorithm that user that forum user posts experiences.
The technical scheme that the present invention solves the problems of the technologies described above is as follows: a kind of network waterborne troops account recognition methods based on particle cluster algorithm, specifically comprises the following steps:
Step 1: collect β user's user profile, extract λ index of correlation from described each user profile, obtain β indicator vector, each indicator vector comprises λ index, and described β and λ are greater than 1 natural number;
Step 2: the index of the λ in each indicator vector is assigned weight as required, obtain the weight vectors of each index;
Step 3: β indicator vector sorted according to the size of each index in same dimension, obtain λ sequence, wherein adjacent indicator vector neighbours each other;
Step 4: choose each user for particle, choosing user's indicator vector is particle position information, the dot product of weight selection vector and user's indicator vector is as the good and bad measurement index in position, application particle cluster algorithm, the particle that wherein user's weight vectors and the dot product of user's indicator vector is less than to setting value is saved in waterborne troops's user list;
Step 5: choose all suspicious users of waterborne troops in waterborne troops's user list, search for respectively the described suspicious user's of waterborne troops neighbours from each dimension, qualified all neighbours are added to waterborne troops's user list.
The invention has the beneficial effects as follows: the thought of particle cluster algorithm has been used for reference in the network waterborne troops account recognition methods that the present invention proposes, do not need to evaluate one by one each forum user or evaluate one by one each user and post, just can rapid screening go out the waterborne troops's account in forum; Only need to periodically to forum user, carry out sifting sort does not need each model to process in real time, so aspect real-time, the network waterborne troops account recognition methods based on improving particle cluster algorithm that this patent proposes is more suitable.
On the basis of technique scheme, the present invention can also do following improvement.
Further, the particle cluster algorithm in describedstep 4 specifically comprises the following steps:
Step 4.1: choose each user for particle, choosing user's indicator vector is particle position vector, and the dot product of weight selection vector and user's indicator vector is as the good and bad measurement index in position;
Step 4.2: calculate the velocity vector of each particle, successively each dimension of each dimension of current location vector and velocity vector is added, obtain dimension values;
Step 4.3: the sequence obtaining according tostep 3, moves to a nearest position by each particle;
Step 4.4: iteration execution step 4.3, until whether the number of times that particle moves reaches pre-determined number or three particles converge to same position;
Step 4.5: meet pre-conditioned particle and be saved in waterborne troops's user list all in iterative process.
Further, describedstep 5 comprises the following steps:
Step 5.1: choose a suspicious user of waterborne troops in waterborne troops's user list, the sequence obtaining according tostep 3 is searched this neighbours of the suspicious user of waterborne troops in each dimension;
Step 5.2: judge whether all neighbours' weight vectors and the dot product of user's indicator vector are less than setting value, and if so, the neighbours that all dot products are less than to setting value are saved in waterborne troops's user list, and carry out next step; Otherwise, jump to step 5.4;
Step 5.3: using the neighbours in waterborne troops's user list as a suspicious user of waterborne troops, jump to step 5.1;
Step 5.4: judge whether to have the suspicious user of waterborne troops who does not search neighbours, if had, jump to step 5.1; Otherwise, finish.
Further, position more new formula be:
xi(t+1)=xi(t)+vi(t+1) (1)
Speed more new formula is:
vi(t+1)=ω ⅹ vi(t)+c1 ⅹ rand() ⅹ (pi(t)-xi(t)+c2 ⅹ rand() ⅹ(gi(t)-xi(t)) (2)
Wherein, i particle position is expressed as vector xi; The position rate of i particle is vector vi; ω is inertia weight; piand gibe respectively historical optimal location and the global history optimal location of this particle; C1 and c2, for the study factor, have represented each particle have been pulled to piand githe weight of the random acceleration term of position; vmaxand xmaxrepresentation speed limits and position limitation respectively.
Further, in described formula (2), c1 and c2 are preferably set as respectively 1.4 and 0.6;
Inertia weight ω is preferably the random number between [0,1].
Particle cluster algorithm is a kind of search procedure based on population, and wherein each individuality is called particle, is defined as the potential solution of problem to be optimized in M dimension search volume, the memory of preserving the optimal location of its historical optimal location, present speed and all particles.Every evolution generation, the component that the information of particle is combined and regulates the speed about on every one dimension, is used to calculate new particle position then.Particle constantly changes their state in multi-dimensional search space, until arrive equilibrium state, or till having surpassed calculating restriction.
If search volume is a dimension, i particle position is expressed as vector xi=(xi1, xi2 ..., xia); The historical optimal location of i particle is pi=(pi1,pi2 ..., pia); , p whereinjfor all pi(i=1,2 ..., the historical optimal location in b); The position rate of i particle is vector vi=(vi1,vi2 ..., vib).Every evolution generation, the position of each particle changes according to current community information, and position more new formula is:
xi(t+1)=xi(t)+vi(t+1)
Speed more new formula is:
vi(t+1)=ω ⅹ vi(t)+c1 ⅹ rand() ⅹ (pi(t)-xi(t)+c2 ⅹ rand() ⅹ(gi(t)-xi(t))
Wherein ω is inertia weight; piand gibe respectively historical optimal location and the global history optimal location of this particle; C1 and c2, for the study factor, have represented each particle have been pulled to piand githe weight of the random acceleration term of position; vmaxand xmaxrepresentation speed limits and position limitation respectively.
The present invention chooses user characteristics description amount as particle position information, chooses the weight vector of evaluation index and the dot product of user characteristics description vectors as the good and bad measurement index in position.For example, when system selects particle that ID is 01 as primary, the current location that this particle is corresponding is [0.20.50.30.10.10.5], and the index of weighing this position quality is the weight vector of evaluation index and the dot product of user characteristics value vector, i.e. [0.20.50.30.10.10.5] [5,4,3,3,2,1]=4.9.It should be noted that, the user of waterborne troops should have lower custom Measure Indexes, lower social rangeability figureofmerit, lower normalization online hours, lower login rule index, lower money order receipt to be signed and returned to the sender frequency Measure Indexes and the lower user name length measuring index browsed, therefore, when judgement is good and bad, dot product is lower, the feature that more meets the user of waterborne troops, namely in particle cluster algorithm, position is more excellent.Obtain by statistics, the good and bad measurement index in the user's of waterborne troops position has 95% probability to be less than 1.2, so in this patent, the good and bad measurement index in position is less than to 1.2 particle and regards the suspicious user of waterborne troops as.
Technical matters to be solved by this invention is, for the deficiencies in the prior art, provide a kind of and guaranteeing that the user of network waterborne troops differentiates under the prerequisite of accuracy, improved largely the network waterborne troops account recognition methods based on particle cluster algorithm that user that forum user posts experiences.
The technical scheme that the present invention solves the problems of the technologies described above is as follows: a kind of network waterborne troops account recognition system based on particle cluster algorithm, comprising: index extraction module, weight allocation module, order module, particle cluster algorithm module and neighbor seaching module;
Described index extraction module, for collecting β user's user profile, is extracted λ index of correlation from described each user profile, obtains β indicator vector, and each indicator vector comprises λ index, and described β and λ are greater than 1 natural number;
Described weight allocation module, for the λ of each an indicator vector index is assigned weight as required, obtains the weight vectors of each index;
Described order module for β indicator vector sorted according to the size of same each index of dimension, obtains λ sequence, wherein adjacent indicator vector neighbours each other;
Described particle cluster algorithm module is used for choosing each user for particle, choosing user's indicator vector is particle position information, the dot product of weight selection vector and user's indicator vector is as the good and bad measurement index in position, application particle cluster algorithm, the particle that wherein user's weight vectors and the dot product of user's indicator vector is less than to setting value is saved in waterborne troops's user list;
Described neighbor seaching module is used for choosing all suspicious users of waterborne troops of waterborne troops's user list, searches for respectively the described suspicious user's of waterborne troops neighbours from each dimension, and qualified all neighbours are added to waterborne troops's user list.
The invention has the beneficial effects as follows: the thought of particle cluster algorithm has been used for reference in the network waterborne troops account recognition methods that the present invention proposes, do not need to evaluate one by one each forum user or evaluate one by one each user and post, just can rapid screening go out the waterborne troops's account in forum; Only need to periodically to forum user, carry out sifting sort does not need each model to process in real time, so aspect real-time, the network waterborne troops account recognition methods based on improving particle cluster algorithm that this patent proposes is more suitable.
On the basis of technique scheme, the present invention can also do following improvement.
Further, described n and m are greater than 1 natural number.
Further, described particle cluster algorithm module comprises that particle is chosen module, dimension values computing module, particle mobile module, iteration module and module is preserved in list;
It is particle position vector for choosing each user for particle, choosing user's indicator vector that described particle is chosen module, and the dot product of weight selection vector and user's indicator vector is as the good and bad measurement index in position;
Described dimension values computing module, for calculating the velocity vector of each particle, is added each dimension of each dimension of current location vector and velocity vector successively, obtains dimension values;
The sequence that described particle mobile module obtains according to order module, moves to a nearest position by each particle;
Described iteration module is carried out the movement of particle mobile module for iteration, until whether the number of times that particle moves reaches pre-determined number or three particles converge to same position;
Described list is preserved module for meeting pre-conditioned particle and be saved in waterborne troops's user list iterative process being all.
Further, described neighbor seaching module comprises that neighbours search module and neighbours add module;
Described neighbours search module for choosing all suspicious user of waterborne troops of waterborne troops's user list, and the sequence obtaining according to order module is searched this neighbours of the suspicious user of waterborne troops in each dimension;
Described neighbours add module and are saved in waterborne troops's user list for the dot product of all weight vectors and user's indicator vector being less than to the neighbours of setting value.
Further, position more new formula be:
xi(t+1)=xi(t)+vi(t+1) (1)
Speed more new formula is:
vi(t+1)=ω ⅹ vi(t)+c1 ⅹ rand() ⅹ (pi(t)-xi(t)+c2 ⅹ rand() ⅹ(gi(t)-xi(t)) (2)
Wherein, i particle position is expressed as vector xi; The position rate of i particle is vector vi; ω is inertia weight; piand gibe respectively historical optimal location and the global history optimal location of this particle; C1 and c2, for the study factor, have represented each particle have been pulled to piand githe weight of the random acceleration term of position; vmaxand xmaxrepresentation speed limits and position limitation respectively.
Further, in described formula (2), c1 and c2 are preferably set as respectively 1.4 and 0.6;
Inertia weight ω is preferably the random number between [0,1].
Accompanying drawing explanation
Fig. 1 is a kind of network waterborne troops account recognition methods process flow diagram based on particle cluster algorithm described in the specific embodiment of theinvention 1;
Fig. 2 is the process flow diagram of particle cluster algorithm in the method described in the specific embodiment of theinvention 1;
Fig. 3 is a kind of network waterborne troops account recognition system block diagram based on particle cluster algorithm described in the specific embodiment of theinvention 2;
Fig. 4 is that the system described in the specific embodiment of theinvention 3 is carried out the process flow diagram of waterborne troops's account identification to registered user in forum;
Fig. 5 is the process flow diagram that the specific embodiment of theinvention 3 application particle cluster algorithms are found the user of waterborne troops.
In accompanying drawing, the list of parts of each label representative is as follows:
1, index extraction module, 2, weight allocation module, 3, order module, 4, particle cluster algorithm module, 5, neighbor seaching module, 41, particle is chosen module, and 42, dimension values computing module, 43, particle mobile module, 44, iteration module, 45, module is preserved in list, and 51, neighbours search module, 52, neighbours add module.
Embodiment
Below in conjunction with accompanying drawing, principle of the present invention and feature are described, example, only for explaining the present invention, is not intended to limit scope of the present invention.
As shown in Figure 1, a kind of network waterborne troops account recognition methods based on particle cluster algorithm for described in the specific embodiment of theinvention 1, specifically comprises the following steps:
Step 1: collect β user's user profile, extract λ index of correlation from described each user profile, obtain β indicator vector, each indicator vector comprises λ index, and described β and λ are greater than 1 natural number;
Step 2: the index of the λ in each indicator vector is assigned weight as required, obtain the weight vectors of each index;
Step 3: β indicator vector sorted according to the size of each index in same dimension, obtain λ sequence, wherein adjacent indicator vector neighbours each other;
Step 4: choose each user for particle, choosing user's indicator vector is particle position information, the dot product of weight selection vector and user's indicator vector is as the good and bad measurement index in position, application particle cluster algorithm, the particle that wherein user's weight vectors and the dot product of user's indicator vector is less than to setting value is saved in waterborne troops's user list;
Step 5: choose a suspicious user of waterborne troops in waterborne troops's user list, the sequence obtaining according tostep 3 is searched this neighbours of the suspicious user of waterborne troops in each dimension;
Step 6: judge whether all neighbours' weight vectors and the dot product of user's indicator vector are less than setting value, and if so, the neighbours that all dot products are less than to setting value are saved in waterborne troops's user list, and carry out next step; Otherwise, jump to step 8;
Step 7: using the neighbours in waterborne troops's user list as a suspicious user of waterborne troops, jump to step 5;
Step 8: judge whether to have the suspicious user of waterborne troops who does not search neighbours, if had, jump to step 5; Otherwise, finish.
Described λ and β are greater than 1 natural number.
As shown in Figure 2, the particle cluster algorithm in describedstep 4 specifically comprises the following steps:
Step 4.1: choose each user for particle, choosing user's indicator vector is particle position vector, and the dot product of weight selection vector and user's indicator vector is as the good and bad measurement index in position;
Step 4.2: calculate the velocity vector of each particle, successively each dimension of each dimension of current location vector and velocity vector is added, obtain dimension values;
Step 4.3: the sequence obtaining according tostep 3, moves to a nearest position by each particle;
Step 4.4: iteration execution step 4.3, until whether the number of times that particle moves reaches pre-determined number or three particles converge to same position;
Step 4.5: meet pre-conditioned particle and be saved in waterborne troops's user list all in iterative process.
Position more new formula is:
xi(t+1)=xi(t)+vi(t+1) (1)
Speed more new formula is:
vi(t+1)=ω ⅹ vi(t)+c1 ⅹ rand() ⅹ (pi(t)-xi(t)+c2 ⅹ rand() ⅹ(gi(t)-xi(t)) (2)
Wherein, i particle position is expressed as vector xi; The position rate of i particle is vector vi; ω is inertia weight; piand gibe respectively historical optimal location and the global history optimal location of this particle; C1 and c2, for the study factor, have represented each particle have been pulled to piand githe weight of the random acceleration term of position; vmaxand xmaxrepresentation speed limits and position limitation respectively.
In described formula (2), c1 and c2 are preferably set as respectively 1.4 and 0.6;
Inertia weight ω is preferably the random number between [0,1].
As shown in Figure 3, a kind of network waterborne troops account recognition system based on particle cluster algorithm for described in the specific embodiment of theinvention 2, comprising:index extraction module 1,weight allocation module 2,order module 3, particlecluster algorithm module 4 andneighbor seaching module 5;
Describedindex extraction module 1, for collecting β user's user profile, is extracted λ index of correlation from described each user profile, obtains β indicator vector, and each indicator vector comprises λ index, and described β and λ are greater than 1 natural number;
Describedweight allocation module 2, for the λ of each an indicator vector index is assigned weight as required, obtains the weight vectors of each index;
Describedorder module 3, for β indicator vector sorted according to the size of same each index of dimension, obtains λ sequence, wherein adjacent indicator vector neighbours each other;
Described particlecluster algorithm module 4 is for choosing each user for particle, choosing user's indicator vector is particle position information, the dot product of weight selection vector and user's indicator vector is as the good and bad measurement index in position, application particle cluster algorithm, the particle that wherein user's weight vectors and the dot product of user's indicator vector is less than to setting value is saved in waterborne troops's user list;
Describedneighbor seaching module 5, for choosing all suspicious users of waterborne troops of waterborne troops's user list, is searched for respectively the described suspicious user's of waterborne troops neighbours from each dimension, qualified all neighbours are added to waterborne troops's user list.
Described λ and β are greater than 1 natural number.
Described particlecluster algorithm module 4 comprises that particle is chosenmodule 41, dimensionvalues computing module 42, particlemobile module 43,iteration module 44 andmodule 45 is preserved in list;
It is particle position vector for choosing each user for particle, choosing user's indicator vector that described particle is chosenmodule 41, and the dot product of weight selection vector and user's indicator vector is as the good and bad measurement index in position;
Described dimensionvalues computing module 42, for calculating the velocity vector of each particle, is added each dimension of each dimension of current location vector and velocity vector successively, obtains dimension values;
The sequence that described particlemobile module 43 obtains according to order module, moves to a nearest position by each particle;
Describediteration module 44 is carried out the movement of particle mobile module for iteration, until whether the number of times that particle moves reaches pre-determined number or three particles converge to same position;
Described list is preservedmodule 45 for meeting pre-conditioned particle and be saved in waterborne troops's user list iterative process being all.
Describedneighbor seaching module 5 comprises that neighbours searchmodule 51 and neighbours addmodule 52;
Described neighbours searchmodule 51 for choosing all suspicious user of waterborne troops of waterborne troops's user list, and the sequence obtaining according to order module is searched this neighbours of the suspicious user of waterborne troops in each dimension;
Described neighbours addmodule 52 and are saved in waterborne troops's user list for the dot product of all weight vectors and user's indicator vector being less than to the neighbours of setting value.
Position more new formula is:
xi(t+1)=xi(t)+vi(t+1) (1)
Speed more new formula is:
vi(t+1)=ω ⅹ vi(t)+c1 ⅹ rand() ⅹ (pi(t)-xi(t)+c2 ⅹ rand() ⅹ(gi(t)-xi(t)) (2)
Wherein, i particle position is expressed as vector xi; The position rate of i particle is vector vi; ω is inertia weight; piand gibe respectively historical optimal location and the global history optimal location of this particle; C1 and c2, for the study factor, have represented each particle have been pulled to piand githe weight of the random acceleration term of position; vmaxand xmaxrepresentation speed limits and position limitation respectively.
In described formula (2), c1 and c2 are preferably set as respectively 1.4 and 0.6;
Inertia weight ω is preferably the random number between [0,1].
Below specific embodiments of the invention 3:
The research of sociology ASSOCIATE STATISTICS shows, the network waterborne troops in social forum generally exists following feature: 1 user account is short; 2 online hours are short; 3 is low with other users' degree of contact; 4 times of posting concentrated and are regular; 5 is mutual hardly with other users.This patent is weighed from above several aspects each user, by the mode quantizing, represent user's matching degree in all fields, and the result after quantizing is formed to a vector characterize this user's feature, and then by particle cluster algorithm, in numerous user vectors, find and meet the vector of specific criteria as the recognition result of network waterborne troops account.The flow process that system is carried out waterborne troops's account identification to registered user in forum as shown in Figure 4.
Waterborne troops's account proper vector representational framework
In network forum, a user includes very many-sided information, such as user name, password, hour of log-on, login times, nearest login time, browsing history, the record of posting, money order receipt to be signed and returned to the sender record, circle of friends, browsing history, collection model record, login IP record etc.And there are notable difference in normal netizen and waterborne troops in some information dimensions, such as, normal netizen's user name generally has specific meanings, so user name length is generally between 4-12 English character, and waterborne troops is for simple to operate, often can use the user name of one or two English character; The login time of normal netizen in forum is random, and waterborne troops is generally at special time period the online time, and during making a show of power about certain topic, waterborne troops can frequently login forum and post, after the task of making a show of power finishes, seldom login forum; Normal netizen browses login forum of model ,Er waterborne troops just in order to issue water, to paste, so they are used for the most of the time posting, can browse other models hardly; Normal netizen can with regard to certain problem, to beam back note mutual with other netizens in forum, and network waterborne troops is only responsible for the model to particular topic issue certain content, money order receipt to be signed and returned to the sender is explained not within their working range, so this type of water pastes seldom to replying the capable processing of exchange premium; Normal netizen has corresponding circle of friends in the constructed environment of forum, forms a small-sized social networks ,Er network waterborne troops and only take and post as object, can have social circle hardly.This patent utilizes normal users and waterborne troops in the difference of above-mentioned dimension, a kind of user account representational framework based on proper vector has been proposed, each user is carried out to quantization means in above-mentioned dimension, and then whether waterborne troops judges to each user in conjunction with particle cluster algorithm.
User name length evaluation index
According to relevant The Study of Sociology, show, in general social forum, normal netizen's the user people have specific meanings, its character quantity generally 4-12 not etc., and most waterborne troops is for simplicity, the user name of using is nonsensical and brief, generally between 1-3 character length.Therefore, utilize this point, can be using user name length as weighing whether one of the standard of waterborne troops's account of an account.As shown in formula (1), this patent be take the unification of user name length 12 to be normalized as standard, that is, the normalization user name length of certain account is that its real user name length is divided by 12.
Normalization user name length U=user length u/12 (1)
Such as user name abc, its normalization user name length is 0.25; User name aven, its normalization user name length is 0.33.Obvious, normalization user name length is less, represents that its user name length is shorter, and this account is that the possibility of waterborne troops's account is also just higher.
Forum's line duration evaluation index
From the online rule of forum, as shown in table 1, the time of normal users login forum is random, there is no very strong regularity, each online hours are different in size, and log in from each time being registered between the last login, should be stochastic distribution from the time.And the work of waterborne troops is the issue water subsides of login forum, so they login the time majority of forum, be regular; From online hours, waterborne troops has generally completed the task Hou Buhui forum that posts and has stayed, so their each online hours are all shorter; Generally can be for an account of a propaganda task registration from login frequency Lai Jiang, waterborne troops, continue for some time with this account note that floods, just abandoned need not for this account afterwards.
Table 1 normal users and forum of waterborne troops line duration Characteristic Contrast table
In order to utilize above-mentioned aspect to distinguish normal users and waterborne troops, this patent has proposed normalization online hours and has logged in the concept of rule index, for the forum's line duration to user and login forum rule, carries out quantitative evaluation.As shown in formula (2), normalization online hours representative of consumer online hours account for the ratio that user registers total duration, and the user of take registers total duration as standard, user's online hours are normalized, thereby measure forum's active degree of this user.As shown in formula (3), log in rule index and represent that the last duration between logging in and logging in for the first time accounts for the ratio of total hour of log-on, significantly, if certain user's normalization online hours are very little, illustrate that this user is registered to the seldom login later of this website; If user's login rule index is very little, illustrate that this user is only active among forum within very concentrated a period of time, and these two kinds of situation ,Dou Shi users' of waterborne troops key character.That is,, if certain user's normalization online hours and to log in rule index lower, this user is that the user's of waterborne troops possibility is larger.
The online total duration t of normalization online hours T=useronline/ user registers total duration Tall(2)
Login rule index D=(the last login time tlast– is landing time t for the first timefirlst)/user registers total duration Tall(3)
Such as, certain user is before the hour of log-on of forum is 2 months, and its duration that logs in forum is total up to 2 hours, and this user's normalization online hours are: 2 hours/(2*30*24 hour).Such as certain user's hour of log-on and all previous landing time are as shown in the table, logging in rule index is 0.0036.
Table 2 user registration and landing time sample table
Browse custom Measure Indexes
The object of login forum of waterborne troops is just pasted in order to issue water, so they are for posting the most of the time, can browse hardly other users' model ,Zhe Shi waterborne troops and normal netizen's very large difference, utilize this feature, domestic consumer and waterborne troops can be distinguished.For certain user of quantitatively evaluating browses the frequent degree of model, this patent has been introduced and has been browsed custom Measure Indexes, be used for characterizing a user and browse the ratio of the shared duration of other user's models in total duration, as shown in formula (4), the time online hours total with respect to this user that this index is browsed user other user's models are normalized, thereby when measurement user is online, browse the time length of other models.When certain user browse custom Measure Indexes when low especially, can think in these user's online hours that most time is posted for oneself, rather than browse others' model, this user is that the user's of waterborne troops likelihood ratio is larger.
Browse custom Measure Indexes L=user and browse model duration tsurfthe online total duration T of/useronline(4)
Such as, according to forum's operation note of certain user A, its duration of browsing other user's models is 3 hours, and its total forum login duration is 240 hours, can obtaining it, to browse custom Measure Indexes be 1/80.
Money order receipt to be signed and returned to the sender frequency Measure Indexes
Waterborne troops is only responsible for particular topic to send out the model of certain content, and can be as normal netizen with regard to certain problem, to beam back note mutual with other netizens, utilize this feature, also normal netizen and waterborne troops can be distinguished, as shown in formula (5), this patent has been introduced money order receipt to be signed and returned to the sender frequency Measure Indexes, and the number of times that others puts question to by user response therewith user always pastes the ratio of number, in representing that this user posts, the proportion that the model exchanging with other people accounts for.When certain user's money order receipt to be signed and returned to the sender frequency Measure Indexes is lower, can infer itself and the exchanging seldom of other users, this user is that the user's of waterborne troops possibility is just larger.
Money order receipt to be signed and returned to the sender frequency Measure Indexes H=replys other people and puts question to subsides number nreply/ total N postsall(5)
Social rangeability figureofmerit
Sociology correlative study shows, social network sites is to have expanded interchange scope to maximum effect of people, makes people's interchange break away to a great extent the restriction of geographic area.Normal netizen has corresponding circle of friends in the constructed environment of forum, forms a small-sized social networks, and correlative study shows, in normal netizen's social circle, friend's number be 100 people to 500 people not etc.And network waterborne troops only take and posts as object, can have hardly social networks circle, so this is also to distinguish whether certain user is an importance of waterborne troops.For the social scope of certain forum user of quantitatively evaluating, this patent is introduced the concept of social rangeability figureofmerit, is used for characterizing the size of user social circle in forum.As shown in formula (6), this index is normalized good friend's quantity of certain user with respect to 500, normal netizen's social rangeability figureofmerit floats between 0.2-1, and there is hardly social scope in the user of waterborne troops, so their social rangeability figureofmerit generally equals or close to 0.
Social rangeability figureofmerit S=good friend quantity n/500 (6)
According to above-mentioned indices, a user's feature just can be described.Therefore this patent characterizes user characteristics with the vector that above-mentioned indices forms, such as certain user's user name length measuring index is 0.5, normalization online hours are 0.3, login rule index is 0.1, browsing custom Measure Indexes is 0.2, money order receipt to be signed and returned to the sender frequency Measure Indexes is 0.1, and social rangeability figureofmerit is 0.5, and this user can describe with [0.50.30.10.20.10.5] this vector so.
Each index relative weighting
Because above-mentioned each index is for judging that whether certain user is the influence degree difference of waterborne troops, therefore, in decision process, should distribute different weights to These parameters.According to relevant The Study of Sociology result, the influence factor of judging for network waterborne troops, by influence degree, sorting successively is from big to small: browse the social rangeability figureofmerit of custom Measure Indexes > > and be accustomed to online Measure Indexes > money order receipt to be signed and returned to the sender frequency Measure Indexes > user name length measuring index.In order to represent that each index is different for the influence degree of judging, this patent has distributed different weights to each index, as shown in the table:
Evaluation indexWeights
Browsecustom Measure Indexes5
Social rangeability figureofmerit4
Onlinecustom Measure Indexes3
Money order receipt to be signed and returned to the senderfrequency Measure Indexes2
User namelength measuring index1
Each index weight value allocation table of table 3
Waterborne troops's account recognition methods based on particle cluster algorithm
Identification to waterborne troops's account, essence is according to certain standard, finds qualified element in set.All users in a fairly simple method Shi Dui forum travel through successively, judge whether it is the user of waterborne troops, but when forum user is huge, this work obviously need to spend the plenty of time.In view of there being a lot of similaritys between the user of waterborne troops, therefore in traversal, can utilize the similarity between them, that is: confirmed that a user is for after the user of waterborne troops, with the higher customer group of its similarity in more easily find the user of waterborne troops.This patent, based on above thought, is used in improved particle cluster algorithm in the search procedure of forum user, thereby has accelerated the recognition speed of waterborne troops's account.
User characteristics description vectors is prepared
As previously mentioned, can obtain the feature description vectors of each user in each relative weighting table of judging index and forum.Suppose and the weights of each evaluation index are distributed as shown in table 2, in forum, the corresponding relation of user ID and feature description vectors is as shown in table 4,
IDFeature description vectors
01[0.2 0.5 0.3 0.1 0.1 0.5]
02[0.1 0.5 0.4 0.2 0.01 0.1]
…………
07[0.01 0.1 0.1 0.05 0.1 0.1]
…………
The corresponding table of table 4 user ID-feature description vectors
Then by all users' feature description vectors, for each index dimension, sort, current system has 6 evaluation indexes to user, can produce 6 sequencing queues, as shown in the table:
The 1st dimension sequence the 2nd dimension sequence the 3rd dimension sequence the 4th dimension sequence the 5th dimension sequence the 6th dimension sequence
010103010707
020302060202
070201030303
………………………………
Table 5 user characteristics description vectors sequencing table
The particle cluster algorithm that now starts application enhancements screens the waterborne troops's account in forum.
Particle cluster algorithm modeling
Particle cluster algorithm is a kind of search procedure based on population, and wherein each individuality is called particle, is defined as the potential solution of problem to be optimized in M dimension search volume, the memory of preserving the optimal location of its historical optimal location, present speed and all particles.Every evolution generation, the component that the information of particle is combined and regulates the speed about on every one dimension, is used to calculate new particle position then.Particle constantly changes their state in multi-dimensional search space, until arrive equilibrium state, or till having surpassed calculating restriction.
If search volume is m dimension, i particle position is expressed as vector xi=(xi1, xi2 ..., xim); The historical optimal location of i particle is pi=(pi1,pi2 ..., pim); , p whereinjfor all pi(i=1,2 ..., the historical optimal location in n); The position rate of i particle is vector vi=(vi1,vi2 ..., vim).Every evolution generation, the position of each particle changes according to current community information, and its position more new formula is:
xi(t+1)=xi(t)+vi(t+1) (7)
Speed more new formula is:
vi(t+1)=ω ⅹ vi(t)+c1 ⅹ rand()ⅹ(pi(t)-xi(t))+c2 ⅹ rand()ⅹ(gi(t)-xi(t)) (8)
Wherein ω is inertia weight; piand gibe respectively historical optimal location and the global history optimal location of this particle; C1 and c2, for the study factor, have represented each particle have been pulled to piand githe weight of the random acceleration term of position; vmaxand xmaxrepresentation speed limits and position limitation respectively.
This patent is chosen user characteristics description amount as particle position information, chooses the weight vector of evaluation index and the dot product of user characteristics description vectors as the good and bad measurement index in position.For example, when system selects particle that ID is 01 as primary, the current location that this particle is corresponding is [0.2 0.5 0.3 0.1 0.1 0.5], and the index of weighing this position quality is the weight vector of evaluation index and the dot product of user characteristics value vector, i.e. [0.2 0.5 0.3 0.1 0.1 0.5] [5,4,3,3,2,1]=4.9.It should be noted that, the user of waterborne troops should have lower custom Measure Indexes, lower social rangeability figureofmerit, lower normalization online hours, lower login rule index, lower money order receipt to be signed and returned to the sender frequency Measure Indexes and the lower user name length measuring index browsed, therefore, when judgement is good and bad, dot product is lower, the feature that more meets the user of waterborne troops, namely in particle cluster algorithm, position is more excellent.Obtain by statistics, the good and bad measurement index in the user's of waterborne troops position has 95% probability to be less than 1.2, so in this patent, the good and bad measurement index in position is less than to 1.2 particle and regards the suspicious user of waterborne troops as.
In speed and position more aspect new formula, because the particle position in this method is discrete, directly application of formula (7) is as position new formula more, so this patent keeps the speed computing formula of particle cluster algorithm constant, and formula is upgraded in position, makes following modification:
For position new formula (7) more, after obtaining present speed vector, successively each dimension of each dimension of current location vector and velocity vector is added, often obtains the value of a dimension, according to each dimension sequencing table of user characteristics vector shown in table 5, move to a nearest position again.Such as, current certain particle position is [0.01 0.1 0.1 0.05 0.1 0.1], the position that is 07 at ID, and present speed is (0.2,0.4,0.2,0.05,0,0.4), from first dimension, particle is carried out to move operation so.Concerning first dimension, positional value accekeration is 0.21, finds all user vectors in the sequence of the first dimension, and the user vector that to find from ID be 07 is nearest, feature value vector the first dimension values is 0.21 or approaches 0.21 particle, the user who is 01 for ID herein.Therefore after the first dimension being calculated, particle has moved to the position that ID is 01.Similarly, for second dimension, positional value accekeration is 0.5, and the eigenwert of the particle that ID is 01 in second dimension was just 0.5 originally, does not need mobile.Similarly, can obtain the positions that this particle moves after 6 times is that ID is 01.
Aspect parameter selection, according to simulating, verifying comparison, the parameter c inparticle cluster algorithm 1 and c2 are set as respectively to 1.4 and can reach speed of convergence faster at 0.6 o'clock.In addition, we arrange inertia weight for the random number between [0,1], choosing Population Size is registered user's quantity in forum, initial population quantity is set to 3, and regulation maximum iteration time is 30, and termination condition is that iterations reaches 30 times or three particle positions converge to same position.
It should be noted that, original particle cluster algorithm is in order to find the best particle position in population, and in this patent, be for all qualified particle positions in searching system, therefore this patent improves predecessor group algorithm, do not change the search procedure of population, but in search procedure, all qualified particle positions are all used as Search Results and preserve, after finishing, algorithm just can find the user ID list of a suspicious waterborne troops and a most suspicious waterborne troops user ID, now from each dimension, search for respectively again this most suspicious user's of waterborne troops neighbours, because " user of waterborne troops is owing to having similar feature, therefore the distribution of the particle that represents the user of waterborne troops in population is close ", in the neighbor node of the most suspicious particle, travel through its neighbours.If neighbours are also suspicious particles, then travel through neighbours' neighbours, until neighbours' neighbours have not been.The neighbours that meet threshold condition are added in the user ID list of suspicious waterborne troops, until its neighbours no longer meet the user of waterborne troops threshold condition.
The particle cluster algorithm screening waterborne troops account of application enhancements
As shown in Figure 5, idiographic flow is as follows for the application particle cluster algorithm searching user's of waterborne troops process flow diagram:
First at random choose three user ID as initial population, such as select ID be 01,07 and 02 user as initial population, the current location that now just can obtain these three particles is respectively [0.20.5 0.3 0.1 0.1 0.5], [0.1 0.5 0.4 0.2 0.01 0.1] and [0.01 0.1 0.1 0.050.1 0.1].
Next calculate its present speed vector.Concerning current location is the particle of [0.2 0.5 0.3 0.1 0.1 0.5], its matching degree computing formula is: [0.2 0.5 0.3 0.1 0.10.5] [5,4,3,3,2,1]=4.9, the matching degree that in like manner can obtain two other particle is 4.42 and 1.2, for current location, is therefore the particle of [0.01 0.1 0.1 0.05 0.1 0.1], can obtain its velocity vector according to formula (8) is [0,0,0,0, the velocity vector that 0,0], in like manner can obtain two other particle is:
1.4ⅹrand()ⅹ([0.2 0.5 0.3 0.1 0.1 0.5]–[0.2 0.5 0.3 0.1 0.10.5])+0.6ⅹrand()ⅹ([0.01 0.1 0.1 0.05 0.1 0.1]–[0.2 0.5 0.30.1 0.1 0.5])
With
1.4ⅹrand()ⅹ([0.1 0.5 0.4 0.2 0.01 0.1]–[0.1 0.5 0.4 0.2 0.010.1])+0.6ⅹrand()ⅹ([0.01 0.1 0.1 0.05 0.1 0.1]–[0.1 0.5 0.40.2 0.01 0.1])
Getting rand () is 0.5, can obtain its velocity vector and be respectively: [0.54-1.2-0.6-0.150-1.2] and [0.27-1.2-0.9-0.45 0.27 0].And when upgrading particle position vector, according to improved position vector update strategy in this patent, can not directly initial position and velocity vector are added and obtain current location vector according to the method for predecessor group algorithm.But after will being added, from front to back particle is moved a certain distance.Such as, for No. 01 particle, its velocity vector is [0.54-1.2-0.6-0.15 0-1.2], from front to back, it need to move 0.54 distance backward in first dimension, second dimension, moves backward 1.2 distances, the 3rd dimension, move backward 0.6 distance, at four dimensions, move backward 0.15 distance, the 5th dimension, keep motionless, the 6th dimension, move backward 1.2 distances.For mobile distance, if not in sorted lists, find the position nearest from this position as shift position.The last mobile position obtaining of No. 01 particle is [0.01 0.1 0.1 0.05 0.1 0.1] so, and in like manner, the position that No. 02 particle moves to is [0.01 0.1 0.1 0.05 0.1 0.1].So just completed one and taken turns evolutionary process.In evolutionary process, if the particle matching degree of process reach the user of waterborne troops and identify threshold value, user ID corresponding to this particle added to the user ID list of suspicious waterborne troops.
Pass through some evolutions of taking turns, can after meeting termination condition, finish particle cluster algorithm, obtain particle optimal location.
In predecessor group algorithm, obtain particle optimal location and can finish algorithm flow, but the object of this patent is to filter out all qualified individualities in set, therefore after obtaining optimum particle position, need to around selecting particle, all directions screen, until screen ineligible particle.The position of particle can represent with vector, and all particle vectors can sort in each dimension, just can in each dimension, find respectively its neighbours.Such as, the vector (1,1) of two dimension, its neighbours are (0,1) (1,0) (1,2) (2,1).
The foregoing is only preferred embodiment of the present invention, in order to limit the present invention, within the spirit and principles in the present invention not all, any modification of doing, be equal to replacement, improvement etc., within all should being included in protection scope of the present invention.

Claims (10)

CN201310611396.9A2013-11-262013-11-26Method and system for network navy account number identification based on particle swarm optimizationExpired - Fee RelatedCN103617235B (en)

Priority Applications (1)

Application NumberPriority DateFiling DateTitle
CN201310611396.9ACN103617235B (en)2013-11-262013-11-26Method and system for network navy account number identification based on particle swarm optimization

Applications Claiming Priority (1)

Application NumberPriority DateFiling DateTitle
CN201310611396.9ACN103617235B (en)2013-11-262013-11-26Method and system for network navy account number identification based on particle swarm optimization

Publications (2)

Publication NumberPublication Date
CN103617235Atrue CN103617235A (en)2014-03-05
CN103617235B CN103617235B (en)2017-01-25

Family

ID=50167938

Family Applications (1)

Application NumberTitlePriority DateFiling Date
CN201310611396.9AExpired - Fee RelatedCN103617235B (en)2013-11-262013-11-26Method and system for network navy account number identification based on particle swarm optimization

Country Status (1)

CountryLink
CN (1)CN103617235B (en)

Cited By (13)

* Cited by examiner, † Cited by third party
Publication numberPriority datePublication dateAssigneeTitle
CN103955714A (en)*2014-04-092014-07-30中国科学院信息工程研究所Navy detection model construction method and system and navy detection method
CN104484359A (en)*2014-12-022015-04-01北京锐安科技有限公司Public opinion analysis method and public opinion analysis device based on social graph
CN106095813A (en)*2016-05-312016-11-09北京奇艺世纪科技有限公司A kind of identification method of user identifier and device
CN106330837A (en)*2015-06-302017-01-11阿里巴巴集团控股有限公司Suspicious network user identification method and device
CN106611104A (en)*2016-10-312017-05-03中南大学Analog computation method and system for complex metallurgical process
CN107332931A (en)*2017-08-072017-11-07合肥工业大学The recognition methods of waterborne troops of machine type forum and device
CN108197696A (en)*2018-01-312018-06-22湖北工业大学A kind of network navy account recognition methods and system
CN108431832A (en)*2015-12-102018-08-21渊慧科技有限公司 Amplify Neural Networks with External Memory
CN109559245A (en)*2017-09-262019-04-02北京国双科技有限公司A kind of method and device identifying specific user
CN110728543A (en)*2019-10-152020-01-24秒针信息技术有限公司Abnormal account identification method and device
CN112115324A (en)*2020-08-102020-12-22微梦创科网络科技(中国)有限公司Method and device for confirming praise-refreshing user based on power law distribution
CN112667876A (en)*2020-12-242021-04-16湖北第二师范学院Opinion leader group identification method based on PSOTVCF-Kmeans algorithm
CN113890756A (en)*2021-09-262022-01-04网易(杭州)网络有限公司User account number chaos degree detection method, device, medium and computing equipment

Citations (4)

* Cited by examiner, † Cited by third party
Publication numberPriority datePublication dateAssigneeTitle
CN102571484A (en)*2011-12-142012-07-11上海交通大学Method for detecting and finding online water army
CN102629904A (en)*2012-02-242012-08-08安徽博约信息科技有限责任公司Detection and determination method of network navy
CN103095499A (en)*2013-01-172013-05-08上海交通大学Method for capturing water armies on microblog platforms
CN103198161A (en)*2013-04-282013-07-10中国科学院计算技术研究所Microblog ghostwriter identifying method and device

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication numberPriority datePublication dateAssigneeTitle
CN102571484A (en)*2011-12-142012-07-11上海交通大学Method for detecting and finding online water army
CN102629904A (en)*2012-02-242012-08-08安徽博约信息科技有限责任公司Detection and determination method of network navy
CN103095499A (en)*2013-01-172013-05-08上海交通大学Method for capturing water armies on microblog platforms
CN103198161A (en)*2013-04-282013-07-10中国科学院计算技术研究所Microblog ghostwriter identifying method and device

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
刘芳 等: "基于粒子群优化算法的社交网络可视化", 《浙江大学学报(工学版)》*

Cited By (20)

* Cited by examiner, † Cited by third party
Publication numberPriority datePublication dateAssigneeTitle
CN103955714A (en)*2014-04-092014-07-30中国科学院信息工程研究所Navy detection model construction method and system and navy detection method
CN104484359A (en)*2014-12-022015-04-01北京锐安科技有限公司Public opinion analysis method and public opinion analysis device based on social graph
CN106330837A (en)*2015-06-302017-01-11阿里巴巴集团控股有限公司Suspicious network user identification method and device
CN108431832A (en)*2015-12-102018-08-21渊慧科技有限公司 Amplify Neural Networks with External Memory
US12299575B2 (en)2015-12-102025-05-13Deepmind Technologies LimitedAugmenting neural networks with external memory
CN106095813A (en)*2016-05-312016-11-09北京奇艺世纪科技有限公司A kind of identification method of user identifier and device
CN106611104B (en)*2016-10-312021-04-20中南大学 Simulation calculation method and system for complex metallurgical process
CN106611104A (en)*2016-10-312017-05-03中南大学Analog computation method and system for complex metallurgical process
CN107332931A (en)*2017-08-072017-11-07合肥工业大学The recognition methods of waterborne troops of machine type forum and device
CN109559245A (en)*2017-09-262019-04-02北京国双科技有限公司A kind of method and device identifying specific user
CN109559245B (en)*2017-09-262022-02-25北京国双科技有限公司Method and device for identifying specific user
CN108197696A (en)*2018-01-312018-06-22湖北工业大学A kind of network navy account recognition methods and system
CN110728543B (en)*2019-10-152022-08-09秒针信息技术有限公司Abnormal account identification method and device
CN110728543A (en)*2019-10-152020-01-24秒针信息技术有限公司Abnormal account identification method and device
CN112115324A (en)*2020-08-102020-12-22微梦创科网络科技(中国)有限公司Method and device for confirming praise-refreshing user based on power law distribution
CN112115324B (en)*2020-08-102023-10-24微梦创科网络科技(中国)有限公司Method and device for confirming praise and praise users based on power law distribution
CN112667876A (en)*2020-12-242021-04-16湖北第二师范学院Opinion leader group identification method based on PSOTVCF-Kmeans algorithm
CN112667876B (en)*2020-12-242024-04-09湖北第二师范学院Opinion leader group identification method based on PSOTVCF-Kmeans algorithm
CN113890756A (en)*2021-09-262022-01-04网易(杭州)网络有限公司User account number chaos degree detection method, device, medium and computing equipment
CN113890756B (en)*2021-09-262024-01-02网易(杭州)网络有限公司Method, device, medium and computing equipment for detecting confusion of user account

Also Published As

Publication numberPublication date
CN103617235B (en)2017-01-25

Similar Documents

PublicationPublication DateTitle
CN103617235A (en)Method and system for network navy account number identification based on particle swarm optimization
US11659050B2 (en)Discovering signature of electronic social networks
CN107330115B (en) Method and device for recommending information
CN111444395A (en)Method, system and equipment for obtaining relation expression between entities and advertisement recalling system
CN110532351A (en)Recommend word methods of exhibiting, device, equipment and computer readable storage medium
WO2007149341A2 (en)System to associate a demographic to a user of an electronic system
CN109460519B (en)Browsing object recommendation method and device, storage medium and server
CN112258223B (en)Marketing advertisement click prediction method based on decision tree
CN107918825A (en)A kind of method and apparatus that age of user section is judged based on application installation preference
CN108304853A (en)Acquisition methods, device, storage medium and the electronic device for the degree of correlation of playing
CN109889436A (en) A method for discovering spammers in social networks
CN113609394A (en)Information flow-oriented safety recommendation system
CN112463859B (en) User data processing method and server based on big data and business analysis
CN108304452A (en)Article processing method and processing device, storage medium
CN108388652B (en)Method and device for sending song list identification
CN110457387A (en)A kind of method and relevant apparatus determining applied to user tag in network
CN112651790B (en)OCPX self-adaptive learning method and system based on user touch in quick-elimination industry
CN108171538A (en)User data processing method and system
CN112132589A (en)Method for constructing fraud recognition model based on multiple times of fusion
CN107562793A (en)A kind of big data method for digging
Ying et al.PFrauDetector: a parallelized graph mining approach for efficient fraudulent phone call detection
CN109636057A (en)Based on the link prediction technique and storage medium for improving artificial immune system
CN104462412A (en)Keyword detection method and device for release of internet keywords
CN110598973A (en)IAP-based risk evaluation method for authentication process of green furniture product
CN110209953A (en)A kind of calculation method towards uncertain social computing problem

Legal Events

DateCodeTitleDescription
PB01Publication
PB01Publication
C10Entry into substantive examination
SE01Entry into force of request for substantive examination
C14Grant of patent or utility model
GR01Patent grant
CF01Termination of patent right due to non-payment of annual fee
CF01Termination of patent right due to non-payment of annual fee

Granted publication date:20170125


[8]ページ先頭

©2009-2025 Movatter.jp