CN107866072A - A kind of system that plug-in detection is carried out using increment decision-making tree - Google Patents
A kind of system that plug-in detection is carried out using increment decision-making tree Download PDFInfo
- Publication number
- CN107866072A CN107866072A CN201711045371.1A CN201711045371A CN107866072A CN 107866072 A CN107866072 A CN 107866072A CN 201711045371 A CN201711045371 A CN 201711045371A CN 107866072 A CN107866072 A CN 107866072A
- Authority
- CN
- China
- Prior art keywords
- tree
- plug
- model
- player
- decision
- 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
Links
Classifications
-
- A—HUMAN NECESSITIES
- A63—SPORTS; GAMES; AMUSEMENTS
- A63F—CARD, BOARD, OR ROULETTE GAMES; INDOOR GAMES USING SMALL MOVING PLAYING BODIES; VIDEO GAMES; GAMES NOT OTHERWISE PROVIDED FOR
- A63F13/00—Video games, i.e. games using an electronically generated display having two or more dimensions
- A63F13/70—Game security or game management aspects
- A63F13/75—Enforcing rules, e.g. detecting foul play or generating lists of cheating players
-
- A—HUMAN NECESSITIES
- A63—SPORTS; GAMES; AMUSEMENTS
- A63F—CARD, BOARD, OR ROULETTE GAMES; INDOOR GAMES USING SMALL MOVING PLAYING BODIES; VIDEO GAMES; GAMES NOT OTHERWISE PROVIDED FOR
- A63F13/00—Video games, i.e. games using an electronically generated display having two or more dimensions
- A63F13/55—Controlling game characters or game objects based on the game progress
- A63F13/58—Controlling game characters or game objects based on the game progress by computing conditions of game characters, e.g. stamina, strength, motivation or energy level
-
- A—HUMAN NECESSITIES
- A63—SPORTS; GAMES; AMUSEMENTS
- A63F—CARD, BOARD, OR ROULETTE GAMES; INDOOR GAMES USING SMALL MOVING PLAYING BODIES; VIDEO GAMES; GAMES NOT OTHERWISE PROVIDED FOR
- A63F13/00—Video games, i.e. games using an electronically generated display having two or more dimensions
- A63F13/80—Special adaptations for executing a specific game genre or game mode
- A63F13/822—Strategy games; Role-playing games
-
- A—HUMAN NECESSITIES
- A63—SPORTS; GAMES; AMUSEMENTS
- A63F—CARD, BOARD, OR ROULETTE GAMES; INDOOR GAMES USING SMALL MOVING PLAYING BODIES; VIDEO GAMES; GAMES NOT OTHERWISE PROVIDED FOR
- A63F2300/00—Features of games using an electronically generated display having two or more dimensions, e.g. on a television screen, showing representations related to the game
- A63F2300/80—Features of games using an electronically generated display having two or more dimensions, e.g. on a television screen, showing representations related to the game specially adapted for executing a specific type of game
- A63F2300/807—Role playing or strategy games
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Human Computer Interaction (AREA)
- Business, Economics & Management (AREA)
- Computer Security & Cryptography (AREA)
- General Business, Economics & Management (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of system that plug-in detection is carried out using increment decision-making tree, including：Data preprocessing module, is cleaned and characteristic vector pickup to the initial data of player actions；Model generates and interactive module, and the characteristic vector of the data preprocessing module is come into generation model as the input of model and exported, while receives feedback and model is adjusted；High level view visualization model, dynamic tree graph, recommendation and display panel and accuracy rate/recall rate line chart are generated according to the output model of model generation and interactive module；The present invention shows different periods decision process using dynamic decision tree, analyzes the characteristics of plug-in, is judged as the reason for plug-in, it is found that it significantly distinguishes over some features of normal player；And due to model characteristics, some characteristics of plug-in differentiation can be excavated；In addition user can also add the knowledge of oneself and beta pruning is carried out to decision tree, and combined by each view and carry out other analysis processes.
Description
Technical field
It is more particularly to a kind of that plug-in detection is carried out using increment decision-making tree the present invention relates to plug-in detection technique field of playing
System.
Background technology
More online Role Playing Games of people are created from the virtual society world that each corner player in the whole world participates in.Play
It can interact, and be taken a significant amount of time in culture their game role of upgrading, energy and money between family.Investigation hair
Now only 2016, the more online RPGs of people in the whole world brought 19,800,000,000 dollars of income.However, just due to this game
Intimately, it also becomes the hotbed of some network crimes.Among this kind of critically important behavior be unofficial permission, approval show
Gold transaction, this trading activity is that some virtual objects are traded with real money.Some advanced game are plug-in existing
This kind of related unlawful practice is being exclusively carried out, is obtaining profit, but is having a strong impact on game balance, injury game operation,
Game company takes in and player experience, game economy system or even the sustainable development entirely played.
Therefore, gaming operators take a large amount of measures, the plug-in method of common detection has three big kinds, is divided into client
Side, network side and server side detection method.Client-side is embedded in subscriber's main station game client similar kill with similar
The principle of malicious software is plug-in to detect, but some elaborate plug-in can not be detected by this now；Network side needs
Network traffics are analyzed, the problems such as this can cause excessive network load and network delay again；Server end detection side
Method is by analyzing the daily record data of player on server.This method is now popular method, and they are often
Regard plug-in test problems as abnormality detection problem.But existing method has some shortcomings, for example plug-in design is more multiple
Miscellaneous, simplex algorithm input and output understand difficult very big for analysis personnel；Still further aspect, advanced plug-in also renewal upgrading
Function so that plug-in detection becomes increasingly difficult to.At this time, the introduction of visual analysis technology is just particularly important.
There are some visual analysis systems on abnormality detection.Such as Jian Zhao et al. #FluxFlow systems
The forwarding behavior of Twitter user is visualized, with reference to contextual informations such as user properties, existed to study abnormal information
The feature spread on Twitter.Nan Cao et al. TargetVue is changed over time by one assigns user's suspicion journey
The model of degree understands potential social robot on social media platform (for example those sends waste advertisements to help to analyze personnel
Robot account), including their exchange activity, behavioural characteristic and social interaction etc..These systems are by integrating the neck of people
The internal procedure of algorithm model is regarded as a flight data recorder by domain knowledge to optimize the process of abnormality detection, analyzes personnel
The relevant information of algorithm model can not be understood by their system.
Although existing method provides abundant contextual information, researcher still can not deeply participate in model
The process of structure.Therefore, it is intended that system can allow for user participate in model construction go with it is more preferable help understand model,
Potential explain is found for some phenomenons.
The content of the invention
The invention provides a kind of system that plug-in detection is carried out using increment decision-making tree, can help to analyze personnel's detection
It is plug-in, understand plug-in detection model, excavate the feature that the behavior of plug-in and normal player or action change over time, moreover it is possible to
Allow analysis personnel to be interacted with model, interact the exploration of formula visual analysis.
A kind of system that plug-in detection is carried out using increment decision-making tree, including：
Data preprocessing module, is cleaned and characteristic vector pickup to the initial data of player actions；
Model generates and interactive module, is generated using the characteristic vector of the data preprocessing module as the input of model
Model simultaneously exports, while receives feedback and model is adjusted；
High level view visualization model, the tree of the decision tree changed between mainly being represented therewith with icicle figure placed side by side
Structure, dynamic tree graph, recommendation and display panel and accuracy rate/call together are generated according to the output model of model generation and interactive module
Return rate line chart；
The dynamic tree graph shows the tree construction of decision tree, the icicle with a kind of compact mode (using icicle figure)
The each node in figure the inside represents the split vertexes of a decision tree, and multiple icicle figures are placed to embody decision tree with the time side by side
Change；
In the recommendation and display panel, chosen recommending panel to be illustrated in the form that can be sorted in the dynamic tree graph
The information of all nodes (rectangle of decision tree) of the icicle figure of (one or more), including accuracy rate, recall rate, accurate rate and
Frequency of occurrence etc.；A node is chosen in dynamic tree graph if being switched in display panel, is shown with the form of radar distribution map
The situation of player is included in this node；
The accuracy rate/recall rate line chart and the dynamic tree graph arrange up and down according to corresponding time relationship, represent with
Accuracy rate/recall rate of every decision tree prediction of change of time.
The initial data of player actions stores according to various different actions, daily every kind of one file of action (ratio
Such as on January 1st, 2017, daily record is logged in), the game one shares hundreds of action.Count within the period interested, each
The frequency of action of the people on different time piece.Notice that the period here both can be (time-division date natural time
Second) or playtime (what action is 1 grade, 2 grades ... done respectively, and what the corresponding frequency is).So each player
Each timeslice has a corresponding characteristic vector.It is more additionally, due to amount of action all too, one point has been carried out to action
Class (task is related, and attribute is related, and fight is related, and article is related).This classification both can be that (for example Game analysis is special by user
Family) oneself specify, can also be some complete inducing classification methods in some existing literatures.
Decision tree presents the flow of a decision-making, and from root node, down each nonleaf node is a bar judged
Whether part, some attribute that can judge it according to the example come in meet the condition of this non-leaf node.Example gets to leaf
After node, leaf node can provide the label of a classification, to tell you which kind of this example belongs to.Traditional decision tree
Training process, it is exactly constantly recursively to the leaf node of the decision tree generated, according to some indexs, (for example information increases
Benefit, Gini indexes etc.) determine the child node below this leaf node is used as division with what attribute.
As shown in figure 1, decision tree refers to, for example will consider to go out today to climb the mountain, such a decision process is had.
Here each square frame is a node, has several Rule of judgment below him, is made decisions according to different condition.Square can be passed through
Shape represents the node of tree, but each node also has several options, for example humidity is high or humidity is normal, strong wind or gentle breeze,
Along tree always down until leaf node just has judged result so there is (weather in the path do not climbed:Rain) → do not climb, (weather:
It is fine) → (humidity:Greatly) → do not climb, (weather:It is cloudy → to have wind:Strong wind) → do not climb.
The model that the present invention can use is a kind of Hoeffding adaptive decision-making trees using Gauss discretization
(Hoeffding adaptive tree with Gaussian discretization).This is an on-line Algorithm, and he is sharp
With Hoeffding circle characteristic so that decision tree can accomplish on-line training, i.e. data have carried out can and bring training, and
Only using once in decision tree；Rather than lot data are needed as traditional decision tree, each data will be used for
Judge that splitting condition is multiple.
Hoeffding definitions, a stochastic variable, scope R, true average deviates it after n independent observation
Estimate is not more than with probability 1- δWhen making and dividing of that attribute in one node of judgement
Wait, find that information gain is maximum and the two second largest attributes, calculate the difference of their information gains, if being more than this ∈,
It can guarantee that a positive tree node splitting effect.Such a boundary can be helped at only part or low volume data,
Training sets out, without that have to wait until that all data have all come just train complete.This decision tree is referred to as Hoeffding
Decision tree.
On the basis of this method, the present invention is had made some improvements using prior art again.Firstly, since Hoeffding
Decision tree only supports discrete value attribute, employs sane increment type Gauss discretization method so that supports Continuous valued attributes.Its
It is secondary, the characteristics of concept drift in data also be present, so-called concept drift refers to such a case：Data generation may be not
It is stable, its generating process may change.Correspond in game data, plug-in behavior be able to may also occur
Change, it is outer to prevent from continuing being sealed off because plug-in company can perceive the plug-in of oneself and change some features by after title
Hang.On the other hand, ADapative WINdowing (ADWIN) technology is employed, Hoeffding decision-making of this method in script
A window and corresponding estimator are added in tree, changes detector etc., to find that concept drift above-mentioned shows
As.
So whole method, just it is referred to as the Hoeffding adaptive decision-making trees using Gauss discretization.To data profit
With this method, as data constantly flow into, decision tree can constantly grow, when detecting some significant changes, decision-making
Setting some subtrees can replace, become other stalk tree.Then a tree changed over time has been obtained, and each
In timeslice, a people will have one to be judged as YES plug-in or normal person judged result (0 normal player, 1 plug-in object for appreciation
Family).
Specifically, to establish process as follows for model：
Step 1：Initialize the state of decision tree：
A root node is established, a statistic, entitled A are initialized in root nodeijk, this statistic is ADWIN methods
(in a part for step 2).To each example (example (x, y), a certain personal action in an x namely period
Characteristic vector, and whether it is plug-in decision content y, 0 is normal person, and 1 is plug-in), first tested with the tree generated, then
Hoeffding into below step 2 adaptively sets growth step.
Step 2:The growth of decision tree
First, (x, y) above is first grouped into some leaf node along existing decision tree by a decision path,
Update the estimator A of the node passed through in this all path and leaf nodeijk.If current leaf node l has one can replace
Change tree TaltIf, it (is exactly this current step that this, which alternatively sets also Hoeffding corresponding to execution and adaptively sets growth step,
Hoeffding adaptively set).It (can certainly be the finger commonly used in other decision trees to calculate the information gain G each acted
Mark), the effect of this step is to assess the present condition of leaf node if appropriate for work to divide.
The computational methods of information gain are such:There are a common concept, referred to as comentropy first, for one point
Cloth, than having three classifications A, B, a C according to collection if any number, the ratio that each classification accounts for be p (A)=0.2, p (B)=0.3, p (C)=
1-0.2-0.3=0.5 then comentropy H=- ∑sipilog(pi).Herein, be exactly 0.2log (0.2)+0.3log (0.3)+
0.5log (0.5)=0.301.In general, entropy is smaller, more orderly：Such as when data set only has a classification, entropy 0, this
It is the minimum value of entropy, represents that data set is complete ordering.On this basis, an also conditional information entropy, and H (Y | X)=∑x∈Xp
(x) H (Y | X=x).Here Y is to be judged as whether being plug-in, this conditional entropy is a description to division, and X is certain
Enter the number of copy in action, such as game.The difference of entropy before and after the division acted for certain, and H (X)-H (Y | X) i.e.
For information gain.Above-mentioned i is exactly a kind of traversal mode, and i just represents A, these three classifications of B, C, it can be common thatIf set N={ 1,2,3 ..., n } so ∑si∈NI is equal to 1+2+3+ ...+n.
When calculating information gain, it is necessary to the step of using a discretization, if the value of this action be successive value without
If being centrifugal pump, so just it is easy to divide.Here a kind of Gauss discretization method of sane increment type is used, utilization is this
After method, by successive value discretization, following division is carried out.If the information gain of that maximum attribute of information gain,
The value for subtracting that second largest information gain is more thanSo to maximum information gain that act into
Line splitting, and an estimator is initialized to each splitting non-zero branch.The method that division uses two points, i.e., by successive value section
It is divided into more than split point and less than or equal to split point two parts.Here Hoeffding circle concept has been used, its implication
It is a stochastic variable, scope R, true average is deviateed its estimate and is not more than with probability 1- δ after n independent observationIf change detector (change detector, also using ADWIN as change detector) is found
Change has been distributed with caused by data, if leaf node l is not set alternatively, a replaceable tree is created in leaf node
Talt；If it is more accurate replaceable tree to be present, then present node l is just replaced with replaceable tree Talt。
Step 3:ADWIN methods
Initialize sliding window W, variance, sum.
A, when a new example arrives, it is added to window W；
B, for window W any segmentation, W=W0+W1, ifIt is unsatisfactory for, then throws away window
Last element in mouthful, untill this formula meets, wherein,(n0And n1Harmonic-mean),WithIt is the statistical average in window W0 and W1, n0And n1It is window
Length.
C, if step b has the example thrown away, then as change detector, tells external program data distribution to send out
Change is given birth to.
According to model above, and in each timeslice, a people will have one be judged as YES it is plug-in either
The judged result of normal person.Suspicion degree just uses the average value of the value of all judged results of current time piece.A such as people
It is 0,0,0,0,1,1,0 in the judged result of timeslice 1 to 7, then he is set to 2/7 in the suspicion degree of the 7th timeslice.
Recommending to represent a kind of information with the form in the recommendation label in display panel, each row, accuracy rate, recall
Rate, accurate rate and frequency of occurrence etc.；Some node in the icicle figure chosen is represented per a line.Grey strip length in form
It is proportional with relative size of its value in this row.Click form new line, i.e. accuracy rate, recall rate, accurate rate and appearance
The frequency, can be with the sequence by column progress from high to low or from low to high.
The algorithm of radar distribution map：
Projection is that the point of some higher-dimensions (is exactly multi-C vector, it is because being generally more than three-dimensional thing why to say higher-dimension
Cannot directly it draw, people can not imagine four-dimensional and four-dimensional thing above, so he is converted into low dimensional,
Draw and allow people to see), want it to become drawing in the plane for two dimension, it is more in certain period that each point represents a player
The characteristic vector of dimension.
Constraint 1：Ensure that the distance in higher dimensional space and the distance in low dimensional space are as far as possible close.Here distance is general
With regard to Euclidean distance, so calculating, it is assumed that for example have the point of two 4 dimensions, x1(x11,x12,x13,x14),x2(x21,x22,
x23,x24), they are exactly at distance
They project become two dimension point distance be exactly then it is usually well known that apart from expression way.
Constraint 2：Consider it is a radar projections in addition, be round, and it is the big of suspicion degree index to put to circle center distance
It is small.So this projection just more constraintss of such a radius.
Consideration represents to project the point of later two dimensional surface with polar coordinates：
pi=(ri·s(ki),θi)
Polar coordinates are exactly (ri,θi), the former is that radius the latter is central angle.Being converted to xy coordinates (cartesian coordinate) is exactly
(ricosθi,risinθi), but more here individual thing, s (k herei) why presence is because " to ensure empty in higher-dimension
Between distance and distance in low dimensional space it is as far as possible close ", be to try to approach here, a little difficulty is accomplished because also having in fact
Radius Constraint condition, so this is a polar disturbance term.HereSoft is one
Individual parameter, instruct disturbance size, here 0.2, can also be selected as needed.
Imitate the projection calculation of multidimensional scaling (multidimensional scaling, MDS) so that projected
Point and point between the difference of distance and original higher dimensional space distance minimize, that is, minimize this function：∑ij
(distij-||pi-pj||)2
Here | | pi-pj| | a kind of and distance, just use Euclidean distance here, why so expression be in order to distinguish on
The dist in faceij.In polar coordinates generation, is entered, last this function that bring minimum above reforms into this appearance.
Then with gradient descent method with regard to k can be obtainedi,θiValue, and then can is drawn on two-dimentional disc, its flute
Coordinate (i.e. xy coordinates) is (r under karr coordinate systemis(ki)cosθi,ris(ki)sinθi)。
Point usually characterizes risk factor to the distance in the center of circle on radar map, but also mentions above, due to constraint 1 about
Beam 2 will meet simultaneously, thus will some points can deviate the position of a little actual accurate suspicion degree index.Then sometimes
It is possible to can be appreciated that some points (point of cluster), it is closer from the center of circle but be not again these cans completely on the center of circle
Object as observation.
Preferably, there are many points in the radar distribution map, each point represents state of the player in current time,
Whether each various operating frequencies of player are that (0 or 1,0 represents normal plays plug-in mark as multi-C vector, and with one
Family, 1 represents plug-in)；
These multi-C vectors are projected on the disc of two dimension, for showing relation between player, and the danger of player
Degree；
On the disc, the Euclidean distance between player, which is tried one's best, keeps the distance of original multi-C vector；And player to circle
The distance of the heart subtracts suspicion degree with 1 and represents that (suspicion degree is the value between 0~1, to export player indicia to current time segment model
The average of value)；
Preferably, the point is round dot, provided with transparency.The point is translucent, so as fruit dot aggregation is more
Local color will be deeper.A kind of such design is the design for having used for reference military radar in fact, because radar is closer to circle
The heart is more dangerous.
Preferably, the high level view visualization model also includes the thumbnail of tree, the breviary as the dynamic tree graph
Figure, is arranged between the dynamic tree graph and the accuracy rate/recall rate line chart.(for decision tree it is more when it is useful)；
The thumbnail of tree, it can be used as from the point of view of time shaft, user can be brushed above and select different thumbnails, and the tree of so top also can
These are restricted to by the tree scope of frame choosing.
Including personal panel and packet panel preferably, in addition to detailed view visualization model,；Multiple row is mutually related
Inter-view；
The personal panel is used to show the situation that individual player's behavior, action and suspicion degree change over time；
The packet panel is used to showing that (for example one group to be plug-in player, and one group is normal player by two groups of players choosing；
Either chosen in radar distribution map two groups) distribution of the value of attribute situation about changing over time.
For preferably display data, user is facilitated to find out plug-in player, it is preferred that everyone is every in the personal panel
The individual period represents that change information is represented with a full line over time for everyone with a rectangle frame and its internal bar chart,
One quantitative value for representing various actions under a kind of behavior of each bar chart adds up, and each bar chart has 4 to represent player's row
To be divided into 4 types herein.
For preferably display data, user is facilitated to find out plug-in player, it is preferred that to double-click some on the personal panel
The bar chart of color, then displaying are sorted in the line chart of each detailed action of the action under this behavior, showed each
The situation that kind action changes over time.
For preferably display data, user is facilitated to find out plug-in player, it is preferred that in the packet panel, often to go
Represent it is a kind of act, two groups of players situation about changing over time, every group of player is a horizontal bar shaped, show maximum and
The section of minimum value.
The operation content of present system is as follows：
User mutual
Focus and context:Choosing is brushed on tree thumbnail (time shaft), above dynamic tree indication range can also change therewith.
Mouse frame selects several trees in dynamic tree graph, this several trees can by the expansion of transverse direction, and it is not selected can laterally be reduced, this
Sample prescription just observes details.And panel is recommended to show the various information of the node in the tree chosen therewith.In radar distribution map
After middle double-click, local radial amplification mode is opened；Point around mouse can be exaggerated in radial direction, and other point can be corresponded to
Be compressed
When mouse-over is on the node in some tree, the node consistent with it can be linked up with line
Ash point is clicked in personal panel can shrink and (flatten grey belt).When bar chart deploys, certain is double-clicked
The bar chart of color, the situation of everything under this behavior of more details can be showed, represented with corresponding color line chart.Institute
There is action all to be indicated with the id of the action of beginning.
Search：In the upper right corner, search box can input the id (multiple centres are separated with comma) of one or more player,
Then can be shown in lower left corner individual's panel.
Filtering：In personal panel, some color bar chart is clicked, the bar shaped ash of remaining color can be darkened, current
Color is constant, facilitate the simple more this behavior of analysis personnel change over time and people and people between relation.
Pull：In personal panel, compare for convenience, small ash point can be pinned and pulled, position is exchanged, carry out comparative analysis people
Member two or more players interested.
View linkage：When mouse-over is when on node, node can also be shown corresponding to the display panel on the right.
Amplification：When left button clicks a node in dynamic tree graph, corresponding personal and packet panel can all be shown；
Radar distribution map occurs in display panel simultaneously.
Interacted with model：System supports the function-beta pruning interacted with model.Decision tree beta pruning is a common interaction,
By stopping the division of some nodes, to control the process that tree grows.
The analysing content of present system is as follows：
1st, show the process of the dynamic evolution of decision tree, show the flow of the decision-making of decision tree, how about a people is judged to
Break to be plug-in or non-plug-in.
2nd, varigrained player's behavior, the evolution of action are analyzed.
3rd, have for analysis personnel interaction and timely feed back, prompt, and the information of appropriate context.According to these letters
Breath, can help user to find some pattern Producing reasons.
4th, analysis personnel can interact in itself with model.The professional knowledge of analysis personnel should be added to Analysis of Policy Making
During go, for adjusting this model.Also help to analyze other patterns simultaneously.
The operating process of present system：
Initially enter model training.
After having trained, a series of tree can be produced, is sequenced from left to right in dynamic tree graph.
Mouse-over is in tree node, it can be seen that presence of the same node point in different trees；Frame selects one piece of region, it can be seen that
Selected tree is exaggerated, and surrounding is set and reduced.Tree has the small tree thumbnail of a row below, it can be worked as time shaft, had
Individual telescopic selection of time frame, help to choose the tree to show in dynamic tree graph.
Exaggerated tree is chosen for frame, the right is recommended panel to show the various indexs of tree interior joint, such as accurate rate, called together
The rate of returning, (computational methods are F1) and frequency of occurrence etc..
The node in some tree is clicked, the display panel for having side occurs radar distribution map, shows what is occurred in node
One player relationship information of player and player's risk factor information.Simultaneously, personal panel and packet panel can also update.It is opened up
Show and had been introduced above content.In use, because time shaft is alignment, it is very convenient by comparison.It is simultaneously at all levels
Details can all show can then correspond to the object for appreciation letter from home in display area in radar distribution map chosen area, personal panel simultaneously
Breath.
Packet panel typically shows the distribution of plug-in in the personal panel and frequency under each action of regular player
Aggregation information.If but clicked in personal panel and compare button, can in radar map, frame selects one piece again, then correspond in individual
Also there is the detailed view of another a group of people in panel, and below figure dotted box portion is compared at this.
Beneficial effects of the present invention：
The system that plug-in detection is carried out using increment decision-making tree of the present invention, is showed different periods using dynamic decision tree and determined
Plan process, analyze the characteristics of plug-in, be judged as the reason for plug-in, it is found that it significantly distinguishes over some spies of normal player
Sign；And due to model characteristics, some characteristics of plug-in differentiation can be excavated；In addition user can also add the knowledge of oneself and fight to the finish
Plan tree carries out beta pruning, and is combined by each view and carry out other analysis processes.
Brief description of the drawings
Fig. 1 be present system high level view visualization model in dynamic tree graph, tree thumbnail and accuracy rate/
The imaging schematic diagram of recall rate line chart.
Fig. 2 is the imaging schematic diagram of the personal panel of present system.
Fig. 3 is the result schematic diagram of the line chart for the action that Fig. 2 double-clicks bar chart displaying.
Fig. 4 is the imaging schematic diagram of the packet panel of present system.
Fig. 5 is the imaging schematic diagram of the recommendation panel of present system.
Fig. 6 is the imaging schematic diagram of the display panel of present system.
Fig. 7 is the enlarged diagram of three trees in the black surround chosen in Fig. 1.
Fig. 8 is the imaging schematic diagram of the radar distribution map under a certain data cases of the display panel of present system.
Embodiment
The present invention is described in detail below in conjunction with the accompanying drawings, the purpose of the present invention and effect will be apparent.
As shown in Fig. 1~7, the structure and construction of the system that plug-in detection is carried out using increment decision-making tree of the present embodiment
Comprise the following steps and content：
Step 1：Visual design：
Data preprocessing module, to log data cleaning and characteristic vector pickup；
Model generates and interactive module, and characteristic vector above is come into generation model as the input of model and exported；Together
When it can receive the feedback of visualization model model is adjusted；
High level view visualization model, the tree of the decision tree changed between mainly being represented therewith with icicle figure placed side by side
Structure, including：
Dynamic tree graph, Fig. 1 top halfs, major part, by the tree construction of decision tree with a kind of compact mode (icicle
Figure) show, each node in the inside represents the split vertexes of a decision tree；Multiple icicle figures are placed side by side, embody decision-making
Tree is changed with time, as shown in Figure 7.
Recommendation and display panel, are selected as shown in figure 5, being illustrated in panel is recommended with the form that can be sorted in dynamic tree graph
In (one or more) tree all nodes information, including accuracy rate, recall rate, accurate rate and frequency of occurrence etc.；Such as Fig. 6 institutes
Show, be switched to display panel, if choosing a node in dynamic tree graph, the player that the node includes it can be shown herein
Between relation and situations such as risk factor；Each rectangle (being exactly node) on each tree is the equal of one and acts, such as 5300040
It is to receive an assignment.The numeral of band behind its colon, equivalent to above said option.It is classified as if 5300040 numbers are more than 6.4
One kind, it is classified as less than 6.4 another kind of.
The thumbnail of tree, Fig. 1 center sections, it can be used as from the point of view of time shaft, user can brush above and select different contractings
Sketch map, the tree of so top can also be restricted to these by the tree scope of frame choosing；
Accuracy rate/recall rate line chart, Fig. 1 the latter half, change over time, it is accurate that every decision tree is predicted
Rate/recall rate can also be changed over time, and these values are shown with line chart；
Detailed view visualization model, multiple row are mutually related inter-view, including：
Personal panel, as shown in Fig. 2 the feelings changed over time for showing individual player's behavior, action and suspicion degree
Condition.
The bar chart of some color is double-clicked on personal panel, then the action being sorted under this behavior it is detailed each
The line chart of kind action, show the situation that each action changes over time, as shown in Figure 3.
Panel is grouped, as shown in figure 4, for showing that (for example one group is plug-in player, and one group is normally to play by two groups of players
Family；Either chosen in radar distribution map two groups) distribution of the value of attribute situation about changing over time.
Initial data stores according to various different actions, daily every kind of one file of action (such as in January, 2017
1, log in daily record), the game one shares hundreds of action.Count within the period interested, everyone is when different
Between action on piece the frequency.The period for paying attention to here both can be the natural time (date Hour Minute Second) or
Playtime (what action is 1 grade, 2 grades ... done respectively, and what the corresponding frequency is).So each each timeslice of player
There is a corresponding characteristic vector.It is more additionally, due to amount of action all too, action has been carried out a classification (task is related,
Attribute is related, and fight is related, and article is related).This classification both can be that user oneself specifies, can also be some existing literatures
In some complete inducing classification methods.
Decision tree presents the flow of a decision-making, and from root node, down each nonleaf node is a bar judged
Whether part, some attribute that can judge it according to the example come in meet the condition of this non-leaf node.Example gets to leaf
After node, leaf node can provide the label of a classification, to tell you which kind of this example belongs to.Traditional decision tree
Training process, it is exactly constantly recursively to the leaf node of the decision tree generated, according to some indexs, (for example information increases
Benefit, Gini indexes etc.) determine the child node below this leaf node is used as division with what attribute.
Here the model used is a kind of Hoeffding adaptive decision-making trees (Hoeffding using Gauss discretization
adaptive tree with Gaussian discretization).This is an on-line Algorithm, and he utilizes Hoeffding
The characteristic on boundary so that decision tree can accomplish on-line training, i.e. data have carried out can and bring training, and only in decision tree
Using once；Rather than lot data are needed as traditional decision tree, each data will be used to judge splitting condition
Repeatedly.
Hoeffding definitions, a stochastic variable, scope R, true average deviates it after n independent observation
Estimate is not more than with probability 1- δWhen making and dividing of that attribute in one node of judgement
Wait, find that information gain is maximum and the two second largest attributes, calculate the difference of their information gains, if being more than this ∈,
It can guarantee that a positive tree node splitting effect.Such a boundary can be helped at only part or low volume data,
Training sets out, without that have to wait until that all data have all come just train complete.This decision tree is referred to as Hoeffding
Decision tree.
Had made some improvements on the basis of this method, and using prior art.Firstly, since Hoeffding decision trees
Discrete value attribute is only supported, employs sane increment type Gauss discretization method so that supports Continuous valued attributes.Secondly, number
The characteristics of concept drift in also be present, so-called concept drift refers to such a case：Data generation may not be to put down
Steady, its generating process may change.Correspond in game data, some changes be able to may also occur for plug-in behavior
Change, it is outer to prevent from continuing being sealed off because plug-in company can perceive the plug-in of oneself and change some features by after title
Hang.On the other hand, ADapative WINdowing (ADWIN) technology is employed, Hoeffding decision-making of this method in script
A window and corresponding estimator are added in tree, changes detector etc., to find that concept drift above-mentioned shows
As.
So whole method, just it is referred to as the Hoeffding adaptive decision-making trees using Gauss discretization.Data are utilized
This method, as data constantly flow into, decision tree can constantly grow, when detecting some significant changes, decision tree
Some subtrees can replace, become other stalk tree.Then a tree changed over time has been obtained, and when each
Between on piece, a people will have one to be judged as YES plug-in or normal person judged result (0 normal player, 1 plug-in object for appreciation
Family).
According to model above, and in each timeslice, a people will have one be judged as YES it is plug-in either
The judged result of normal person.Suspicion degree just uses the average value of the value of all judged results of current time piece.A such as people
It is 0,0,0,0,1,1,0 in the judged result of timeslice 1 to 7, then he is set to 2/7 in the suspicion degree of the 7th timeslice.
Recommending to represent a kind of information with the form in the recommendation label in display panel, each row, accuracy rate, recall
Rate, accurate rate and frequency of occurrence etc.；Some node in the icicle figure chosen is represented per a line.Grey strip length in form
It is proportional with relative size of its value in this row.Click form new line, i.e. accuracy rate, recall rate, accurate rate and appearance
The frequency, can be with the sequence by column progress from high to low or from low to high.
Recommend and display panel in displaying label in radar map, have many points in radar distribution map, each put generation
State of one player of table in current time.Each various operating frequencies of player mark (0 as multi-C vector, and with one
Or 1,0 represents normal player, and 1 represents plug-in).Then these multi-C vectors are projected on the disc of two dimension, played for showing
Relation between family, and the degree of danger of player.On this disc, the Euclidean distance between player keeps original multidimensional as far as possible
The distance of vector；And player subtracts suspicion degree to the distance in the center of circle with 1 represents that (suspicion degree is the value between 0~1, for current
Time segment model exports the average of player indicia value).
A kind of such design is the design for having used for reference military radar in fact, because radar is more dangerous closer to the center of circle
's.It is this to design the design for having used for reference military radar.
Personal panel in detailed view visualization model, everyone with a rectangle frame and its internal bar each period
Shape figure represents that change information is represented with a full line over time for everyone.One of each bar chart is represented under a kind of behavior
The quantitative value of various actions adds up, and each bar chart has 4 to represent player's behavior to be divided into 4 types herein.
The bar chart of some color is double-clicked on personal panel, then the action being sorted under this behavior it is detailed each
The line chart of kind action, show the situation that each action changes over time.
In panel is grouped, often goes and represent a kind of action, the situation that two groups of players change over time, every group of player is one
Individual horizontal bar shaped, the left end right-hand member of this bar shaped respectively show the minimax value of certain value acted of this group of player.Exhibition
The section of existing maximum and minimum value.
Step 2：User mutual
Focus and context:Choosing is brushed on tree thumbnail (time shaft), above dynamic tree indication range can also change therewith.
Mouse frame selects several trees in dynamic tree graph, this several trees can by the expansion of transverse direction, and it is not selected can laterally be reduced, this
Sample prescription just observes details.And panel is recommended to show the various information of the node in the tree chosen therewith.In radar distribution map
After middle double-click, local radial amplification mode is opened；Point around mouse can be exaggerated in radial direction, and other point can be corresponded to
Be compressed.
When mouse-over is on the node in some tree, the node consistent with it can be linked up with line
Ash point is clicked in personal panel can shrink and (flatten grey belt).When bar chart deploys, certain is double-clicked
The bar chart of color, the situation of everything under this behavior of more details can be showed, represented with corresponding color line chart.Institute
There is action all to be indicated with the id of the action of beginning.
Search：In the upper right corner, search box can input the id (multiple centres are separated with comma) of one or more player,
Then can be shown in lower left corner individual's panel.
Filtering：In personal panel, some color bar chart is clicked, the bar shaped ash of remaining color can be darkened, current
Color is constant, facilitate the simple more this behavior of analysis personnel change over time and people and people between relation.
Pull：In personal panel, compare for convenience, small ash point can be pinned and pulled, position is exchanged, carry out comparative analysis people
Member two or more players interested.
View linkage：When mouse-over is when on node, node can also be shown corresponding to the display panel on the right.
Amplification：When left button clicks a node in dynamic tree graph, corresponding personal and packet panel can all be shown；
Radar distribution map occurs in display panel simultaneously.
Interacted with model：System supports the function-beta pruning interacted with model.Decision tree beta pruning is a common interaction,
By stopping the division of some nodes, to control the process that tree grows.
Step 3：Analysis task
1st, show the process of the dynamic evolution of decision tree, show the flow of the decision-making of decision tree, how about a people is judged to
Break to be plug-in or non-plug-in.
2nd, varigrained player's behavior, the evolution of action are analyzed.
3rd, have for analysis personnel interaction and timely feed back, prompt, and the information of appropriate context.According to these letters
Breath, can help user to find some pattern Producing reasons.
4th, analysis personnel can interact in itself with model.The professional knowledge of analysis personnel should be added to Analysis of Policy Making
During go, for adjusting this model.Also help to analyze other patterns simultaneously.
Actual mechanical process
Initially enter model training.
After having trained, a series of tree can be produced, is sequenced from left to right in dynamic tree graph.
Mouse-over is in tree node, it can be seen that presence of the same node point in different trees；Frame selects one piece of region, it can be seen that
Selected tree is exaggerated, and surrounding is set and reduced.Tree has the small tree thumbnail of a row below, it can be worked as time shaft, had
Individual telescopic selection of time frame, help to choose the tree to show in dynamic tree graph.
Exaggerated tree is chosen for frame, the right is recommended panel to show the various indexs of tree interior joint, such as accurate rate, called together
The rate of returning, (computational methods are F1) and frequency of occurrence etc..
If clicking the node in some tree, the display panel for having side occurs radar distribution map, shows and go out in node
Existing one player relationship information of player and player's risk factor information.Simultaneously, personal panel and packet panel can also update.
It shows and had been introduced above content.In use, because time shaft is alignment, it is very convenient by comparison.Each layer simultaneously
Secondary details can all show.Simultaneously the object for appreciation in display area can be then corresponded in radar distribution map chosen area, personal panel
Family's information.
Packet panel typically shows the distribution of plug-in in the personal panel and frequency under each action of regular player
Aggregation information.If but clicked in personal panel and compare button, can in radar map, frame selects one piece again, then correspond in individual
Also there is the detailed view of another a group of people in panel, and below figure dotted box portion is compared at this.
The first detects plug-in method：
Action is all chosen in control panel first, and these actions have been divided into 4 classes.After model running, it can be seen that
A series of tree, the structure of these trees are developing over time.Tree can slowly grow, if it is bad to have arrived certain phase effect
The subtree for having part is replaced.Some information, the possibility strategy of such as plug-in its behavior pattern of change can be obtained from above.
It is suspended in by mouse on tree node and sees those red lines, it can be found that some nodes is that occur always from the beginning to the end
, some are that a period of time just disappearance only occur.High-rise node can typically last long, and it is relatively good to illustrate these
Available for the attribute of differentiation, for example receive an assignment, task is often what is attained wealth, and attaining wealth contributes to plug-in profit
So plug-in tend to largely receive an assignment, so this respect might have obvious differentiation effect.
A node is selected, that continues one section of disappearance, caused by under thinking being what reason.Clicking node can be on the right
Display panel shows radar distribution map, as shown in Figure 8, it is seen that has 2 cluster points above, (so-called cluster is just collected on one
Point).The interaction for opening partial enlargement can be double-clicked, then this two clusters point is studied.
Personal panel comparing function is opened, after looking at outside this two batches that hanging over some time point starts, is had not quite alike
Behavior and action change with time pattern.Be to discriminate between originally on this node plug-in and normal person (acquiescence is only just
Ordinary person and plug-in), due to new plug-in appearance, this separating capacity declines.Similar conclusion can be obtained in group's panel, it is such as plug-in
Covered with normal person section.
Make such a guess:Because there are two clusters plug-in, there is the plug-in of less identical behavior pattern in other words, newly enter
The plug-in pattern of office may be not quite alike, and as new Incoming is plug-in more and more, the action that can be used for distinguishing originally can
It is able to can fail (information gain is not much of that) gradually, so causing such case.
The plug-in method of second of detection：
All concentrated on for that group plug-in among radar, it appears that color is very deep.Because these points are all transparencies,
So must be stack just can color it is very deep.Illustrate to have accumulated many players here.
Either from the point of view of projection result, or (the personal face from the point of view of actual selection data (being selected in radar map upper frame)
The column diagram of plate) all it is highly consistent, plug-in conduct Bulk Product is still more approximate.The folding of personal panel is just observed at last
Line chart can also obtain similar conclusion；And it distributed for normal player is relatively just more scattered.
Here reason is plug-in usually Bulk Product, so each side is more similar and in accordance with convention.Because it is
Seeking results maximizations, can it produce in batches when plug-in design.And comparatively, relatively just more scattered point of normal player
Cloth.
Claims (8)
- A kind of 1. system that plug-in detection is carried out using increment decision-making tree, it is characterised in that including：Data preprocessing module, is cleaned and characteristic vector pickup to the initial data of player actions；Model generates and interactive module, carrys out generation model using the characteristic vector of the data preprocessing module as the input of model And export, while receive feedback and model is adjusted；High level view visualization model, dynamic tree graph, recommendation and exhibition are generated according to the output model of model generation and interactive module Show panel and accuracy rate/recall rate line chart；The dynamic tree graph shows the tree construction of decision tree using icicle figure, and each node represents one inside the icicle figure The split vertexes of decision tree, multiple icicle figures are placed changed with time with embodying decision tree side by side；In the recommendation and display panel, the ice chosen in the dynamic tree graph is illustrated in the form that can be sorted recommending panel The information of all nodes of post figure；If choosing a node in dynamic tree graph in display panel, with the form of radar distribution map Show the situation that player is included in this node；The accuracy rate/recall rate line chart and the dynamic tree graph arrange up and down according to corresponding time relationship, represent with when Between change every decision tree prediction accuracy rate/recall rate.
- 2. the system of plug-in detection is carried out using increment decision-making tree as claimed in claim 1, it is characterised in that the radar point There are many points in Butut, each point represents state of the player in current time, each various operating frequency conducts of player Multi-C vector, and whether be plug-in mark with one；These multi-C vectors are projected on the disc of two dimension, for showing relation between player, and the degree of danger of player；On the disc, Euclidean distance between player keeps the distance of original multi-C vector, and player is to the distance in the center of circle Suspicion degree is subtracted with 1 to represent.
- 3. the system of plug-in detection is carried out using increment decision-making tree as claimed in claim 2, it is characterised in that the point is circle Point, provided with transparency.
- 4. the system of plug-in detection is carried out using increment decision-making tree as claimed in claim 1, it is characterised in that the high level regards Figure visualization model also includes the thumbnail of tree, as the thumbnail of the dynamic tree graph, is arranged in the dynamic tree graph and institute State between accuracy rate/recall rate line chart.
- 5. the system of plug-in detection is carried out using increment decision-making tree as claimed in claim 1, it is characterised in that also including details View visualization model, including personal panel and packet panel；The personal panel is used to show the situation that individual player's behavior, action and suspicion degree change over time；The packet panel is used for the situation for showing that the distribution of the value for the two groups of player attributes chosen changes over time.
- 6. the system of plug-in detection is carried out using increment decision-making tree as claimed in claim 5, it is characterised in that the personal face Everyone is represented each period with a rectangle frame and its internal bar chart in plate, everyone over time change information with one Full line represents that a quantitative value for representing various actions under a kind of behavior of each bar chart adds up.
- 7. the system of plug-in detection is carried out using increment decision-making tree as claimed in claim 6, it is characterised in that the personal face The bar chart of some color is double-clicked on plate, then displaying is sorted in the broken line of each detailed action of the action under this behavior Figure, show the situation that each action changes over time.
- 8. the system of plug-in detection is carried out using increment decision-making tree as claimed in claim 5, it is characterised in that in the packet In panel, often going and represent a kind of action, the situation that two groups of players change over time, every group of player is a horizontal bar shaped, Show the section of maximum and minimum value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711045371.1A CN107866072B (en) | 2017-10-31 | 2017-10-31 | System for detecting plug-in by adopting incremental decision tree |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711045371.1A CN107866072B (en) | 2017-10-31 | 2017-10-31 | System for detecting plug-in by adopting incremental decision tree |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107866072A true CN107866072A (en) | 2018-04-03 |
| CN107866072B CN107866072B (en) | 2020-06-16 |
Family
ID=61753546
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711045371.1A Active CN107866072B (en) | 2017-10-31 | 2017-10-31 | System for detecting plug-in by adopting incremental decision tree |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107866072B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108905209A (en) * | 2018-06-11 | 2018-11-30 | 网易（杭州）网络有限公司 | Plug-in detection method and system, electronic equipment and storage medium in game |
| CN111558226A (en) * | 2020-04-28 | 2020-08-21 | 腾讯科技(成都)有限公司 | Method, device, equipment and storage medium for detecting abnormal operation behaviors |
| CN115208737B (en) * | 2022-07-08 | 2023-07-04 | 电子科技大学 | Communication method based on maximum distance separable code and intelligent reflecting surface |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101187959A (en) * | 2006-11-17 | 2008-05-28 | 中兴通讯股份有限公司 | Game cheat detection method based on decision tree |
| CN105930934A (en) * | 2016-04-27 | 2016-09-07 | 北京物思创想科技有限公司 | Prediction model demonstration method and device and prediction model adjustment method and device |
| US20170032026A1 (en) * | 2011-11-04 | 2017-02-02 | BigML, Inc. | Interactive visualization of big data sets and models including textual data |
| CN106991425A (en) * | 2016-01-21 | 2017-07-28 | 阿里巴巴集团控股有限公司 | The detection method and device of commodity transaction quality |
| CN107281755A (en) * | 2017-07-14 | 2017-10-24 | 网易（杭州）网络有限公司 | Construction method, device, storage medium, processor and the terminal of detection model |
-
2017
- 2017-10-31 CN CN201711045371.1A patent/CN107866072B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101187959A (en) * | 2006-11-17 | 2008-05-28 | 中兴通讯股份有限公司 | Game cheat detection method based on decision tree |
| US20170032026A1 (en) * | 2011-11-04 | 2017-02-02 | BigML, Inc. | Interactive visualization of big data sets and models including textual data |
| CN106991425A (en) * | 2016-01-21 | 2017-07-28 | 阿里巴巴集团控股有限公司 | The detection method and device of commodity transaction quality |
| CN105930934A (en) * | 2016-04-27 | 2016-09-07 | 北京物思创想科技有限公司 | Prediction model demonstration method and device and prediction model adjustment method and device |
| CN107281755A (en) * | 2017-07-14 | 2017-10-24 | 网易（杭州）网络有限公司 | Construction method, device, storage medium, processor and the terminal of detection model |
Non-Patent Citations (2)
| Title |
|---|
| 孟辉等: "基于多元图形特征融合原理的降维方法研究", 《燕山大学学报》 * |
| 李彦龙等: "树比较可视化方法综述", 《软件学报》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108905209A (en) * | 2018-06-11 | 2018-11-30 | 网易（杭州）网络有限公司 | Plug-in detection method and system, electronic equipment and storage medium in game |
| CN111558226A (en) * | 2020-04-28 | 2020-08-21 | 腾讯科技(成都)有限公司 | Method, device, equipment and storage medium for detecting abnormal operation behaviors |
| CN111558226B (en) * | 2020-04-28 | 2023-04-18 | 腾讯科技(成都)有限公司 | Method, device, equipment and storage medium for detecting abnormal operation behaviors |
| CN115208737B (en) * | 2022-07-08 | 2023-07-04 | 电子科技大学 | Communication method based on maximum distance separable code and intelligent reflecting surface |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107866072B (en) | 2020-06-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109464807A (en) | Detect game plug-in method, apparatus and terminal | |
| CN106202316A (en) | Merchandise news acquisition methods based on video and device | |
| CN107866072A (en) | A kind of system that plug-in detection is carried out using increment decision-making tree | |
| CN108351985A (en) | Method and apparatus for large-scale machines study | |
| JP2004529406A (en) | Method and apparatus for dynamic real-time market segmentation | |
| CN106951471A (en) | A kind of construction method of the label prediction of the development trend model based on SVM | |
| CN109598278A (en) | Clustering processing method, apparatus, electronic equipment and computer readable storage medium | |
| Wilkinson | Classification and regression trees | |
| CN110232331A (en) | A kind of method and system of online face cluster | |
| Schertler et al. | User guided movement analysis in games using semantic trajectories | |
| US20080244463A1 (en) | Method and apparatus for providing a presumptive drafting solution | |
| Xiao et al. | R&b: Region and boundary aware zero-shot grounded text-to-image generation | |
| Camacho | Bio-inspired Clustering: basic features and future trends in the era of Big Data | |
| CN105354720B (en) | A method of mixed recommendation is carried out to consumption place based on visual cluster | |
| Carroll et al. | Assessing the Height of Mobility Barriers: A Methodology and an Empirical Test in the UK Retail Grocery Industry 1 | |
| KR102643159B1 (en) | A matching method that finds empty space in lcl containers in real time during container import and export | |
| Rajesh et al. | Player recommendation system for fantasy premier league using machine learning | |
| US20040006567A1 (en) | Decision support system using narratives for detecting patterns | |
| CN110175191B (en) | Modeling method for data filtering rule in data analysis | |
| Adilah et al. | An augmented method of selecting fashion talent by adding social media characteristic | |
| Xie et al. | CausalFlow: Visual analytics of causality in event sequences | |
| García-Saravia et al. | Application of an eye tracker over facility layout problem to minimize user fatigue | |
| US20230008377A1 (en) | Information processing device, control method, and storage medium | |
| Horton | Algorithms for the analysis of spatio-temporal data from team sports | |
| Kim et al. | Engaging in sports via the metaverse? An examination through analysis of metaverse research trends in sports |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |