CN103530540B - User identity attribute detection method based on man-machine interaction behavior characteristics - Google Patents
User identity attribute detection method based on man-machine interaction behavior characteristics Download PDFInfo
- Publication number
- CN103530540B CN103530540B CN201310454565.2A CN201310454565A CN103530540B CN 103530540 B CN103530540 B CN 103530540B CN 201310454565 A CN201310454565 A CN 201310454565A CN 103530540 B CN103530540 B CN 103530540B
- Authority
- CN
- China
- Prior art keywords
- man
- identity attribute
- user
- machine interaction
- attribute
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/30—Authentication, i.e. establishing the identity or authorisation of security principals
- G06F21/31—User authentication
- G06F21/32—User authentication using biometric data, e.g. fingerprints, iris scans or voiceprints
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2221/00—Indexing scheme relating to security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F2221/21—Indexing scheme relating to G06F21/00 and subgroups addressing additional information or applications relating to security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F2221/2133—Verifying human interaction, e.g., Captcha
Abstract
The invention discloses a user identity attribute detection method based on man-machine interaction behavior characteristics. The man-machine interaction behavior characteristics are extracted through analysis of man-machine interaction behaviors generated when man-machine interaction equipment, such as a mouse, a keyboard and a touch screen, operates in an interaction process between a user and an intelligent computer system, a user identity attribute template is established based on the man-machine interaction behavior characteristics, and identity attributes comprising genders, ages, races and the like of the user are detected and distinguished. The user identity attribute detection method has the advantages that the man-machine interaction behavior characteristics fill up blanks in analysis of identity attributes of operators in the intelligent computer system, and a brand new thought is provided for information perception analysis of users of computers and mobile networks. In addition, the user can be continuously analyzed through the method in the interaction process between the user and the intelligent computer system, and normal behaviors of the user cannot be disturbed.
Description
Technical field
The present invention relates to a kind of computer and mobile network user information Perception analytical technology, particularly to a kind of based on meter
Calculation machine and the identity attribute detection method of smart phone user man-machine interaction behavior characteristicss.
Background technology
With the propulsion of social informatization, networking spring tide, the perception to user profile in computer and mobile network
Analysis becomes more and more important.On the one hand, in the network virtualization economic activity such as ecommerce, Internet bank, businessman is urgent
Want to sufficiently understand client, to provide targetedly commodity or service thus improving the success rate of business activity as far as possible;
On the other hand, computer network and mobile network information criminal activity are also increasingly severe, extract and analysis is present in calculating net
The identity attribute such as the sex of the electronic evidence in network system and then determination operator, age, race, language can be the network crime
The discovery of activity and containment provide important help.
In recent years, research worker is had to propose to detect information or the identity attribute of user based on biological characteristic, they are according to people
The physiological features such as face, fingerprint, iris, palmmprint detect to information such as the sex of user, age, races, but such method
Need using specific biomedical information acquisition equipment, such as photographic head, fingerprint sensor etc., it is not suitable for existing calculating network ring
Border.There is presently no can in existing calculating network environment large-scale application analysis detect user identity attribute technology
Or method.
For the demand, the present invention is proposed a kind of analysis based on man-machine interaction behavior characteristicss and detects user identity attribute
Techniques or methods.
Content of the invention
It is an object of the invention to provide a kind of computer based on man-machine interaction behavior characteristicss and smart phone user identity
Detection of attribute technology, in particular with interbehavior feature produced during user operation human-computer interaction device as foundation
Method to detect the identity attribute of operator.
For reaching object above, the present invention adopts the following technical scheme that and is achieved:
A kind of user identity attribute detection method based on man-machine interaction behavior characteristicss is it is characterised in that include setting up body
Part attribute model and detection two parts of identity attribute:
(1)Set up identity attribute model, comprise the steps:
The first step, during computer and smart phone user are normally using human-computer interaction device, gather and records
The man-machine interaction behavioral data of user, including mouse interbehavior data, keystroke interbehavior data, touches interbehavior number
According to;
Second step, for all computers and smart phone user, defines the species of identity attribute, and every kind of identity attribute
Division classification;
3rd step, was divided to the man-machine interaction behavioral data of record with fixed observer time span T for the cycle, was formed
Multiple time spans are the man-machine interaction behavioral data block of T;The species of the identity attribute according to user and every kind of identity attribute
Divide and these behavioral data blocks are marked;
4th step, for the data block of each labelling, extracts and labelling man-machine interaction behavior characteristicss vector, by different pieces of information
Man-machine interaction behavior characteristicss vector combination in block forms the identity attribute characteristic vector training set of user;
5th step, for every kind of identity attribute, according to the labelling of identity attribute species, obtains each identity attribute corresponding
Characteristic vector training set;Divide the labelling of classification according to each identity attribute, obtain the labelling of each characteristic vector in training set;
Respectively using corresponding for each identity attribute characteristic vector training set as training sample, simultaneously by the mark of characteristic vector in training set
It is denoted as the labelling for training sample, identity attribute model is built respectively to each identity attribute.
(2)Detection identity attribute, comprises the steps:
The first step, after user logins computer or smart mobile phone, the man-machine interaction behavior of capture active user, with length T
For the cycle, obtain in T user's man-machine interaction behavioral data and extract corresponding man-machine interaction behavior characteristicss vector, so generate right
Answer the characteristic vector of each identity attribute;
Second step, detects to the identity attribute of active user:Using the identity attribute characteristic vector generating as built
The input of vertical identity attribute presumption model, obtains the detected value of user identity attribute, and the identity attribute of user is judged.
In said method, the described identity attribute characteristic vector setting up formation user in identity attribute model part the 4th step
The comprising the following steps that of training set:
(1)In the man-machine interaction behavioral data block for T for the observation interval, travel through man-machine interaction sequence of events, successively
Isolate different types of interbehavior event, including mouse interbehavior event, keystroke interbehavior event, touch interaction row
For event;
(2)For different types of interbehavior event, extract interbehavior characteristic vector, including mouse behavior characteristicss to
Amount, keystroke behavior characteristicss vector, touch behavior characteristicss vector;
(3)Man-machine interaction behavior characteristicss vector in different pieces of information block is combined, formed identity attribute feature to
Amount training set.
The man-machine interaction behavioral data of described computer or smart phone user generation is basic man-machine interaction event composition
Sequence, the form of basic man-machine interaction event is:{ interaction time stabs, interactive screen position, including mouse, keyboard or touch
The interactive device type of plate, alternative events type }.
Described identity attribute refers to computer and the intrinsic physiology of smart phone user or behavioral trait, including user's
Sex, age, race, language, right-hand man's use habit, schooling, computer use proficiency level, occupation, finger health
Situation.
Described identity attribute model of setting up is realized by one or more Classifier combination, and described grader includes Weighted random
Forest classified device, artificial nerve network classifier, support vector machine classifier.Wherein, set up by Weighted random forest classified device
The concretely comprising the following steps of identity attribute model:
1)It is N that initialization training sample concentrates the number of feature samples, and in each feature samples, the number of characteristic component is
M, the number of decision tree is P, and the number of the decision-making feature of each decision tree is m, and m is much smaller than M;
2)In order to eliminate the impact of different feature dimensions, each dimensional feature vector is normalized, by its value
It is limited between [0,1];
3)Using Bagging algorithm, N number of feature samples are sampled P time, obtain P characteristic set;
4)Each random tree is randomly selected with a characteristic set, and this decision tree is estimated and error analyses,
For each of tree node, randomly choose the m characteristic component put based on this, and for different classes of feature samples,
Give different weights to find optimal partitioning scheme;
5)According to classifying quality best characteristic node node division is Liang Ge branch, then recursive call step 4)Until
This tree can Accurate classification training sample set, or all properties were used;After the complete growth of decision tree, not right
It carries out beta pruning;
6)Repeat step 3)、4)、5)Until establishing whole P decision trees;
7)Using the most methods voted based on weighting Lai the comprehensive classification results determining multiple decision trees, that is, added
The classification results of power random forest grader.
The present invention describes the behavior spy that user embodies in interactive process in the form of man-machine interaction sequence of events
Property, the identity attribute of operator is detected with this, be computer and the analysis of mobile network user information Perception provide a kind of complete
New thinking.Its advantage is:First, identity attribute analysis desired data can directly obtain from interactive process, need not join
Standby extra instrument and equipment;Secondly, identity attribute analysis is based on man-machine interaction behavior characteristicss, need not remember or carry,
It is difficult to be imitated and forge;In addition, can persistently catch during computer user and smart phone user operation equipment
Obtain human-machine interactive information produced by user operation, therefore can be based on man-machine interaction behavior characteristicss persistently to user identity attribute
Carry out discriminatory analysis, and the normal behaviour without interference with user, there is extensive safety and the suitability.
Brief description
With reference to the accompanying drawings and detailed description the present invention is described in further detail.
Fig. 1 is the step block diagram of the inventive method.
Fig. 2 is the identity attribute feature generation step block diagram of man-machine interaction behavior in the inventive method.
Fig. 3 is the identity attribute method for establishing model step block diagram in the inventive method based on Weighted random forest.
Fig. 4 is the experimental result picture detecting computer user's identity attribute using the inventive method.The mistake of in figure black
Bar represents the standard deviation of the identity attribute accuracy rate after 20 random data samplings.
Specific embodiment
System structure
Referring to Fig. 1, the computer based on man-machine interaction behavior characteristicss for the present invention and the detection of smart phone user identity attribute
Method, can be used for user identity attribute aware in the e-commerce initiatives such as electronic emporium, the Internet bank, to provide targetedly business
Product or service;Can also be used for the information forensics analysis in enterprise information system, safeguard protection is carried out to important information system.This
Bright comprise to set up identity attribute model and identity attribute detects two parts, specific implementation steps are as follows:
1)Set up identity attribute model part to comprise the steps:
(1)During computer and smart phone user are normally using human-computer interaction device, gather and record user
Man-machine interaction behavioral data, including mouse interbehavior data, keystroke interbehavior data, touch interbehavior data, enter
And form identity attribute model and set up required interbehavior data set;The form of basic human-machine interaction data is:{ interaction time
Stamp, interactive screen position, type of interaction, alternative events type }, type of interaction includes mouse interaction, keyboard mutuality and touch screen
Interaction, alternative events type includes click and moving event, keyboard key stroke event, finger pressing on the touchscreen and touches
Touch moving event;
(2)For all computers and smart phone user, define the species of identity attribute, and the drawing of every kind of identity attribute
Sub-category;
(3)For the cycle, the man-machine interaction behavioral data of record is divided with fixed observer time span T, formed multiple
Time span is the man-machine interaction behavioral data block of T;The species of the identity attribute according to user and the division of every kind of identity attribute
These behavioral data blocks are marked;
(4)For the data block of each labelling, extract and labelling man-machine interaction behavior characteristicss vector, special including mouse behavior
Levy vector, keystroke behavior characteristicss vector, touch behavior characteristicss vector.Wherein mouse behavior characteristicss vector refers to be produced by mouse is mobile
A series of behavior measure amounts that the operation such as raw space-time geometric locus and click is derived, it is possible to use once move
The pass in geometric locus, rate curve, the relation of accelerating curve, average translational speed and distance, average translational speed and direction
Relation, motion track distance and the displacement in the relation of system, average translational acceleration and distance, average translational acceleration and direction
Ratio is as feature;Keystroke behavior characteristicss vector refers to that the time serieses pressed by each key of keyboard and the event of upspringing is formed derive
A series of behavior measure amounts obtaining, it is possible to use the persistent period of singly-bound button and the interval time of adjacent key are as spy
Levy;Touch behavior characteristicss vector and refer to that finger moves the operation such as space-time geometric locus and pressing of generation on the touchscreen and derived
A series of behavior measure amounts going out, it is possible to use screen touch pressure, touch click time, touch motion track, touch are moved
Rate curve, touch traveling time are as feature.Man-machine interaction behavior characteristicss vector in different pieces of information block is combined,
Form the identity attribute characteristic vector training set of user;
(5)For each identity attribute(The sex of user, age, race, language, right-hand man's use habit, cultural journey
Degree, computer use the attribute such as proficiency level, occupation, finger health status), according to the labelling of identity attribute species, obtain every
The corresponding characteristic vector training set of individual identity attribute;Divide the labelling of classification according to each identity attribute, obtain every in training set
The labelling of individual characteristic vector;Respectively using corresponding for each identity attribute characteristic vector training set as training sample, will instruct simultaneously
Practice the labelling concentrating characteristic vector as the labelling of training sample, each identity attribute is built respectively based on Weighted random forest
Identity attribute detection model., by the characteristic vector training with Sex-linked marker taking the gender attribute detection model of user as a example
Collection, as the training data of model, the detection of gender attribute is considered as two classification problems(Man or female), thus build being based on
The gender attribute detection model of man-machine interaction behavior.
2)Identity attribute detection part comprises the steps:
(1)During user uses the intelligence system such as computer or smart mobile phone, the man-machine friendship of capture active user
Mutually behavior, with length T(T typically can be set to 30 seconds or the longer time)For the cycle, obtain in T user's human-machine interaction data and carry
Take behavior characteristicss, generate identity attribute characteristic vector;
(2)Using the identity attribute characteristic vector generating as the input of identity attribute detection model, obtain user identity and belong to
The detected value of property, by the threshold epsilon of this detected value and corresponding identity attribute model(ε is chosen according to the precision of model training,
Typically may be set to 50%)It is compared, differentiate the corresponding identity attribute of user.So that the gender attribute of user detects as a example, will be from
The correspondence other identity attribute vector extracting in the T time cycle, as the input of the gender attribute detection model of foundation, obtains
To the detected value of this model, detected value is compared with corresponding threshold value, if detected value is more than threshold value, judges active user
Sex be male;If detected value is less than threshold value, judge the sex of active user as women.
Identity attribute detection model based on Weighted random forest
Above-mentioned 1)Set up the of identity attribute model part(5)In step, the identity attribute based on Weighted random forest detects mould
Type sets up process referring to Fig. 3, comprises the following steps that:
(1)It is N that initialization training characteristics concentrate the number of feature samples, and in each feature samples, the number of characteristic component is
M, the number of decision tree is P, and the number of the decision-making feature of each decision tree is m (m is much smaller than M);
(2)In order to eliminate the impact of different feature dimensions, each dimensional feature vector is normalized, by its value
It is limited between [0,1];
(3)Using Bagging algorithm, N number of feature samples are sampled P time, obtain P characteristic set;
(4)Each random tree is randomly selected with a characteristic set, and this decision tree is estimated and error analyses.
For each of tree node, randomly choose the m characteristic component put based on this, and for different classes of feature samples,
Give different weights to find optimal partitioning scheme;
(5)According to classifying quality best characteristic node node division is Liang Ge branch, then recursive call step(4)Directly
To this tree can Accurate classification training sample set, or all properties were used;After the complete growth of decision tree, no
Beta pruning is carried out to it;
(6)Repeat step(3)、(4)、(5)Until establishing whole P decision trees;
(7)Using the most methods voted based on weighting Lai the comprehensive classification results determining multiple decision trees, that is, obtain
The classification results of Weighted random forest classified device.
The system of selection of decision-making characteristic variable number, the description of optimal partitioning scheme
The of " the identity attribute detection model based on Weighted random forest "(1)The choosing of decision-making characteristic variable number m in step
Select and refer to that in every decision tree of construction be need to randomly select m dimensional feature from feature samples, and choose classification in this m dimensional feature
The best characteristic node of effect.In the construction process of whole random forest, m is a constant, and we choose m=int (log2m+
1), wherein int is bracket function.
The(4)In step, optimal partitioning scheme is to instigate the categorical data on each node to be derived from same category as far as possible,
So that the impurity level of each node reaches the partitioning scheme of minimum(When the categorical data in certain node i is all from same
Classification, then the impurity level of this node is 0).During every decision tree construction, its generation is followed top-down recurrence and is divided
Split principle, start successively training set to be divided from root node.For each node, former according to node impurity level minimum
Then, it is split into left sibling and has node, they comprise a subset of training data respectively, so that node is continued according to same rule
Continuous division, until branch stops growing.If the categorical data in node i is both from same category, impurity level I of this node
(i)=0.The measure of impurity level is based on Gini impurity level criterion, that is, assume P (wj) it is that w is belonged on node ijClass sample
Number accounts for the frequency of training sample sum, then Gini impurity level criterion is expressed as:
Decision method based on weighted majority ballot
The of " the identity attribute detection model based on Weighted random forest "(7)The side of the most ballots based on weighting in step
Method refers to that the classification few to feature samples number gives bigger weights.In identity attribute detection process, with gender information's
As a example detection(2 class classification problems:Man or female), a sample x is through each decision tree classifier TiAfterwards, will produce 2 defeated
Go out result, be 2 confidence values, c ∈ { 1,2 }, each confidence levelp(f(x)=j) illustrate the probability that this sample x belongs to jth class
Value, the weighted value based on all decision tree results for the final judgement, it is shown below.
Wherein weights αiCircular be the category votes be multiplied by the repeated sampling for the category time
Number, and it is normalized.
Finally, the decision value obtaining is compared with decision-making value ε, the identity attribute of user is differentiated.If F >=
ε, then the sex of active user is male;If F<ε, then the sex of active user is women.Wherein, the selection of ε can be instructed in model
Adopt cross validation when practicing, be adjusted by the value being continually changing ε and optimize, to obtain preferable model training result.
The step that gender attribute is detected according to the present invention
The first step, defines the classification of gender attribute, in the present embodiment gender attribute is divided into 2 classes:The first kind is man
Property user;Equations of The Second Kind is female user.
Second step, captures man-machine interaction Behavioral training data.Man-machine to record for the cycle with fixed observer time span T
Interbehavior data is divided, and forms the man-machine interaction behavioral data block that multiple time spans are T, and the sex according to user
Attribute is marked to these behavioral data blocks.
3rd step, generates the features training collection of gender attribute.For the data block of each labelling, extract and the man-machine friendship of labelling
Mutually behavior characteristicss vector;Man-machine interaction behavior characteristicss vector in different pieces of information block is combined, is formed and belong to sex
The gender attribute characteristic vector training set of property labelling;Each characteristic vector that this gender attribute features training is concentrated can be whole
Man-machine interaction behavior characteristicss vector or the subvector of the vector of the man-machine interaction behavior characteristicss after feature selection.
4th step, sets up gender attribute detection model.The test problems of gender attribute are considered as the problem of 2 classification, with property
The corresponding characteristic vector training set of other attribute as training sample, simultaneously using the Sex-linked marker of each characteristic vector as training sample
This labelling, builds the identity attribute detection model based on Weighted random forest to gender attribute.
5th step, the real time human-machine interaction behavioral data of monitoring user simultaneously extracts behavior characteristicss vector.Access meter in user
When calculation machine or smart mobile phone, with the new man-machine interaction behavioral data of observation time T capture;From the man-machine interaction for T for the time span
The corresponding characteristic vector of extracting data gender attribute.
6th step, generates the testing result of gender attribute.Using the corresponding characteristic vector of gender attribute generating as built
The input of vertical gender attribute detection model, obtains the detected value of user's gender attribute;This detected value is carried out with decision-making value
Relatively, the gender attribute of user is judged.
The step that age attribute is detected according to the present invention
The first step, defines the classification of age attribute, in the present embodiment age attribute is divided into 3 classes:The first kind is year
The user that age is less than 30 years old;Equations of The Second Kind is user between 30 years old to 60 years old for the age;3rd class is the user more than 60 years old.
Second step, captures man-machine interaction Behavioral training data.Man-machine to record for the cycle with fixed observer time span T
Interbehavior data is divided, and forms the man-machine interaction behavioral data block that multiple time spans are T, and the age according to user
Attribute is marked to these behavioral data blocks.
3rd step, generates the features training collection of age attribute.For the data block of each labelling, extract and the man-machine friendship of labelling
Mutually behavior characteristicss vector;Man-machine interaction behavior characteristicss vector in different pieces of information block is combined, forms band has age and belong to
The age attribute characteristic vector training set of property labelling;Each characteristic vector that this age attribute features training is concentrated can be whole
Man-machine interaction behavior characteristicss vector or the subvector of the vector of the man-machine interaction behavior characteristicss after feature selection.
4th step, sets up age attribute detection model.The test problems of age attribute are considered as the problem of 3 classification, with year
Age attribute corresponding characteristic vector training set as training sample, simultaneously using the age indicator of each characteristic vector as training sample
This labelling, builds the identity attribute detection model based on Weighted random forest to age attribute.
5th step, the real time human-machine interaction behavioral data of monitoring user simultaneously extracts behavior characteristicss vector.Access meter in user
When calculation machine or smart mobile phone, with the new man-machine interaction behavioral data of observation time T capture;From the man-machine interaction for T for the time span
The corresponding characteristic vector of extracting data age attribute.
6th step, generates the testing result of age attribute.Using the corresponding characteristic vector of age attribute generating as built
The input of vertical age attribute detection model, obtains the detected value of age of user attribute;This detected value is carried out with decision-making value
Relatively, the age attribute of user is judged.
The step that linguistic property is detected according to the present invention
The first step, the classification of definitional language attribute, in the present embodiment linguistic property is divided into 2 classes:The first kind is English
Language is the user of mother tongue;Equations of The Second Kind is the user of mother tongue for non-english.
Second step, captures man-machine interaction Behavioral training data.Man-machine to record for the cycle with fixed observer time span T
Interbehavior data is divided, and forms the man-machine interaction behavioral data block that multiple time spans are T, and the language according to user
Attribute is marked to these behavioral data blocks.
3rd step, the features training collection of production language attribute.For the data block of each labelling, extract and the man-machine friendship of labelling
Mutually behavior characteristicss vector;Man-machine interaction behavior characteristicss vector in different pieces of information block is combined, is formed and belong to language
The linguistic property characteristic vector training set of property labelling;Each characteristic vector that this linguistic property features training is concentrated can be whole
Man-machine interaction behavior characteristicss vector or the subvector of the vector of the man-machine interaction behavior characteristicss after feature selection.
4th step, sets up linguistic property detection model.The test problems of linguistic property are considered as the problem of 2 classification, with language
Say attribute corresponding characteristic vector training set as training sample, simultaneously using the language tag of each characteristic vector as training sample
This labelling, builds the identity attribute detection model based on Weighted random forest to linguistic property.
5th step, the real time human-machine interaction behavioral data of monitoring user simultaneously extracts behavior characteristicss vector.Access meter in user
When calculation machine or smart mobile phone, with the new man-machine interaction behavioral data of observation time T capture;From the man-machine interaction for T for the time span
The corresponding characteristic vector of extracting data linguistic property.
6th step, the testing result of production language attribute.Using the corresponding characteristic vector of linguistic property generating as built
The input of vertical linguistic property detection model, obtains the detected value of user language attribute;This detected value is carried out with decision-making value
Relatively, the linguistic property of user is judged.
The step that right-hand man's use habit attribute is detected according to the present invention
The first step, defines the classification of right-hand man's use habit attribute, in the present embodiment by right-hand man's use habit attribute
It is divided into 2 classes:The first kind is the user with left hand for custom operation human-computer interaction device;Equations of The Second Kind is with the right hand for custom operation
The user that man-machine interaction sets.
Second step, captures man-machine interaction Behavioral training data.Man-machine to record for the cycle with fixed observer time span T
Interbehavior data is divided, and forms the man-machine interaction behavioral data block that multiple time spans are T, and the left and right according to user
Handss use habit attribute is marked to these behavioral data blocks.
3rd step, generates the features training collection of right-hand man's use habit attribute.For the data block of each labelling, extract simultaneously
Labelling man-machine interaction behavior characteristicss vector;Man-machine interaction behavior characteristicss vector in different pieces of information block is combined, is formed
Right-hand man's use habit attribute feature vector training set with right-hand man's use habit attribute labelling;This right-hand man's use habit
Each characteristic vector in attribute character training set can be for whole man-machine interaction behavior characteristicss vector or through feature
The subvector of the man-machine interaction behavior characteristicss vector after selection.
4th step, sets up right-hand man's use habit detection of attribute model.Test problems by right-hand man's use habit attribute
It is considered as the problem of 2 classification, using the corresponding characteristic vector training set of right-hand man's use habit attribute as training sample, simultaneously with every
Right-hand man's use habit labelling of individual characteristic vector, as the labelling of training sample, builds to right-hand man's use habit attribute and is based on
The identity attribute detection model of Weighted random forest.
5th step, the real time human-machine interaction behavioral data of monitoring user simultaneously extracts behavior characteristicss vector.Access meter in user
When calculation machine or smart mobile phone, with the new man-machine interaction behavioral data of observation time T capture;From the man-machine interaction for T for the time span
The corresponding characteristic vector of extracting data right-hand man's use habit attribute.
6th step, generates the testing result of right-hand man's use habit attribute.The right-hand man's use habit attribute pair that will generate
The characteristic vector answered, as the input of right-hand man's use habit detection of attribute model of foundation, obtains user right-hand man using habit
The detected value of used attribute;This detected value and decision-making value are compared, right-hand man's use habit attribute of user is sentenced
Disconnected.
The step that schooling attribute is detected according to the present invention
The first step, define schooling attribute classification, in the present embodiment by schooling Attribute transposition be 3 classes:The
One class is schooling in primary school and following user;Equations of The Second Kind is the user in junior middle school to senior middle school for the schooling;3rd class is
Schooling is in university and above user.
Second step, captures man-machine interaction Behavioral training data.Man-machine to record for the cycle with fixed observer time span T
Interbehavior data is divided, and forms the man-machine interaction behavioral data block that multiple time spans are T, and the culture according to user
Degree attribute is marked to these behavioral data blocks.
3rd step, generates the features training collection of schooling attribute.For the data block of each labelling, extract and labelling people
Machine interbehavior characteristic vector;Man-machine interaction behavior characteristicss vector in different pieces of information block is combined, is formed with literary composition
The schooling attribute feature vector training set of change degree attribute labelling;Each in this schooling attribute character training set is special
Levying vector can be whole man-machine interaction behavior characteristicss vector or the man-machine interaction behavior characteristicss after feature selection
The subvector of vector.
4th step, sets up schooling detection of attribute model.The test problems of schooling attribute are considered as asking of 3 classification
Topic, using schooling attribute corresponding characteristic vector training set as training sample, simultaneously with the cultural journey of each characteristic vector
Scale designation, as the labelling of training sample, builds the identity attribute based on Weighted random forest and detects mould to schooling attribute
Type.
5th step, the real time human-machine interaction behavioral data of monitoring user simultaneously extracts behavior characteristicss vector.Access meter in user
When calculation machine or smart mobile phone, with the new man-machine interaction behavioral data of observation time T capture;From the man-machine interaction for T for the time span
The corresponding characteristic vector of extracting data schooling attribute.
6th step, generates the testing result of schooling attribute.The corresponding characteristic vector of schooling attribute that will generate
As the input of the schooling detection of attribute model of foundation, obtain the detected value of user's schooling attribute;This is detected
Value is compared with decision-making value, and the schooling attribute of user is judged.
The step that computer uses proficiency level's attribute is detected according to the present invention
The first step, defines the classification that computer uses proficiency level's attribute, in the present embodiment by computer using skilled
Degree Attribute transposition is 3 classes:The first kind is very unskilled user(There is no corresponding human-computer interaction device using experience);Second
Class is general skilled user(Using corresponding human-computer interaction device between 1 month to 3 months);3rd class is very skilled
User(Using corresponding human-computer interaction device more than 3 months).
Second step, captures man-machine interaction Behavioral training data.Man-machine to record for the cycle with fixed observer time span T
Interbehavior data is divided, and forms the man-machine interaction behavioral data block that multiple time spans are T, and the calculating according to user
Machine is marked to these behavioral data blocks using proficiency level's attribute.
3rd step, generates the features training collection that computer uses proficiency level's attribute.For the data block of each labelling, carry
Take and labelling man-machine interaction behavior characteristicss vector;Man-machine interaction behavior characteristicss vector in different pieces of information block is combined,
Form the computer using proficiency level's attribute labelling with computer and use proficiency level's attribute feature vector training set;This meter
It can be whole man-machine interaction behavior characteristicss vector that calculation machine uses each characteristic vector in proficiency level's attribute character training set,
It can also be the subvector of the man-machine interaction behavior characteristicss vector after feature selection.
4th step, sets up computer and uses proficiency level's detection of attribute model.Computer is used proficiency level's attribute
Test problems are considered as the problem of 3 classification, proficiency level attribute corresponding characteristic vector training set are used as training using computer
Sample, is used proficiency level's labelling as the labelling of training sample using the computer of each characteristic vector simultaneously, computer is made
Build the identity attribute detection model based on Weighted random forest with proficiency level's attribute.
5th step, the real time human-machine interaction behavioral data of monitoring user simultaneously extracts behavior characteristicss vector.Access meter in user
When calculation machine or smart mobile phone, with the new man-machine interaction behavioral data of observation time T capture;From the man-machine interaction for T for the time span
Extracting data computer uses the corresponding characteristic vector of proficiency level's attribute.
6th step, generates the testing result that computer uses proficiency level's attribute.The computer generating is used skilled journey
The degree corresponding characteristic vector of attribute uses the input of proficiency level's detection of attribute model as the computer of foundation, obtains user
Computer uses the detected value of proficiency level's attribute;This detected value and decision-making value are compared, the computer of user is made
Judged with proficiency level's attribute.
The step that professional attribute is detected according to the present invention
The first step, the classification of the professional attribute of definition, is 2 classes by professional Attribute transposition in the present embodiment:The first kind is meter
The user of calculation machine working;The user that Equations of The Second Kind is obtained employment for non-computer.
Second step, captures man-machine interaction Behavioral training data.Man-machine to record for the cycle with fixed observer time span T
Interbehavior data is divided, and forms the man-machine interaction behavioral data block that multiple time spans are T, and the occupation according to user
Attribute is marked to these behavioral data blocks.
3rd step, generates the features training collection of professional attribute.For the data block of each labelling, extract and the man-machine friendship of labelling
Mutually behavior characteristicss vector;Man-machine interaction behavior characteristicss vector in different pieces of information block is combined, is formed and belong to occupation
The professional attribute feature vector training set of property labelling;Each characteristic vector in this professional attribute character training set can be whole
Man-machine interaction behavior characteristicss vector or the subvector of the vector of the man-machine interaction behavior characteristicss after feature selection.
4th step, sets up professional detection of attribute model.The test problems of professional attribute are considered as the problem of 2 classification, with duty
Industry attribute corresponding characteristic vector training set as training sample, simultaneously using the labelling of each characteristic vector as training sample
Labelling, builds the identity attribute detection model based on Weighted random forest to professional attribute.
5th step, the real time human-machine interaction behavioral data of monitoring user simultaneously extracts behavior characteristicss vector.Access meter in user
When calculation machine or smart mobile phone, with the new man-machine interaction behavioral data of observation time T capture;From the man-machine interaction for T for the time span
The extracting data occupation corresponding characteristic vector of attribute.
6th step, generates the testing result of professional attribute.Using the corresponding characteristic vector of professional attribute generating as built
The input of vertical professional detection of attribute model, obtains the detected value of user's occupation attribute;This detected value is carried out with decision-making value
Relatively, the professional attribute of user is judged.
The step that finger health attribute is detected according to the present invention
The first step, define finger health attribute classification, in the present embodiment by finger health Attribute transposition be 2 classes:The
One class is the user of finger health;Equations of The Second Kind is the user of finger non-health.
Second step, captures man-machine interaction Behavioral training data.Man-machine to record for the cycle with fixed observer time span T
Interbehavior data is divided, and forms the man-machine interaction behavioral data block that multiple time spans are T, and the finger according to user
Healthy attribute is marked to these behavioral data blocks.
3rd step, generates the features training collection of finger health attribute.For the data block of each labelling, extract and labelling people
Machine interbehavior characteristic vector;Man-machine interaction behavior characteristicss vector in different pieces of information block is combined, is formed and carry handss
Refer to the finger health attribute feature vector training set of healthy attribute labelling;Each in this finger health attribute character training set is special
Levying vector can be whole man-machine interaction behavior characteristicss vector or the man-machine interaction behavior characteristicss after feature selection
The subvector of vector.
4th step, sets up finger health detection of attribute model.The test problems of finger health attribute are considered as asking of 2 classification
Topic, using finger health attribute corresponding characteristic vector training set as training sample, is made with the labelling of each characteristic vector simultaneously
For the labelling of training sample, the identity attribute detection model based on Weighted random forest is built to finger health attribute.
5th step, the real time human-machine interaction behavioral data of monitoring user simultaneously extracts behavior characteristicss vector.Access meter in user
When calculation machine or smart mobile phone, with the new man-machine interaction behavioral data of observation time T capture;From the man-machine interaction for T for the time span
The extracting data finger health corresponding characteristic vector of attribute.
6th step, generates the testing result of finger health attribute.By the finger the generating health corresponding characteristic vector of attribute
As the input of the finger health detection of attribute model of foundation, obtain the detected value of user's finger health attribute;This is detected
Value is compared with decision-making value, and the finger health attribute of user is judged.
Explanation with regard to identity attribute category setting in embodiment
Identity attribute category setting in an embodiment, is only used as a kind of embodiment of present invention.Practical application
In can have other identity attribute category setting schemes.As age attribute can also be divided into 4 classes as needed:The first kind is
The user that age is less than 15 years old;Equations of The Second Kind be the age 15-30 year between user;3rd class is more than the user in 30-50 year,
4th class is the user more than 50 years old.Now only need to set using identical classification during setting up identity attribute detection model
Fixed, you can using method of the present invention.
Experimental result according to detection part identity attribute of the present invention
The keystroke behavioral data of mouse behavioral data and 51 users by collecting 58 users sets up man-machine interaction row
For data set.By experiment, method proposed by the present invention and technology are verified.Table 1 lists the detection of identity attribute information
Result.
The statistical result of table 1. identity attribute infomation detection.
*:Upper hurdle represents the result carrying out sex-screening using keyboard behavior, and lower hurdle represents and carries out sex using mouse behavior
The result of detection.
As shown in the experimental result of table 1 and Fig. 4, method proposed by the present invention can be believed to the identity attribute of user exactly
Breath is detected.When being detected to user identity attribute information using human-machine interactive information, accuracy rate is above 85%.When
When the ethnic information of user being detected using keyboard mutuality data, related accuracy rate is 87.32%.This result verification
The feasibility of method proposed by the invention, shows that the method can be computer and mobile network user information Perception analyzes offer one
Plant effective technological means.
Claims (4)
1. a kind of user identity attribute detection method based on man-machine interaction behavior characteristicss is it is characterised in that include setting up identity
Attribute model and detection two parts of identity attribute:
(1) set up identity attribute model, comprise the steps:
The first step, during computer and smart phone user are normally using human-computer interaction device, gather and records user
Man-machine interaction behavioral data, including mouse interbehavior data, keystroke interbehavior data, touch interbehavior data;
Second step, for all computers and smart phone user, defines the species of identity attribute, and the drawing of every kind of identity attribute
Sub-category;
3rd step, was divided to the man-machine interaction behavioral data of record with fixed observer time span T for the cycle, was formed multiple
Time span is the man-machine interaction behavioral data block of T;The species of the identity attribute according to user and the division of every kind of identity attribute
These behavioral data blocks are marked;
4th step, for the data block of each labelling, extracts and labelling man-machine interaction behavior characteristicss vector, by different pieces of information block
Man-machine interaction behavior characteristicss vector combination formed user identity attribute characteristic vector training set;
5th step, for every kind of identity attribute, according to the labelling of identity attribute species, obtains the corresponding feature of each identity attribute
Vectorial training set;Divide the labelling of classification according to each identity attribute, obtain the labelling of each characteristic vector in training set;Respectively
Using corresponding for each identity attribute characteristic vector training set as training sample, the labelling of characteristic vector in training set is made simultaneously
For the labelling of training sample, identity attribute model is built respectively to each identity attribute;
(2) detect identity attribute, comprise the steps:
The first step, after user logins computer or smart mobile phone, the man-machine interaction behavior of capture active user, with length T as week
Phase, obtain in T user's man-machine interaction behavioral data and extract corresponding man-machine interaction behavior characteristicss vector, and then generate corresponding every
The characteristic vector of individual identity attribute;
Second step, detects to the identity attribute of active user:Using the identity attribute characteristic vector generating as foundation
The input of identity attribute presumption model, obtains the detected value of user identity attribute, and the identity attribute of user is judged;
Described identity attribute model of setting up is realized by one or more Classifier combination, and described grader includes Weighted random forest
Grader, artificial nerve network classifier, support vector machine classifier;
Concretely comprising the following steps of identity attribute model is set up by Weighted random forest classified device:
1) initialization training sample concentrates the number of feature samples is N, and in each feature samples, the number of characteristic component is M, certainly
The number of plan tree is P, and the number of the decision-making feature of each decision tree is m, and m is much smaller than M;
2) in order to eliminate the impact of different feature dimensions, each dimensional feature vector is normalized, its value is limited
Between [0,1];
3) use Bagging algorithm that N number of feature samples are sampled P time, obtain P characteristic set;
4) characteristic set is randomly selected to each random tree, and this decision tree is estimated and error analyses, for
Each of tree node, randomly chooses the m characteristic component put based on this, and for different classes of feature samples, gives
Different weights are to find optimal partitioning scheme;
5) according to classifying quality best characteristic node node division is Liang Ge branch, then recursive call step 4) until this
Tree can Accurate classification training sample set, or all properties were used;After the complete growth of decision tree, it is not entered
Row beta pruning;
6) repeat step 3), 4), 5) until establishing whole P decision trees;
7) carry out, using the methods of the most ballots based on weighting, the classification results that synthesis determines multiple decision trees, that is, obtain weighting with
The classification results of machine forest classified device.
2. the user identity attribute detection method based on man-machine interaction behavior characteristicss according to claim 1, its feature exists
In the described concrete steps setting up the identity attribute characteristic vector training set forming user in identity attribute model part the 4th step
As follows:
The first step, in the man-machine interaction behavioral data block for T for the observation interval, travels through man-machine interaction sequence of events, successively
Isolate different types of interbehavior event, including mouse interbehavior event, keystroke interbehavior event, touch interaction row
For event;
Second step, for different types of interbehavior event, extracts interbehavior characteristic vector, including mouse behavior characteristicss to
Amount, keystroke behavior characteristicss vector, touch behavior characteristicss vector;
3rd step, the man-machine interaction behavior characteristicss vector in different pieces of information block is combined, formed identity attribute feature to
Amount training set.
3. the user identity attribute detection method based on man-machine interaction behavior characteristicss according to claim 1, its feature exists
In the man-machine interaction behavioral data of described computer or smart phone user generation is the sequence of basic man-machine interaction event composition
Arrange, the form of basic man-machine interaction event is:{ interaction time stabs, and interactive screen position, including mouse, keyboard or touch pad
Interactive device type, alternative events type }.
4. the user identity attribute detection method based on man-machine interaction behavior characteristicss according to claim 1, its feature exists
Refer to computer and the intrinsic physiology of smart phone user or behavioral trait in, described identity attribute, including the sex of user,
Age, race, language, right-hand man's use habit, schooling, computer use proficiency level, occupation, finger health status.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310454565.2A CN103530540B (en) | 2013-09-27 | 2013-09-27 | User identity attribute detection method based on man-machine interaction behavior characteristics |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310454565.2A CN103530540B (en) | 2013-09-27 | 2013-09-27 | User identity attribute detection method based on man-machine interaction behavior characteristics |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103530540A CN103530540A (en) | 2014-01-22 |
CN103530540B true CN103530540B (en) | 2017-02-22 |
Family
ID=49932544
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201310454565.2A Active CN103530540B (en) | 2013-09-27 | 2013-09-27 | User identity attribute detection method based on man-machine interaction behavior characteristics |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103530540B (en) |
Families Citing this family (47)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103870547A (en) * | 2014-02-26 | 2014-06-18 | 华为技术有限公司 | Grouping processing method and device of contact persons |
CN103942489A (en) * | 2014-03-31 | 2014-07-23 | 中国科学院信息工程研究所 | Attack detection method and system on basis of cursor hidden scene |
CN103927467A (en) * | 2014-04-25 | 2014-07-16 | 武汉大学 | Smart phone authentication system and method based on screen touching behavior |
CN105279691A (en) * | 2014-07-25 | 2016-01-27 | 中国银联股份有限公司 | Financial transaction detection method and equipment based on random forest model |
CN104239761B (en) * | 2014-09-15 | 2017-06-27 | 西安交通大学 | The identity for sliding behavioural characteristic based on touch screen continues authentication method |
CN104537252B (en) * | 2015-01-05 | 2019-09-17 | 深圳市腾讯计算机系统有限公司 | User Status list disaggregated model training method and device |
CN104778387B (en) * | 2015-04-23 | 2017-12-08 | 西安交通大学 | Cross-platform identity authorization system and method based on man-machine interaction behavior |
CN104809377B (en) * | 2015-04-29 | 2018-01-05 | 西安交通大学 | Network user identity monitoring method based on webpage input behavior feature |
CN104951544A (en) * | 2015-06-19 | 2015-09-30 | 百度在线网络技术(北京)有限公司 | User data processing method and system and method and system for providing user data |
CN105306496B (en) * | 2015-12-02 | 2020-04-14 | 中国科学院软件研究所 | User identity detection method and system |
CN105607741B (en) * | 2015-12-31 | 2019-01-15 | 联想(北京)有限公司 | A kind of control method and electronic equipment |
RU2626337C1 (en) * | 2016-02-18 | 2017-07-26 | Акционерное общество "Лаборатория Касперского" | Method of detecting fraudulent activity on user device |
CN105760516A (en) * | 2016-02-25 | 2016-07-13 | 广州视源电子科技股份有限公司 | Method and device for distinguishing users |
CN109074622B (en) | 2016-04-27 | 2022-04-05 | 北京嘀嘀无限科技发展有限公司 | System and method for determining transportation service route |
CN107798446A (en) * | 2016-09-07 | 2018-03-13 | 北京嘀嘀无限科技发展有限公司 | The evaluation process method and device of rich ore route |
CN107451437B (en) * | 2016-05-31 | 2021-04-16 | 百度在线网络技术(北京)有限公司 | Locking method and device of mobile terminal |
CN106095915A (en) * | 2016-06-08 | 2016-11-09 | 百度在线网络技术(北京)有限公司 | The processing method and processing device of user identity |
CN106156591A (en) * | 2016-07-12 | 2016-11-23 | 中国科学技术大学苏州研究院 | A kind of smart phone user Transparent Authentication method under cloud environment |
CN106371750A (en) * | 2016-08-30 | 2017-02-01 | 北京奇艺世纪科技有限公司 | Method and device for confirming user gender |
CN106503499A (en) * | 2016-09-22 | 2017-03-15 | 天津大学 | Smart mobile phone touch-screen input recognition method based on machine learning |
CN106572097B (en) * | 2016-11-01 | 2019-07-09 | 南京邮电大学 | A kind of mobile device-based hybrid identity identifying method |
CN106856015B (en) * | 2016-12-20 | 2019-08-16 | 国网山东省电力公司东明县供电公司 | A kind of Work attendance method and device |
CN106503722B (en) * | 2016-12-21 | 2019-09-24 | 湖南文理学院 | Operator's skeleton growth rings method based on touch trajectory image procossing |
CN107194216A (en) * | 2017-05-05 | 2017-09-22 | 中南大学 | A kind of mobile identity identifying method and system of the custom that swiped based on user |
CN107026928A (en) * | 2017-05-24 | 2017-08-08 | 武汉大学 | A kind of behavioural characteristic identification authentication method and device based on mobile phone sensor |
CN107623715B (en) * | 2017-08-08 | 2020-06-09 | 阿里巴巴集团控股有限公司 | Identity information acquisition method and device |
CN107688827A (en) * | 2017-08-24 | 2018-02-13 | 西安交通大学 | A kind of user identity attribute forecast method based on user's daily behavior feature |
CN107644106B (en) * | 2017-10-17 | 2020-10-09 | 厦门市美亚柏科信息股份有限公司 | Method, terminal device and storage medium for automatically mining service middleman |
CN110020155A (en) | 2017-12-06 | 2019-07-16 | 广东欧珀移动通信有限公司 | User's gender identification method and device |
CN109948633A (en) * | 2017-12-20 | 2019-06-28 | 广东欧珀移动通信有限公司 | User gender prediction method, apparatus, storage medium and electronic equipment |
CN109961077A (en) * | 2017-12-22 | 2019-07-02 | 广东欧珀移动通信有限公司 | Gender prediction's method, apparatus, storage medium and electronic equipment |
CN108090216B (en) * | 2017-12-29 | 2021-02-05 | 咪咕文化科技有限公司 | Label prediction method, device and storage medium |
CN108600258A (en) * | 2018-05-09 | 2018-09-28 | 华东师范大学 | A kind of method for auditing safely towards Integrated Electronic System self-generating white list |
CN109067970A (en) * | 2018-06-27 | 2018-12-21 | 上海擎感智能科技有限公司 | Based on the smart phone display methods and system of onboard instruments screen, car-mounted terminal |
CN109145544A (en) * | 2018-09-05 | 2019-01-04 | 郑州云海信息技术有限公司 | A kind of human-computer behavior detection system and method |
CN111147431B (en) * | 2018-11-06 | 2022-06-07 | 北京京东尚科信息技术有限公司 | Method and apparatus for generating information |
CN109583161B (en) * | 2018-11-27 | 2021-08-06 | 咪咕文化科技有限公司 | Information processing method and device and storage medium |
CN110110766B (en) * | 2019-04-23 | 2021-08-13 | 西安交通大学 | Online character analysis method and device based on motion planning control characteristics |
CN110363283B (en) * | 2019-06-06 | 2021-06-22 | 哈尔滨工业大学(深圳) | User attribute prediction method based on deep learning and related device |
CN110347928A (en) * | 2019-07-16 | 2019-10-18 | 韶关市启之信息技术有限公司 | A kind of recommendation method for personalized information and system |
CN111047332B (en) * | 2019-11-13 | 2021-05-07 | 支付宝(杭州)信息技术有限公司 | Model training and risk identification method, device and equipment |
CN111046374A (en) * | 2019-11-26 | 2020-04-21 | 山东浪潮人工智能研究院有限公司 | Method and system for improving user password security based on machine learning |
CN111104301B (en) * | 2019-12-03 | 2021-08-20 | 深圳市联谛信息无障碍有限责任公司 | Method and system for judging barrier user in webpage |
CN111488519A (en) * | 2020-03-02 | 2020-08-04 | 深圳壹账通智能科技有限公司 | Method and device for identifying gender of user, electronic equipment and storage medium |
WO2021243534A1 (en) * | 2020-06-02 | 2021-12-09 | 深圳市欢太科技有限公司 | Behavior control method and apparatus and storage medium |
WO2021262170A1 (en) * | 2020-06-25 | 2021-12-30 | Google Llc | Anomalous user interface input detection |
CN115081334A (en) * | 2022-06-30 | 2022-09-20 | 支付宝(杭州)信息技术有限公司 | Method, system, apparatus and medium for predicting age bracket or gender of user |
-
2013
- 2013-09-27 CN CN201310454565.2A patent/CN103530540B/en active Active
Non-Patent Citations (4)
Title |
---|
基于击键特征的手机用户身份认证系统;张明烁 等;《软件开发与设计》;20091130;第2009年卷(第22期);第39-40、58页 * |
基于击键特征的用户身份认证新方法;朱明 等;《计算机工程》;20021031;第28卷(第10期);第138-139、142页 * |
基于鼠标动力学模型的用户身份认证与监控;房超 等;《西安交通大学学报》;20081031;第42卷(第10期);第1235-1239页 * |
基于鼠标行为特征的用户身份认证与监控;沈超 等;《通信学报》;20100731;第31卷(第7期);第68-75页 * |
Also Published As
Publication number | Publication date |
---|---|
CN103530540A (en) | 2014-01-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103530540B (en) | User identity attribute detection method based on man-machine interaction behavior characteristics | |
Shi et al. | Detecting malicious social bots based on clickstream sequences | |
CN104809377B (en) | Network user identity monitoring method based on webpage input behavior feature | |
CN104504404B (en) | The user on the network's kind identification method and system of a kind of view-based access control model behavior | |
CN106886792A (en) | A kind of brain electricity emotion identification method that Multiple Classifiers Combination Model Based is built based on layering | |
CN110110663A (en) | A kind of age recognition methods and system based on face character | |
CN109376613A (en) | Video brainpower watch and control system based on big data and depth learning technology | |
CN105337987A (en) | Network user identity authentication method and system | |
CN112329816A (en) | Data classification method and device, electronic equipment and readable storage medium | |
Du et al. | An integrated framework based on latent variational autoencoder for providing early warning of at-risk students | |
CN109726747A (en) | Recommend the data fusion sort method of platform based on social networks | |
Zou et al. | A novel network security algorithm based on improved support vector machine from smart city perspective | |
CN111191099A (en) | User activity type identification method based on social media | |
CN104572877A (en) | Detection method and detection system of game public opinion | |
Liu et al. | Coevil: A coevolutionary model for crime inference based on fuzzy rough feature selection | |
CN107247656A (en) | A kind of machine, which is passed the examination, records and counts the method and system of examinee's mouse behavior | |
Wang et al. | Data acquisition model for online learning activity in distance English teaching based on xAPI | |
CN106780258A (en) | A kind of method for building up and device of minor crime decision tree | |
CN116401311B (en) | Three-dimensional visual data management system and method based on GIS | |
KR20150083165A (en) | System and method for analyzing opinion time series | |
CN106066993A (en) | A kind of crowd's semantic segmentation method and system | |
CN109446394A (en) | For network public-opinion event based on modular public sentiment monitoring method and system | |
CN109657703A (en) | Listener clustering method based on space-time data track characteristic | |
CN104298751A (en) | Method and device for obtaining internet virtual identity characteristic information | |
Zuo | Implementation of HCI software interface based on image identification and segmentation algorithms |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant |