CN104331747A - Method for detecting malicious fare evasion - Google Patents
Method for detecting malicious fare evasion Download PDFInfo
- Publication number
- CN104331747A CN104331747A CN201410573517.XA CN201410573517A CN104331747A CN 104331747 A CN104331747 A CN 104331747A CN 201410573517 A CN201410573517 A CN 201410573517A CN 104331747 A CN104331747 A CN 104331747A
- Authority
- CN
- China
- Prior art keywords
- user
- place
- driver
- destination
- personalized
- 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
- 238000000034 method Methods 0.000 title claims abstract description 29
- 238000001514 detection method Methods 0.000 claims description 26
- 238000012512 characterization method Methods 0.000 claims description 19
- 239000011159 matrix material Substances 0.000 claims description 13
- 238000004364 calculation method Methods 0.000 claims description 6
- 230000003203 everyday effect Effects 0.000 claims description 4
- 238000012549 training Methods 0.000 claims description 4
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 238000011156 evaluation Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 235000001674 Agaricus brunnescens Nutrition 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000035622 drinking Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
Landscapes
- Navigation (AREA)
Abstract
The invention discloses a method for detecting malicious fare evasion. The method comprises the steps of performing statistics on the times that different users leave for different destinations from different places of departure, so as to obtain individual user historical data; building associated information between the places of departure and the destinations; building a an individual user preferable destination model; when a first user generates a service request, predicting the probability that the first user leaves for different locations as the destinations by combining the associated information of the positions of different locations in a map in accordance with the place of departure of the first user and the individual user preferable destination model; tracking the track position of a driver during a preset time after a rejection notification message sent by the driver is received, calculating the fare evasion probability of the driver based on the first user in accordance with the predicted probability, and detecting whether the driver has a fare evasion behavior. According to the method provided by the invention, the behavior of malicious fare evasion can be detected with higher accuracy rate; furthermore, if the method is fused with traditional characteristics, the accuracy rate can be improved significantly.
Description
Technical field
The present invention relates to transport information technical field, more particularly, relate to a kind of malice and escape single detection method and device.
Background technology
The rise that generation drives industry has brought a lot of convenience.The business demand that generation drives after drinking, generation of travelling drives, commercial affairs generation drives etc. also gets more and more.Traditional generation drive industry efficiency of service low, charge high, lose industrial advantages gradually.In the generation the substitute is based on mobile Internet, drives industry service mode.
The third party application that smart mobile phone realizes---in generation, drives APP and arises at the historic moment like the mushrooms after rain.User, opening after generation drives APP, can see that neighbouring generation drives driver information, and comprise name driver, drive number of times, evaluation etc. from oneself distance, driving age, generation, user can independently select the generation meeting self-demand to drive driver.Generation drives APP, and not only interface is very directly perceived, and easy to operate, most importantly driver information transparence, makes user feel more relieved, and be also conducive to setting up contact fast, the parent that Consumer's Experience aspect obtains numerous consumer looks at.
But, in actual use, often occur that the single file of escaping of driver is, such as: part driver after being driven APP by generation and receiving user's request, with user's self-dealing, to reach the object escaping information service expense.In existing generation, the system of driving provided simple rule, in order to above-mentioned single file of escaping for distinguishing, as judge driver total refuse single rate, if certain driver total refuse single rate higher than a certain threshold value, then this driver is considered as escaping single driver and processes.But the destination have a lot owing to refusing single reason, such as user initiatively refuses, gone excessively far causes transaction to reach, and therefore whether existing method only weighs driver for escaping list according to total single rate of refusing, and accuracy rate is very low.For another example judge the translational speed of driver after refusing list, if certain driver refuses single rear translational speed and the speed of a motor vehicle is similar, then this driver is considered as escaping single driver.But driver may on the bus going to somewhere, and therefore the accuracy rate of this method also suffers restraints.
Summary of the invention
Goal of the invention of the present invention is the defect for prior art, proposes a kind of malice and escapes single detection method and device, escape single accuracy rate in order to improve detection of malicious.
According to an aspect of the present invention, provide a kind of malice and escape single detection method, comprising:
Statistics different user goes to the number of times of different destination from different departure place, obtain personalized user historical data;
Go to the number of times of different destination according to all users from different departure place, set up the related information of departure place and destination, obtain the position related information of different location in map thus;
According to described personalized user historical data, set up personalized user destination preference pattern; The preference of each user to each place is obtained by the prediction of described personalized user destination preference pattern;
After first user produces services request, according to the departure place of first user and the personalized user destination preference pattern of first user, in conjunction with the position related information of different location in map, prediction first user goes to different location as the probability of destination; Receive that driver sends refuse single notification message after, the track position of driver in Preset Time is followed the tracks of, utilizes and predict that driver described in the probability calculation that obtains escapes single probability based on this first user;
Escape single probability according to described driver based on first user, detect whether described driver produces and escape single file and be.
According to another aspect of the present invention, provide a kind of malice and escape single pick-up unit, comprising:
User's historical data statistical module, to go to the number of times of different destination, obtaining personalized user historical data from different departure place for adding up different user; Go to the number of times of different destination according to all users from different departure place, set up the related information of departure place and destination, obtain the position related information of different location in map thus;
Model building module, for according to described personalized user historical data, sets up personalized user destination preference pattern; The preference of each user to each place is obtained by the prediction of described personalized user destination preference pattern;
Single probability evaluation entity is escaped based on user, for after first user produces services request, according to the departure place of first user and the personalized user destination preference pattern of first user, in conjunction with the position related information of different location in map, prediction first user goes to different location as the probability of destination; Receive that driver sends refuse single notification message after, the track position of driver in Preset Time is followed the tracks of, utilizes and predict that driver described in the probability calculation that obtains escapes single probability based on this first user;
Detection module, for escaping single probability according to described driver based on first user, detecting whether described driver produces and escaping single file and be.
According to such scheme provided by the invention, utilize this user's historical data to set up personalized user destination preference pattern, obtain the preference of each user to each place by the prediction of described personalized user destination preference pattern; Then according to the result of personalized user destination preference pattern prediction, single file is refused for identifying to driver, identify whether for escaping single file be.This programme mainly according to first user to the preference in some place, and the information in the position related information two of different location in map show that driver escapes single probability based on first user, these two aspects all with the information relevant with position recorded in user's historical data.Total relative to driver refuses single rate, the information relevant with position recorded in user's historical data more can reflect the preference of user and associating between place, thus more can reflect whether driver escapes single situation, therefore, utilize this programme to carry out detection of malicious and escape single behavior, accuracy rate is higher.
Accompanying drawing explanation
Fig. 1 shows the schematic flow sheet that malice provided by the invention escapes single detection method embodiment one;
Fig. 2 shows the schematic flow sheet that malice provided by the invention escapes single detection method embodiment two;
Fig. 3 shows the schematic diagram of personalized user destination preference pattern in the present invention;
Fig. 4 shows the functional block diagram that malice provided by the invention escapes single pick-up unit.
Embodiment
For fully understanding the object of the present invention, feature and effect, by following concrete embodiment, the present invention is elaborated, but the present invention is not restricted to this.
Fig. 1 shows the schematic flow sheet that malice provided by the invention escapes single detection method embodiment one.As shown in Figure 1, the method comprises the steps:
Step S100, statistics different user goes to the number of times of different destination from different departure place, obtain personalized user historical data.
User mentioned herein specifically refers to the client that request generation drives service.Drive in system in generation, the history information on services to all users can be recorded, the relevant information of the destination that such as each user goes to from departure place after asking generation to drive service each time.Therefore, will count different user for the system of driving goes to the number of times of different destination as personalized user historical data from different departure place.
Step S101, goes to the number of times of different destination from different departure place according to all users, set up the related information of departure place and destination, obtains the position related information of different location in map thus.
For place A and B, add up using place A as departure place, using place B as the service times of destination, set up the position related information of place A and place B in map according to this service times.If go to the user person-time of place B a lot of from place A, reflect that the position degree of association of place A and place B is stronger.
Step S102, according to described personalized user historical data, sets up personalized user destination preference pattern; The preference of each user to each place is obtained by the prediction of described personalized user destination preference pattern.
The present embodiment utilizes personalized user historical data to set up personalized user destination preference pattern, obtains the preference of each user to each place by the prediction of described personalized user destination preference pattern.
Step S103, after first user produces services request, according to the departure place of first user and the personalized user destination preference pattern of first user, in conjunction with the position related information of different location in map, prediction first user goes to different location as the probability of destination; Receive that driver sends refuse single notification message after, the track position of driver in Preset Time is followed the tracks of, utilizes and predict that driver described in the probability calculation that obtains escapes single probability based on this first user.
By the result of above-mentioned personalized user destination preference pattern prediction, the present embodiment can refuse single file for identifying to driver, identifies whether for escaping single file to be.After first user produces services request, according to the departure place of first user and the personalized user destination preference pattern of first user, in conjunction with the position related information of different location in map, prediction first user goes to different location as the probability of destination.Receive driver send refuse single notification message after, the track position of driver in Preset Time is followed the tracks of, such as follow the tracks of the track position in driver half an hour, prediction first user goes to the probability of the place of arrival in driver half an hour to escape single probability as driver based on first user.
Step S104, escapes single probability according to described driver based on first user, detects whether described driver produces and escapes single file and be.
Driver is that the driver drawn according to two aspect information may escape single probability based on single probability of escaping of first user, and being the preference aspect of first user to some place on the one hand, is in the position related information of different location in map on the other hand.These two aspects all with the information relevant with position recorded in user's historical data.Wherein, driver escapes single probability whether can produce as detection driver the principal character escaped single file and be based on first user, certainly also comes together to detect driver in conjunction with further feature and whether produces that to escape single file be that the present invention does not limit this.
According to the said method that the present embodiment provides, utilize this user's historical data to set up personalized user destination preference pattern, obtain the preference of each user to each place by the prediction of described personalized user destination preference pattern; Then according to the result of personalized user destination preference pattern prediction, single file is refused for identifying to driver, identify whether for escaping single file be.The method that the present embodiment provides mainly according to first user to the preference in some place, and the information in the position related information two of different location in map show that driver escapes single probability based on first user, these two aspects all with the information relevant with position recorded in user's historical data.Total relative to driver refuses single rate, the information relevant with position recorded in user's historical data more can reflect the preference of user and associating between place, thus more can reflect whether driver escapes single situation, therefore, the method utilizing the present embodiment to provide is carried out detection of malicious and is escaped single behavior, and accuracy rate is higher.
Fig. 2 shows the schematic flow sheet that malice provided by the invention escapes single detection method embodiment two.As shown in Figure 2, the method comprises the steps:
Step S200, statistics different user goes to the number of times of different destination from different departure place, obtain personalized user historical data.
In the generation managing each city for system of driving in units of city, drives information on services, in a city, in the form of a grid city is divided for system of driving, each grid is as the three unities unit, in a certain order these grids are numbered, make each grid have a numbering as location information.In generation, drives in system the history information on services recorded to all users in this city, and such as each user goes to the relevant information of destination after asking generation to drive service each time from departure place.Going out different user in this city for system of driving according to these Information Statistics goes to the number of times of different location as personalized user historical data from different departure place.
Alternatively, the form that personalized user historical data ties up matrix with m*n stores, and wherein m is total number of users, and n is place sum, i.e. grid sum.The i-th row data representation i-th user of this matrix goes to the number of times of different location, and jth column data represents that different user goes to the number of times in a jth place.For 4*4 matrix, as follows:
Wherein, A
ijrepresent that i-th user goes to the number of times in a jth place.Such as, A
34=4 represent that the 3rd user goes to the number of times in the 4th place to be 4 times.Missing values in matrix represents that corresponding user did not go to corresponding place.
By matrix decomposition technology, in conjunction with the proximity relations of diverse location in map, be each user, its potential proper vector is set up in each place respectively.
Step S201, goes to the number of times of different destination from different departure place according to all users, set up the related information of departure place and destination, obtains the position related information of different location in map thus.
Except statistics personalized user historical data, also add up the position related information of different location in map, determine the position degree of association of different location in map thus.For example, for place A and B, add up using place A as departure place, using place B as the service times of destination, set up the position related information of place A and place B in map according to this service times.If go to the user person-time of place B a lot of from place A, reflect that the position degree of association of place A and place B is higher.
Step S202, the place potential proper vector corresponding according to the potential proper vector of the user of each user, each place, sets up described personalized user destination preference pattern.
In the present embodiment, the schematic diagram of the personalized user destination preference pattern set up as shown in Figure 3.The matrix U of m*k dimension represents the potential feature of user, and the matrix V of n*k dimension represents the potential feature in place, and wherein k is potential feature vector dimension (k is empirical value).The i-th row in matrix U represents that the k of i-th user ties up potential proper vector, and the jth list in matrix V shows that the k in a jth place ties up potential proper vector.According to model hypothesis A
ijnormal Distribution, its average is the inner product that k that the k of i-th user ties up potential proper vector and a jth place ties up potential proper vector, and its mean square deviation is σ
a.Average obeyed by the prior distribution of each potential proper vector eigenwert is the normal distribution of 0.Wherein the mean square deviation of user characteristics vector characteristics value is σ
u, the mean square deviation of Site characterization vector characteristics value is σ
v.
According to Fig. 3, then have:
Wherein, U
irepresent the potential proper vector of user of i-th user, V
jrepresent the potential proper vector in place in a jth place, n (j) represents the contiguous place in a jth place, V
trepresent the potential proper vector that in the contiguous place in a jth place, t place is corresponding, d
trepresent the distance between a described jth place and described t place, s () is normalized function, and α is linear superposition weight, is determined by empirical value.
The implication directly perceived of above-mentioned formula (1) is, the preference of i-th user to a jth place depends on the potential proper vector in this user and this place, and the potential proper vector in the place of following this distance location nearer.
Step S203, is optimized described personalized user destination preference pattern, obtains the Site characterization vector that each place of user characteristics vector sum of each user is corresponding.
In order to make predicated error as far as possible little, demand gets the Site characterization vector in each place of user characteristics vector sum of each user, makes visible error minimum, namely minimizes following majorized function and be optimized personalized user destination preference pattern:
Wherein, A
ijrepresent that i-th user goes to the actual value of the number of times in a jth place, I
ij() is indicative function, and represent that i-th user goes to the number of times in a jth place to be greater than 0, the difference terms in formula represents the error between actual value and predicted value,
for matrix U mould square,
for matrix V mould square,
with
be regularization term, for preventing data over-fitting.
Alternatively, adopt gradient descent method to ask for the minimum value of L (U, V), obtain the Site characterization vector that each place of user characteristics vector sum of each user is corresponding.Wherein, the solution formula of Gradient Descent is as follows:
Step S204, the Site characterization vector corresponding according to each place of user characteristics vector sum of each user, calculates the preference of each user to each place.
Particularly, Site characterization corresponding for each place of user characteristics vector sum of each user vector is substituted in the preference pattern of described personalized user destination, namely substitutes in formula (1), obtain the preference of each user to each place.
As shown in the above, the main feature of of the present embodiment is, when setting up personalized user destination preference pattern, not only consider the direct relation of different location and different user, if the distance also contemplating two places is closer, the distance of the Site characterization vector both that also should be closer, and that is, user also can be converted into the preference of user to this place to a certain extent to the preference in the place near certain place.As formula (1) is also presented as two-part weighted sum, a part is the relation in user and place, and a part is the relation in the multiple places near user and this place.In a kind of special case, suppose that certain user did not go to place 2, but he went to place 3 near place 2, place 4 and place 6, if only consider the direct relation of place and user, this user and place 2 seem that it doesn't matter, and the preference of the user drawn to place 2 is lower; But in a practical situation, user went to place 3 near place 2, place 4 and place 6, if the number of times in these places is higher near going, so user is also probably interested in place 2, the possibility of again going to is also higher, and the preference of user to place 2 utilizing formula (1) to obtain is higher.Therefore, the each user described by the preference pattern of personalized user destination that the potential proper vector that the present embodiment is corresponding according to the place near the potential proper vector in place corresponding to the potential proper vector of the user of user, place and this place described is set up more can reflect the preference of user to the preference in each place, and then improves detection of malicious and escape single accuracy rate.
Step S205, after first user produces services request, according to the departure place of first user and the personalized user destination preference pattern of first user, in conjunction with the position related information of different location in map, prediction first user goes to different location as the probability of destination, receive that driver sends refuse single notification message after, the track position of driver in Preset Time is followed the tracks of, utilizes and predict that driver described in the probability calculation that obtains escapes single probability based on this first user.
Prediction first user goes to different location to realize especially by following formula (5) as the probability of destination.Receive driver send refuse single notification message after, the track position of driver in Preset Time is followed the tracks of, such as, track position in driver half an hour, substitute into using the place of arrival in driver's Preset Time as destination in formula (5), driver can be calculated and escape single probability based on this first user.
p(target＝V
j|S
h,U
i)∝β·l(A′
ij)+(1-β)·p(V
j|S
h) (5)
Above-mentioned formula (5) is expressed as driver and escapes single Probability p (target=V based on this first user
j| S
h, U
i) be proportional to β l (A '
ij)+(1-β) p (V
j| S
h).
Wherein, U
irepresent first user, V
jrepresent predetermined place of arrival, determine according to the track position of driver in Preset Time, S
hrepresent the departure place of first user, A '
ijrepresent that first user is to the preference of described predetermined place of arrival, p (V
j| S
h) representing the position degree of association between the departure place of described first user and described predetermined place of arrival, β is linear superposition weight, is determined by empirical value.
In the present embodiment, driver based on first user to escape single probability not only relevant with the preference of first user to described predetermined place of arrival, also relevant with the degree of association between the departure place of first user and predetermined place of arrival.The degree of association between the departure place of first user and predetermined place of arrival refers to does not consider first user itself, goes to the probability of predetermined place of arrival (being obtained by above-mentioned steps S201 statistics) from the departure place of first user.More than the user person-time going to predetermined place of arrival from the departure place of first user, the degree of association between the departure place of first user and predetermined place of arrival is also higher.
From foregoing, when the preference of first user to predetermined place of arrival is higher, and when the degree of association between the departure place of first user and predetermined place of arrival is also high, driver based on first user to escape single probability also higher.This conforms to actual conditions: if first user is to certain place very preference, and the frequency that other user arrives this place from current location is also very high, so driver refuse single after still go to this place show driver to escape single possibility larger.
Step S206, described driver is input in SVM classifier as characteristic information based on single probability of escaping of first user together with further feature information, described SVM classifier gives different weighted values to often kind of characteristic information of input in the training process, in order to whether to produce as the described driver of detection the foundation escaped single file and be.
In the present embodiment, be input in SVM classifier based on single probability of escaping of first user by driver together with further feature information, described further feature packets of information is containing one or more of following characteristics information: what driver was total refuse single rate, driver refuse single after translational speed information, driver's order every day number.Described SVM classifier gives different weighted values to often kind of characteristic information of input in the training process, in order to whether to produce as the described driver of detection the foundation escaped single file and be.
What driver was total refuse, and single rate is higher, driver refuse single after translational speed higher, driver's order every day number is lower, these factors also reflect driver to a certain extent and escape single probability, driver is input to SVM classifier trains based on single probability of escaping of first user together with these features, and the testing result drawn is more accurate.
According to the said method that the present embodiment provides, utilize this personalized user historical data to set up personalized user destination preference pattern, obtain the preference of each user to each place by the prediction of described personalized user destination preference pattern; Then according to the result of personalized user destination preference pattern prediction, single file is refused for identifying to driver, identify whether for escaping single file be.The method that the present embodiment provides mainly according to first user to the preference in some place, and the information in the position related information two of different location in map show that driver escapes single probability based on first user, these two aspects all with the information relevant with position recorded in user's historical data.Total relative to driver refuses single rate, the information relevant with position recorded in user's historical data more can reflect the preference of user and associating between place, thus more can reflect whether driver escapes single situation, therefore, the method utilizing the present embodiment to provide is carried out detection of malicious and is escaped single behavior, and accuracy rate is higher.In addition, when setting up personalized user destination preference pattern, not only consider the direct relation of different location and different user, if the distance also contemplating two places is closer, the distance of the Site characterization vector both that also should be closer, that is, user also can be converted into the preference of user to this place to a certain extent to the preference in the place near certain place, and this also improves detection of malicious to a certain extent and escapes single accuracy rate.Finally, draw driver based on first user escape single probability after, it be input in SVM classifier together with further feature information and train, the testing result drawn is more accurate.
Fig. 4 shows the functional block diagram that malice provided by the invention escapes single pick-up unit.As shown in Figure 4, this device comprises: user's historical data statistical module 400, model building module 410, escape single probability evaluation entity 420 and detection module 430 based on user.
User's historical data statistical module 400, to go to the number of times of different destination, obtaining personalized user historical data from different departure place for adding up different user; Go to the number of times of different destination according to all users from different departure place, set up the related information of departure place and destination, obtain the position related information of different location in map thus;
Model building module 410, for according to described personalized user historical data, sets up personalized user destination preference pattern; The preference of each user to each place is obtained by the prediction of described personalized user destination preference pattern;
Single probability evaluation entity 420 is escaped based on user, for after first user produces services request, according to the departure place of first user and the personalized user destination preference pattern of first user, in conjunction with the position related information of different location in map, prediction first user goes to different location as the probability of destination; Receive that driver sends refuse single notification message after, the track position of driver in Preset Time is followed the tracks of, utilizes and predict that driver described in the probability calculation that obtains escapes single probability based on this first user;
Detection module 430, for escaping single probability according to described driver based on first user, detecting whether described driver produces and escaping single file and be.
Further, described model building module 410, specifically for the potential proper vector in place corresponding to the place near: the potential proper vector in place corresponding according to the potential proper vector of the user of each user, each place and this place described, sets up described personalized user destination preference pattern.
Further, described model building module 410 specifically for: described personalized user destination preference pattern is optimized, obtains the Site characterization vector that each place of user characteristics vector sum of each user is corresponding; The Site characterization vector corresponding according to each place of user characteristics vector sum of each user, calculates the preference of each user to each place.
In this device, the form that described user's historical data ties up matrix with m*n stores, and wherein m is total number of users, and n is place sum; It is as follows that described model building module 410 sets up described personalized user destination preference pattern:
Wherein, U
irepresent the potential proper vector of user of i-th user, V
jrepresent the potential proper vector in place in a jth place, n (j) represents the contiguous place in a jth place, V
trepresent the potential proper vector that in the contiguous place in a jth place, t place is corresponding, d
trepresent the distance between a described jth place and described t place, s () is normalized function, and α is linear superposition weight.
Further, described model building module 410 is optimized personalized user destination preference pattern specifically for minimizing following majorized function:
Wherein, A
ijrepresent that i-th user goes to the actual value of the number of times in a jth place, I
ij() is indicative function, and represent that i-th user goes to the number of times in a jth place to be greater than 0, the difference terms in formula represents the error between actual value and predicted value,
represent the mould square of U,
represent the mould square of V.
Further, described model building module 410 specifically for: adopt gradient descent method to ask for the minimum value of L (U, V), obtain the Site characterization vector that each place of user characteristics vector sum of each user is corresponding.
Further, described model building module 410 specifically for: Site characterization corresponding for each place of user characteristics vector sum of each user vector is substituted in described personalized user destination preference pattern, obtains the preference of each user to each place.
Further, describedly single probability evaluation entity 420 is escaped specifically for being calculated by following formula based on user:
p(target＝V
j|S
h,U
i)∝β·l(A′
ij)+(1-β)·p(V
j|S
h)
Above-mentioned formula is expressed as driver and escapes single Probability p (target=V based on first user
j| S
h, U
i) be proportional to β l (A '
ij)+(1-β) p (V
j| S
h);
Wherein, U
irepresent first user, V
jrepresent predetermined place of arrival, determine according to the track position of driver in Preset Time, S
hrepresent the departure place of first user, A '
ijrepresent that first user is to the preference of described predetermined place of arrival, p (V
j| S
h) representing the position degree of association between the departure place of described first user and described predetermined place of arrival, β is linear superposition weight.
Further, described detection module 430 specifically for: described driver is input in SVM classifier as characteristic information based on single probability of escaping of first user together with further feature information, described SVM classifier gives different weighted values to often kind of characteristic information of input in the training process, in order to whether to produce as the described driver of detection the foundation escaped single file and be.Described further feature packets of information is containing one or more of following characteristics information alternatively: what driver was total refuse single rate, driver refuse single after translational speed information, driver's order every day number.
According to the said apparatus that the present embodiment provides, utilize this personalized user historical data to set up personalized user destination preference pattern, obtain the preference of each user to each place by the prediction of described personalized user destination preference pattern; Then according to the result of personalized user destination preference pattern prediction, single file is refused for identifying to driver, identify whether for escaping single file be.The device that the present embodiment provides mainly according to first user to the preference in some place, and the information in the position related information two of different location in map show that driver escapes single probability based on first user, these two aspects all with the information relevant with position recorded in user's historical data.Total relative to driver refuses single rate, the information relevant with position recorded in user's historical data more can reflect the preference of user and associating between place, thus more can reflect whether driver escapes single situation, therefore, the device utilizing the present embodiment to provide carrys out detection of malicious and escapes single behavior, and accuracy rate is higher.In addition, when setting up personalized user destination preference pattern, not only consider the direct relation of different location and different user, if the distance also contemplating two places is closer, the distance of the Site characterization vector both that also should be closer, that is, user also can be converted into the preference of user to this place to a certain extent to the preference in the place near certain place, and this also improves detection of malicious to a certain extent and escapes single accuracy rate.Finally, draw driver based on first user escape single probability after, it be input in SVM classifier together with further feature information and train, the testing result drawn is more accurate.
Finally; what enumerate it is to be noted that above is only specific embodiments of the invention; certain those skilled in the art can change and modification the present invention; if these amendments and modification belong within the scope of the claims in the present invention and equivalent technologies thereof, protection scope of the present invention all should be thought.
Claims (10)
1. malice escapes a single detection method, it is characterized in that, comprising:
Statistics different user goes to the number of times of different destination from different departure place, obtain personalized user historical data;
Go to the number of times of different destination according to all users from different departure place, set up the related information of departure place and destination, obtain the position related information of different location in map thus;
According to described personalized user historical data, set up personalized user destination preference pattern; The preference of each user to each place is obtained by the prediction of described personalized user destination preference pattern;
After first user produces services request, according to the departure place of first user and the personalized user destination preference pattern of first user, in conjunction with the position related information of different location in map, prediction first user goes to different location as the probability of destination; Receive that driver sends refuse single notification message after, the track position of driver in Preset Time is followed the tracks of, utilizes and predict that driver described in the probability calculation that obtains escapes single probability based on this first user;
Escape single probability according to described driver based on first user, detect whether described driver produces and escape single file and be.
2. method according to claim 1, is characterized in that, described according to described personalized user historical data, sets up personalized user destination preference pattern and comprises further:
The potential proper vector in place that place near the potential proper vector in place corresponding according to the potential proper vector of the user of each user, each place and this place described is corresponding, sets up described personalized user destination preference pattern.
3. method according to claim 2, is characterized in that, is describedly obtained the preference of each user to each place comprised further by the prediction of described personalized user destination preference pattern:
Described personalized user destination preference pattern is optimized, obtains the Site characterization vector that each place of user characteristics vector sum of each user is corresponding;
The Site characterization vector corresponding according to each place of user characteristics vector sum of each user, calculates the preference of each user to each place.
4. method according to claim 3, is characterized in that, the form that described user's historical data ties up matrix with m*n stores, and wherein m is total number of users, and n is place sum; Described personalized user destination preference pattern is set up as follows:
Wherein, U
irepresent the potential proper vector of user of i-th user, V
jrepresent the potential proper vector in place in a jth place, n (j) represents the contiguous place in a jth place, V
trepresent the potential proper vector that in the contiguous place in a jth place, t place is corresponding, d
trepresent the distance between a described jth place and described t place, s () is normalized function, and α is linear superposition weight.
5. method according to claim 4, is characterized in that, described be optimized to be specially to personalized user destination preference pattern minimize following majorized function personalized user destination preference pattern be optimized:
Wherein, A
ijrepresent that i-th user goes to the actual value of the number of times in a jth place, I
ij() is indicative function, and represent that i-th user goes to the number of times in a jth place to be greater than 0, the difference terms in formula represents the error between actual value and predicted value,
represent the mould square of U,
represent the mould square of V.
6. method according to claim 5, is characterized in that, is describedly optimized personalized user destination preference pattern, and the Site characterization vector obtaining each place of user characteristics vector sum of each user corresponding is specially:
Adopt gradient descent method to ask for the minimum value of L (U, V), obtain the Site characterization vector that each place of user characteristics vector sum of each user is corresponding.
7. the method according to claim 5 or 6, is characterized in that, the Site characterization vector corresponding according to each place of user characteristics vector sum of each user, calculates the preference of each user to each place and be specially:
Site characterization corresponding for each place of user characteristics vector sum of each user vector is substituted in the preference pattern of described personalized user destination, obtains the preference of each user to each place.
8. method according to claim 1, is characterized in that, the described driver of described calculating realizes especially by following formula based on single probability of escaping of this first user:
p(target＝V
j|S
h,U
i)∝β·l(A′
ij)+(1-β)·p(V
j|S
h)
Above-mentioned formula is expressed as described driver and escapes single Probability p (target=V based on this first user
j| S
h, U
i) be proportional to β l (A '
ij)+(1-β) p (V
j| S
h);
Wherein, U
irepresent first user, V
jrepresent predetermined place of arrival, determine according to the track position of driver in Preset Time, S
hrepresent the departure place of first user, A '
ijrepresent that first user is to the preference of described predetermined place of arrival, p (V
j| S
h) representing the position degree of association between the departure place of described first user and described predetermined place of arrival, β is linear superposition weight.
9. method according to claim 1, is characterized in that, describedly escapes single probability according to described driver based on first user, detects whether described driver produces and escapes single file for specifically to comprise:
Described driver is input in SVM classifier as characteristic information based on single probability of escaping of first user together with further feature information, described SVM classifier gives different weighted values to often kind of characteristic information of input in the training process, in order to whether to produce as the described driver of detection the foundation escaped single file and be.
10. method according to claim 9, is characterized in that, described further feature packets of information is containing one or more of following characteristics information: what driver was total refuse single rate, driver refuse single after translational speed information, driver's order every day number.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410573517.XA CN104331747B (en) | 2014-10-23 | 2014-10-23 | Malice escapes single detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410573517.XA CN104331747B (en) | 2014-10-23 | 2014-10-23 | Malice escapes single detection method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104331747A true CN104331747A (en) | 2015-02-04 |
| CN104331747B CN104331747B (en) | 2017-12-19 |
Family
ID=52406468
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410573517.XA Active CN104331747B (en) | 2014-10-23 | 2014-10-23 | Malice escapes single detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104331747B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106372674A (en) * | 2016-09-30 | 2017-02-01 | 百度在线网络技术（北京）有限公司 | Driver classification method and device applied to online taxi booking service platform |
| CN106971279A (en) * | 2017-05-03 | 2017-07-21 | 百度在线网络技术（北京）有限公司 | Estimate method, device, equipment and the storage medium of driver's order behavior |
| CN108764940A (en) * | 2018-05-24 | 2018-11-06 | 北京嘀嘀无限科技发展有限公司 | In generation, drives as monitoring and managing method, device and server |
| CN110992044A (en) * | 2019-11-12 | 2020-04-10 | 北京三快在线科技有限公司 | Data processing method and device, electronic equipment and readable storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102880642A (en) * | 2012-08-20 | 2013-01-16 | 浙江工业大学 | Bus transfer method based on weighted directed network model |
| CN103325246A (en) * | 2013-07-11 | 2013-09-25 | 河北工业大学 | Dynamic detection method for wagons of multiple vehicle types |
| CN103956044A (en) * | 2014-05-05 | 2014-07-30 | 杭州师范大学 | Intelligent taxi on-call system and method based on mobile internet |
-
2014
- 2014-10-23 CN CN201410573517.XA patent/CN104331747B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102880642A (en) * | 2012-08-20 | 2013-01-16 | 浙江工业大学 | Bus transfer method based on weighted directed network model |
| CN103325246A (en) * | 2013-07-11 | 2013-09-25 | 河北工业大学 | Dynamic detection method for wagons of multiple vehicle types |
| CN103956044A (en) * | 2014-05-05 | 2014-07-30 | 杭州师范大学 | Intelligent taxi on-call system and method based on mobile internet |
Non-Patent Citations (3)
| Title |
|---|
| LING CHEN 等: "A system for destination and future route prediction based on trajectory mining", 《PERVASIVE AND MOBILE COMPUTING》 * |
| 李雯 等: "基于运动趋势的移动对象位置预测", 《通信学报》 * |
| 苏孝强 等: "基于概率的移动用户移动模型", 《微计算机信息》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106372674A (en) * | 2016-09-30 | 2017-02-01 | 百度在线网络技术（北京）有限公司 | Driver classification method and device applied to online taxi booking service platform |
| CN106971279A (en) * | 2017-05-03 | 2017-07-21 | 百度在线网络技术（北京）有限公司 | Estimate method, device, equipment and the storage medium of driver's order behavior |
| CN108764940A (en) * | 2018-05-24 | 2018-11-06 | 北京嘀嘀无限科技发展有限公司 | In generation, drives as monitoring and managing method, device and server |
| CN110992044A (en) * | 2019-11-12 | 2020-04-10 | 北京三快在线科技有限公司 | Data processing method and device, electronic equipment and readable storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104331747B (en) | 2017-12-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chen et al. | Understanding ridesplitting behavior of on-demand ride services: An ensemble learning approach | |
| CN109919347B (en) | Road condition generation method, related device and equipment | |
| CN103905978B (en) | Location-based information-pushing method, pusher and network | |
| Battifarano et al. | Predicting real-time surge pricing of ride-sourcing companies | |
| CN105493109B (en) | Inferred using the air quality of multiple data sources | |
| Wang et al. | Crowdsensing-based consensus incident report for road traffic acquisition | |
| US20190266891A1 (en) | A method to quantitatively analyze the effects of urban built environment on road travel time | |
| Pirdavani et al. | Application of geographically weighted regression technique in spatial analysis of fatal and injury crashes | |
| AU2013205716B2 (en) | Location-based cognitive and predictive communication system | |
| CN109410568B (en) | Get-off site presumption method and system based on user portrait and transfer rule | |
| US20180314998A1 (en) | Resource Allocation in a Network System | |
| CN104809344A (en) | IC (Integrated Circuit) card data-based estimation method for passenger flow in bus station interval | |
| CN103903430A (en) | Dynamic fusion type travel time predicting method with multi-source and isomorphic data adopted | |
| CN104331747A (en) | Method for detecting malicious fare evasion | |
| CN113345229B (en) | Road early warning method based on federal learning and related equipment thereof | |
| CN110118567A (en) | Trip mode recommended method and device | |
| Mikluščák et al. | Using neural networks for route and destination prediction in intelligent transport systems | |
| US20190325239A1 (en) | Determining Vehicle Crowdedness Using Real-Time Location Data | |
| Vedagiri et al. | Estimating modal shift of car travelers to bus on introduction of bus priority system | |
| Tran et al. | A user equilibrium-based fast-charging location model considering heterogeneous vehicles in urban networks | |
| Nasrollahi et al. | Identification and prioritization of connected vehicle technologies for sustainable development in Iran | |
| CN106228830A (en) | A kind of bus arrival time real-time estimate system and method | |
| CN108133353A (en) | Trip intelligent prompt method | |
| CN110570656A (en) | Method and device for customizing public transport line | |
| CN106295868A (en) | Traffic trip data processing method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right |
Denomination of invention: Malicious evasion detection method Granted publication date: 20171219 Pledgee: Zhongguancun Branch of Bank of Beijing Co.,Ltd. Pledgor: BEIJING YIXINYIXING AUTOMOTIVE TECHNOLOGY DEVELOPMENT CO.,LTD. Registration number: Y2024110000209 |