CN104331747B - Malice escapes single detection method - Google Patents
Malice escapes single detection method Download PDFInfo
- Publication number
- CN104331747B CN104331747B CN201410573517.XA CN201410573517A CN104331747B CN 104331747 B CN104331747 B CN 104331747B CN 201410573517 A CN201410573517 A CN 201410573517A CN 104331747 B CN104331747 B CN 104331747B
- Authority
- CN
- China
- Prior art keywords
- user
- place
- mrow
- driver
- msub
- 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
Abstract
The invention discloses a kind of malice to escape single detection method.Wherein method includes：Statistics different user goes to the number of different destinations from different departure places, obtains personalized user historical data；Establish the related information of departure place and destination；Establish personalized user destination preference pattern；After the first user produces service request, according to the departure place of the first user and the personalized user destination preference pattern of the first user, with reference to position related information of the different location in map, the first user of prediction goes to probability of the different location as destination；Receive driver's transmission refuse single notification message after, track position of the driver in preset time is tracked, single probability is escaped based on first user using driver described in the obtained probability calculation of prediction, detects whether the driver produces and escapes single act.Escape single behavior using this programme to detect malice, accuracy rate is higher.If merged with traditional characteristic, accuracy rate can be significantly improved.
Description
Technical field
The present invention relates to transport information technical field, escapes single detection method and device more specifically to a kind of malice.
Background technology
The rise that generation drives industry has brought many convenience.In generation, drives after drinking, generation of travel drives, commercial affairs are for the industry driven etc.
Business demand is also more and more.Drive that industry efficiency of service is low, charge is high traditional generation, gradually lose industrial advantages.Take and generation
Be to drive industry service mode in the generation based on mobile Internet.
The third party application realized on smart mobile phone --- in generation, drives APP and arises at the historic moment like the mushrooms after rain.User
Opening after generation drives APP, it can be seen that in neighbouring generation, drives driver information, including name driver, from a distance from oneself, the driving age,
In generation, drives number, evaluation etc., and user can independently select the generation for meeting self-demand to drive driver.Generation drives APP, and not only interface is very straight
See, easy to operate, most importantly driver information transparence, user is felt more relieved, be also beneficial to quick establish contact, user
The pro-gaze of numerous consumers has been obtained in terms of experience.
However, in actual use, often there is the single act of escaping of driver, such as：Part driver drives by generation
After APP receives user's request, and user's self-dealing, to reach the purpose for escaping information service expense.Existing generation, which drives, is
System provides simple rule, to escape single act to above-mentioned and distinguish, such as judge driver always refuse single rate, if certain driver
Total single rate of refusing is higher than a certain threshold value, then is considered as the driver and escapes single driver and handled.But have very due to refusing list
It is more, for example actively refusal, the destination gone cause transaction not reach to user too far, therefore existing method is according only to total
Refuse single rate to weigh whether driver is to escape list, accuracy rate is very low.For another example translational speed of the driver after list is refused is judged, if certain
Translational speed is similar with speed after driver refuses list, then is considered as the driver and escapes single driver.But driver may go to somewhere
On bus, therefore the accuracy rate of this method also suffers restraints.
The content of the invention
The goal of the invention of the present invention is the defects of being directed to prior art, proposes that a kind of malice escapes single detection method and device,
Escape single accuracy rate to improve detection malice.
According to an aspect of the invention, there is provided a kind of malice escapes single detection method, including：
Statistics different user goes to the number of different destinations from different departure places, obtains personalized user historical data；
The number of different destinations is gone to from different departure places according to all users, establishes associating for departure place and destination
Information, thus obtain position related information of the different location in map；
According to the personalized user historical data, personalized user destination preference pattern is established；Pass through the individual character
With changing customer objective preference pattern is predicted to obtain preference of each user to each place；
After the first user produces service request, according to the departure place of the first user and the personalized user mesh of the first user
Ground preference pattern, with reference to position related information of the different location in map, the first user of prediction goes to different location conduct
The probability of destination；Receive driver transmission refuse single notification message after, to track position of the driver in preset time
It is tracked, driver described in the probability calculation obtained using prediction escapes single probability based on first user；
Single probability is escaped based on the first user according to the driver, detects whether the driver produces and escapes single act.
According to another aspect of the present invention, there is provided a kind of malice escapes single detection means, including：
User's history data statistics module, the secondary of different destinations is gone to from different departure places for counting different user
Number, obtains personalized user historical data；The number of different destinations is gone to from different departure places according to all users, is established out
Hair ground and the related information of destination, thus obtain position related information of the different location in map；
Model building module, for according to the personalized user historical data, establishing personalized user destination preference
Model；Predict to obtain preference of each user to each place by personalized user destination preference pattern；
Single probability evaluation entity is escaped based on user, for when the first user produce service request after, according to the first user
Departure place and the first user personalized user destination preference pattern, associate letter with reference to position of the different location in map
Breath, the first user of prediction go to probability of the different location as destination；Receive driver transmission refuse single notification message it
Afterwards, track position of the driver in preset time is tracked, driver described in the probability calculation obtained using prediction is based on should
First user's escapes single probability；
Detection module, for escaping single probability based on the first user according to the driver, detect whether the driver produces
Single act is escaped in life.
According to such scheme provided by the invention, personalized user destination preference mould is established using the user's history data
Type, predict to obtain preference of each user to each place by personalized user destination preference pattern；Then root
The result predicted according to personalized user destination preference pattern, the single act of refusing to driver are identified, identified whether to escape list
Behavior.Preference of this programme mainly according to the first user to some places, and position of the different location in map are closed
Connection information of both information show that driver escapes single probability based on the first user, the two aspect all with user's history data
The information relevant with position of record.Relative to driver it is total refuse single rate for, what is recorded in user's history data has with position
The information of pass can more reflect associating between the preference of user and place, thus more can reflect whether driver escapes single feelings
Condition, therefore, escape single behavior using this programme to detect malice, accuracy rate is higher.
Brief description of the drawings
Fig. 1 shows that malice provided by the invention escapes the schematic flow sheet of single detection method embodiment one；
Fig. 2 shows that malice provided by the invention escapes the schematic flow sheet of single detection method embodiment two；
Fig. 3 shows the schematic diagram of personalized user destination preference pattern in the present invention；
Fig. 4 shows that malice provided by the invention escapes the functional block diagram of single detection means.
Embodiment
To be fully understood by the purpose of the present invention, feature and effect, by following specific embodiments, the present invention is done in detail
Describe in detail bright, but the present invention is not restricted to this.
Fig. 1 shows that malice provided by the invention escapes the schematic flow sheet of single detection method embodiment one.As shown in figure 1,
This method comprises the following steps：
Step S100, statistics different user go to the number of different destinations from different departure places, obtain personalized user
Historical data.
User mentioned herein refers specifically to the client for driving service in request generation.System is driven in generation, can be recorded pair
The history information on services of all users, such as the destination that each user goes to after the service of driving of the generation of request each time from departure place
Relevant information.Therefore, will count different user for the system of driving goes to the number of different destinations to make from different departure places
For personalized user historical data.
Step S101, the number of different destinations is gone to from different departure places according to all users, establishes departure place and mesh
Ground related information, thus obtain position related information of the different location in map.
For place A and B, count using place A as departure place, using place B as the service times of destination, according to
The position related information of place A and place B in map is established according to the service times.If A goes to place B user from place
It is person-time a lot, reflect that place A and place B the position degree of association are stronger.
Step S102, according to the personalized user historical data, establish personalized user destination preference pattern；Pass through
Personalized user destination preference pattern is predicted to obtain preference of each user to each place.
The present embodiment establishes personalized user destination preference pattern using personalized user historical data, passes through described
Property customer objective preference pattern predict to obtain preference of each user to each place.
Step S103, after the first user produces service request, according to the departure place of the first user and of the first user
Property customer objective ground preference pattern, with reference to position related information of the different location in map, the first user of prediction goes to not
With probability of the place as destination；Receive driver transmission refuse single notification message after, to driver in preset time
Track position be tracked, single probability is escaped based on first user using driver described in the obtained probability calculation of prediction.
The result predicted by above-mentioned personalized user destination preference pattern, the present embodiment can refuse single act to driver
It is identified, identifies whether to escape single act.After the first user produces service request, according to the departure place of the first user and the
The personalized user destination preference pattern of one user, with reference to position related information of the different location in map, prediction first
User goes to probability of the different location as destination.Receive driver transmission refuse single notification message after, driver is existed
Track position in preset time is tracked, such as the track position in driver's half an hour is tracked, and prediction first is used
The probability for the place of arrival that family was gone in driver's half an hour escapes single probability as driver based on the first user.
Step S104, single probability is escaped based on the first user according to the driver, detects whether the driver produces and escapes
Single act.
Escape single probability of the driver based on the first user is that the driver drawn according to two aspect information may escape single probability,
On the one hand it is preference aspect of first user to some places, is on the other hand position related information of the different location in map
Aspect.The two information relevant with position of aspect all with being recorded in user's history data.Wherein, driver is based on the first user
Escape single probability and can escape the principal character of single act as whether detection driver produce, come together certainly also in relation with further feature
Whether detection driver, which produces, is escaped single act, and the present invention is without limitation.
The above method provided according to the present embodiment, personalized user destination preference is established using the user's history data
Model, predict to obtain preference of each user to each place by personalized user destination preference pattern；Then
The result predicted according to personalized user destination preference pattern, the single act of refusing to driver are identified, identified whether to escape
Single act.Preference of the method that the present embodiment provides mainly according to the first user to some places, and different location exist
Information of both position related information in map show that driver escapes single probability based on the first user, the two aspects all with
The information relevant with position recorded in user's history data.Relative to driver it is total refuse single rate for, in user's history data
The information relevant with position of record can more reflect associating between the preference of user and place, thus can more reflect driver
Whether single situation is escaped, and therefore, the method provided using the present embodiment escapes single behavior to detect malice, and accuracy rate is higher.
Fig. 2 shows that malice provided by the invention escapes the schematic flow sheet of single detection method embodiment two.As shown in Fig. 2
This method comprises the following steps：
Step S200, statistics different user go to the number of different destinations from different departure places, obtain personalized user
Historical data.
In the generation for managing each city in units of city for system of driving, drives information on services, and in a city, generation drives system
City is divided in the form of a grid, each grid is as one place unit, in a certain order to these grids
It is numbered so that there is each grid a numbering to be used as location information.Generation, which drives record in system, to be had in the city to all
The history information on services of user, such as each user go to the correlation of destination after the service of driving of the generation of request each time from departure place
Information.In generation, drives system and goes out the number that different user in the city goes to from different departure places different location according to these Information Statistics
As personalized user historical data.
Alternatively, personalized user historical data stores in the form of m*n ties up matrix, and wherein m is total number of users, and n is ground
Point sum, i.e. grid sum.I-th row data of this matrix represent that i-th of user goes to the number of different location, jth columns
The number in j-th of place is gone to according to expression different user.It is as follows by taking 4*4 matrixes as an example：
Wherein, AijRepresent that i-th of user goes to the number in j-th of place.For example, A34=4 the 3rd users of expression go to
The number in the 4th place is 4 times.User corresponding to missing values expression in matrix did not go to corresponding place.
It is each user with reference to proximity relations of the diverse location in map by matrix decomposition technology, each ground
Point establishes its potential characteristic vector respectively.
Step S201, the number of different destinations is gone to from different departure places according to all users, establishes departure place and mesh
Ground related information, thus obtain position related information of the different location in map.
In addition to counting personalized user historical data, position related information of the different location in map is also counted,
Thereby determine that the position degree of association of the different location in map.For example, for place A and B, count using place A as going out
Hair ground, using place B as the service times of destination, the position of place A and place B in map is established according to the service times
Related information.If A goes to place B user person-time many from place, reflect the place A and place B position degree of association compared with
It is high.
Step S202, according to the potential feature in place corresponding to the potential characteristic vector of the user of each user, each place to
Amount, establishes personalized user destination preference pattern.
In the present embodiment, the schematic diagram for the personalized user destination preference pattern established is as shown in Figure 3.M*k dimensions
Matrix U represents the potential feature of user, and the matrix V of n*k dimensions represents the potential feature in place, and wherein k is potential feature vector dimension (k
It is empirical value).The i-th row in matrix U represents that the k of i-th of user ties up potential characteristic vector, and the jth row in matrix V represent jth
The k in individual place ties up potential characteristic vector.According to model hypothesis AijNormal Distribution, its average are that the k dimensions of i-th of user are potential
The k in characteristic vector and j-th of place ties up the inner product of potential characteristic vector, and its mean square deviation is σA.Each potential characteristic vector is special
The prior distribution of value indicative obeys the normal distribution that average is 0.Wherein the mean square deviation of user characteristics vector characteristics value is σU, place spy
The mean square deviation for levying vector characteristics value is σV。
According to Fig. 3, then have：
Wherein, Aij' represent i-th of user to the preference in j-th of place, UiRepresent the potential spy of user of i-th of user
Sign vector, VjThe potential characteristic vector in place in j-th of place is represented, n (j) represents the neighbouring place in j-th of place, VtRepresent jth
Potential characteristic vector, d corresponding to t-th of place in the neighbouring place in individual placetRepresent j-th of place and described t-th
The distance between place, s () are normalized function, and α is linear superposition weight, is determined by empirical value.
The directly perceived of above-mentioned formula (1) is meant that i-th of user depends on the user and the ground to the preference in j-th of place
The potential characteristic vector of point, and the potential characteristic vector with the nearer place of the distance location.
Step S203, personalized user destination preference pattern is optimized, the user for obtaining each user is special
Levy Site characterization vector corresponding to each place of vector sum.
In order that must predict that error is as far as possible small, demand takes the place in each place of user characteristics vector sum of each user special
Sign vector so that visible error is minimum, that is, it is excellent to the progress of personalized user destination preference pattern to minimize following majorized function
Change：
Wherein, AijRepresent that i-th of user goes to the actual value of the number in j-th of place, Iij() is indicative function, is represented
I-th of user goes to the number in j-th of place, and more than 0, the difference terms in formula represent the error between actual value and predicted value,For square of the mould of matrix U,For square of the mould of matrix V,WithIt is regularization term, for preventing from counting
According to over-fitting.
Alternatively, L (U, V) minimum value is asked for using gradient descent method, obtains the user characteristics vector of each user
Ui' and each place corresponding to Site characterization vector Vi'.Wherein, the solution formula that gradient declines is as follows：
Step S204, according to Site characterization vector corresponding to each place of user characteristics vector sum of each user, calculate
Obtain preference of each user to each place.
Specifically, Site characterization vector corresponding to each place of user characteristics vector sum of each user is substituted into described
Property customer objective in preference pattern, that is, substitute into formula (1), obtain preference A of each user to each placeij″。
As shown in the above, the characteristics of one of the present embodiment main is to establish personalized user destination preference
During model, not only consider the direct relation of different location and different user, it is also contemplated that if the distance in two places is closer
If, the distance of that both Site characterization vector also should be closer, that is to say, that user is to the place near some place
Preference can also convert into preference of the user to the place to a certain extent.As formula (1) is also embodied in two-part weighted sum,
A part is the relation in user and place, and a part is user and the relation in multiple places near the place.In a kind of special case
In, it is assumed that certain user did not go to place 2, but his place 3, place 4 and place 6 for going near place 2, if only considering ground
Point and the direct relation of user, the user do not appear to relation with place 2, and the user drawn is relatively low to the preference in place 2；But
In a practical situation, place 3, place 4 and the place 6 that user was gone near place 2, if go the number in these neighbouring places compared with
If height, then user is likely to interested to place 2, and the possibility gone to again is also higher, is obtained using formula (1)
User it is higher to the preference in place 2.Therefore, the present embodiment is according to corresponding to the potential characteristic vector of the user of user, place
The personalized user mesh that potential characteristic vector corresponding to place near the potential characteristic vector in place and the described place is established
Ground preference pattern described by each user the preference of user can more be reflected to the preference in each place, and then improve
Detection malice escapes single accuracy rate.
Step S205, after the first user produces service request, according to the departure place of the first user and of the first user
Property customer objective ground preference pattern, with reference to position related information of the different location in map, the first user of prediction goes to not
With probability of the place as destination, receive driver's transmission refuse single notification message after, to driver in preset time
Track position be tracked, single probability is escaped based on first user using driver described in the obtained probability calculation of prediction.
Predict that the first user goes to different location to be realized as the probability of destination especially by equation below (5).Connecing
Receive driver's transmission refuse single notification message after, track position of the driver in preset time is tracked, such as driver
Track position in half an hour, substituted into using the place of arrival in driver's preset time as destination in formula (5), you can calculate
Obtain driver and single probability is escaped based on first user.
P (target=Vj'|Sh,Ui')∝β·l(Aij″)+(1-β)·p(Vj'|Sh) (5)
Above-mentioned formula (5) is expressed as driver and escapes single Probability p (target=V based on first userj'|Sh,Ui') direct ratio
In β l (Aij″)+(1-β)·p(Vj'|Sh)。
Wherein, Ui' represent that the user characteristics of the first user is vectorial, Vj' represent to be destined to the Site characterization vector in place,
Being destined to place is determined, S according to track position of the driver in preset timehThe departure place of the first user is represented,
Aij" represent the first user to the preference for being destined to place, p (Vj'|Sh) represent the departure place of first user with
The position degree of association being destined between place, β are linear superposition weight, are determined by empirical value.
In the present embodiment, escape single probability of the driver based on the first user is not only destined to place with the first user to described
Preference it is relevant, it is also relevant with the degree of association being destined between place with the departure place of the first user.First user's goes out
Hair ground and the degree of association being destined between place refer to not consider the first user in itself, gone to from the departure place of the first user pre-
Determine the probability (counting to obtain by above-mentioned steps S201) of place of arrival.Gone to from the departure place of the first user and be destined to place
User person-time it is more, the departure place of the first user and the degree of association being destined between place are also higher.
From the foregoing it can be that when the first user is higher to the preference for being destined to place, and first user goes out
In the case that hair ground and the degree of association that is destined between place are also high, escape single probability of the driver based on the first user is also got over
It is high.This is consistent with actual conditions：If the first user is to some place very preference, and other users reach from current location and are somebody's turn to do
The frequency in place is also very high, then driver refuse it is single after still go to the place show driver to escape single possibility larger.
Step S206, the driver is escaped into single probability as characteristic information and further feature information one based on the first user
Rise and be input in SVM classifier, the SVM classifier assigns different power to every kind of characteristic information of input in the training process
Weight values, whether to produce the foundation for escaping single act as the detection driver.
In the present embodiment, escape single probability of the driver based on the first user is input to SVM points together with further feature information
In class device, the one or more of further feature packet information containing following characteristics：Driver it is total refuse single rate, after driver refuses list
Translational speed information, the daily order number of driver.The SVM classifier is assigned to every kind of characteristic information of input in the training process
Different weighted values is given, whether to produce the foundation for escaping single act as the detection driver.
Driver it is total refuse that single rate is higher, driver refuses that the translational speed after list is higher, the daily order number of driver is relatively low, these because
Element also reflects driver and escapes single probability to a certain extent, and driver is escaped into single probability and these features one based on the first user
Rise and be input to SVM classifier and be trained, the testing result drawn is more accurate.
The above method provided according to the present embodiment, personalized user purpose is established using the personalized user historical data
Ground preference pattern, predict to obtain preference of each user to each place by personalized user destination preference pattern
Degree；Then the result predicted according to personalized user destination preference pattern, the single act of refusing to driver are identified, and identification is
No is to escape single act.The method that the present embodiment provides is mainly the preference to some places according to the first user, and different
Information of both position related information of the place in map show that driver escapes single probability based on the first user, the two sides
With position relevant information of the face all with being recorded in user's history data.Relative to driver it is total refuse single rate for, user's history
The information relevant with position recorded in data can more reflect associating between the preference of user and place, thus can more reflect
Go out whether driver escapes single situation, therefore, the method provided using the present embodiment escapes single behavior to detect malice, and accuracy rate is more
It is high.In addition, when establishing personalized user destination preference pattern, different location and the direct pass of different user are not only considered
System, it is also contemplated that if the distance in two places is closer, the distance of that both Site characterization vector also should be closer,
That is, user can also convert into user to the inclined of the place to a certain extent to the preference in the place near some place
Good, this also improves detection malice and escapes single accuracy rate to a certain extent.Finally, showing that driver escaped based on the first user
After single probability, it is input in SVM classifier together with further feature information and is trained, the testing result drawn is more
Accurately.
Fig. 4 shows that malice provided by the invention escapes the functional block diagram of single detection means.As shown in figure 4, the device
Including：User's history data statistics module 400, model building module 410, based on user escape single probability evaluation entity 420 with
And detection module 430.
User's history data statistics module 400, different destinations are gone to from different departure places for counting different user
Number, obtain personalized user historical data；The number of different destinations is gone to from different departure places according to all users, is established
Departure place and the related information of destination, thus obtain position related information of the different location in map；
Model building module 410, for according to the personalized user historical data, it is inclined to establish personalized user destination
Good model；Predict to obtain preference of each user to each place by personalized user destination preference pattern；
Single probability evaluation entity 420 is escaped based on user, for after the first user produces service request, being used according to first
The departure place at family and the personalized user destination preference pattern of the first user, with reference to position association of the different location in map
Information, the first user of prediction go to probability of the different location as destination；Receive driver transmission refuse single notification message
Afterwards, track position of the driver in preset time is tracked, driver described in the probability calculation obtained using prediction is based on
First user's escapes single probability；
Detection module 430, for escaping single probability based on the first user according to the driver, whether detect the driver
Single act is escaped in generation.
Further, the model building module 410 is specifically used for：According to the potential characteristic vector of the user of each user,
The potential characteristic vector in place corresponding to the potential characteristic vector in place corresponding to each place and the place near the described place,
Establish personalized user destination preference pattern.
Further, the model building module 410 is specifically used for：Personalized user destination preference pattern is entered
Row optimization, obtain Site characterization vector corresponding to each place of user characteristics vector sum of each user；According to each user's
Site characterization vector, is calculated preference of each user to each place corresponding to each place of user characteristics vector sum.
In the present apparatus, the user's history data store in the form of m*n ties up matrix, and wherein m is total number of users, and n is ground
Point sum；It is as follows that the model building module 410 establishes personalized user destination preference pattern：
Wherein, Aij' represent i-th of user to the preference in j-th of place, UiRepresent the potential spy of user of i-th of user
Sign vector, VjThe potential characteristic vector in place in j-th of place is represented, n (j) represents the neighbouring place in j-th of place, VtRepresent jth
Potential characteristic vector, d corresponding to t-th of place in the neighbouring place in individual placetRepresent j-th of place and described t-th
The distance between place, s () are normalized function, and α is linear superposition weight.
Further, the model building module 410 is specifically used for minimizing following majorized function to personalized user mesh
Ground preference pattern optimize：
Wherein, AijRepresent that i-th of user goes to the actual value of the number in j-th of place, Iij() is indicative function, is represented
I-th of user goes to the number in j-th of place, and more than 0, the difference terms in formula represent the error between actual value and predicted value,U mould square is represented,Represent V mould square.
Further, the model building module 410 is specifically used for：L (U, V) minimum is asked for using gradient descent method
Value, obtain Site characterization vector corresponding to each place of user characteristics vector sum of each user.
Further, the model building module 410 is specifically used for：By the user characteristics vector sum of each user each
Site characterization vector is substituted into the preference pattern of the personalized user destination corresponding to point, obtains each user to each place
Preference.
Further, single probability evaluation entity 420 of escaping based on user is specifically used for being counted by equation below
Calculate：
P (target=Vj'|Sh,Ui')∝β·l(Aij″)+(1-β)·p(Vj'|Sh)
Above-mentioned formula is expressed as driver and escapes single Probability p (target=V based on the first userj'|Sh,Ui') it is proportional to β
l(Aij″)+(1-β)·p(Vj'|Sh)；
Wherein, Ui' represent that the user characteristics of the first user is vectorial, Vj' represent to be destined to the Site characterization vector in place,
Being destined to place is determined, S according to track position of the driver in preset timehThe departure place of the first user is represented,
Aij" represent the first user to the preference for being destined to place, p (Vj'|Sh) represent the departure place of first user with
The position degree of association being destined between place, β are linear superposition weight.
Further, the detection module 430 is specifically used for：Using the driver based on the first user escape single probability as
Characteristic information is input in SVM classifier together with further feature information, and the SVM classifier is in the training process to input
Every kind of characteristic information assigns different weighted values, whether to produce the foundation for escaping single act as the detection driver.It is optional
The one or more of ground further feature packet information containing following characteristics：Driver it is total refuse single rate, driver refuses the shifting after list
Dynamic velocity information, the daily order number of driver.
The said apparatus provided according to the present embodiment, personalized user purpose is established using the personalized user historical data
Ground preference pattern, predict to obtain preference of each user to each place by personalized user destination preference pattern
Degree；Then the result predicted according to personalized user destination preference pattern, the single act of refusing to driver are identified, and identification is
No is to escape single act.The device that the present embodiment provides is mainly the preference to some places according to the first user, and different
Information of both position related information of the place in map show that driver escapes single probability based on the first user, the two sides
With position relevant information of the face all with being recorded in user's history data.Relative to driver it is total refuse single rate for, user's history
The information relevant with position recorded in data can more reflect associating between the preference of user and place, thus can more reflect
Go out whether driver escapes single situation, therefore, the device provided using the present embodiment escapes single behavior to detect malice, and accuracy rate is more
It is high.In addition, when establishing personalized user destination preference pattern, different location and the direct pass of different user are not only considered
System, it is also contemplated that if the distance in two places is closer, the distance of that both Site characterization vector also should be closer,
That is, user can also convert into user to the inclined of the place to a certain extent to the preference in the place near some place
Good, this also improves detection malice and escapes single accuracy rate to a certain extent.Finally, showing that driver escaped based on the first user
After single probability, it is input in SVM classifier together with further feature information and is trained, the testing result drawn is more
Accurately.
Finally it should be noted that be：Listed above be only the present invention specific embodiment, the technology of certain this area
The present invention can be modified by personnel and modification, if these modifications and variations belong to the claims in the present invention and its equivalent skill
Within the scope of art, protection scope of the present invention is considered as.
Claims (8)
1. a kind of malice escapes single detection method, it is characterised in that including：
Statistics different user goes to the number of different destinations from different departure places, obtains personalized user historical data；It is described
Personalized user historical data stores in the form of m*n ties up matrix, and wherein m is total number of users, and n is place sum；
The number of different destinations is gone to from different departure places according to all users, that establishes departure place and destination associates letter
Breath, thus obtains position related information of the different location in map；
It is specific as follows that personalized user destination preference pattern is established according to the personalized user historical data：
<mrow>
<msup>
<msub>
<mi>A</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</msub>
<mo>&prime;</mo>
</msup>
<mo>=</mo>
<msubsup>
<mi>U</mi>
<mi>i</mi>
<mi>T</mi>
</msubsup>
<mrow>
<mo>(</mo>
<msub>
<mi>&alpha;V</mi>
<mi>j</mi>
</msub>
<mo>+</mo>
<mo>(</mo>
<mrow>
<mn>1</mn>
<mo>-</mo>
<mi>&alpha;</mi>
</mrow>
<mo>)</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>t</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mrow>
<mi>n</mi>
<mrow>
<mo>(</mo>
<mi>j</mi>
<mo>)</mo>
</mrow>
</mrow>
</munderover>
<mi>s</mi>
<mo>(</mo>
<msub>
<mi>d</mi>
<mi>t</mi>
</msub>
<mo>)</mo>
<msub>
<mi>V</mi>
<mi>t</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
Wherein, Aij' represent i-th of user to the preference in j-th of place, UiRepresent the potential feature of user of i-th of user to
Amount, VjThe potential characteristic vector in place in j-th of place is represented, n (j) represents the neighbouring place in j-th of place, VtRepresent j-th of ground
Potential characteristic vector corresponding to t-th of place in the neighbouring place of point, dtRepresent j-th of place and t-th of place
The distance between, s () is normalized function, and α is linear superposition weight；Pass through personalized user destination preference pattern
Prediction obtains preference of each user to each place；
After the first user produces service request, according to the departure place of the first user and the personalized user destination of the first user
Preference pattern, with reference to position related information of the different location in map, the first user of prediction goes to different location as purpose
The probability on ground；Receive driver's transmission refuse single notification message after, track position of the driver in preset time is carried out
Tracking, driver described in the probability calculation obtained using prediction escape single probability based on first user；
Single probability is escaped based on the first user according to the driver, detects whether the driver produces and escapes single act；It is described to escape
Single probability is realized especially by equation below：
<mrow>
<mi>l</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<mrow>
<mn>1</mn>
<mo>+</mo>
<msup>
<mi>e</mi>
<mrow>
<mo>-</mo>
<mi>x</mi>
</mrow>
</msup>
</mrow>
</mfrac>
</mrow>
P (target=Vj'|Sh,Ui')∝β·l(Aij”)+(1-β)·p(Vj'|Sh)
Above-mentioned formula is expressed as the driver and escapes single Probability p (target=V based on first userj'|Sh,Ui') be proportional to
β·l(Aij”)+(1-β)·p(Vj'|Sh)；
Wherein, Ui' represent that the user characteristics of the first user is vectorial, Vj' represent to be destined to the Site characterization vector in place, make a reservation for
Place of arrival is determined, S according to track position of the driver in preset timehRepresent the departure place of the first user, Aij" table
Show the first user to the preference for being destined to place, p (Vj'|Sh) represent the departure place of first user with it is described pre-
Determine the position degree of association between place of arrival, β is linear superposition weight.
2. according to the method for claim 1, it is characterised in that it is described according to the personalized user historical data, establish
Personalized user destination preference pattern further comprises：
According to the potential characteristic vector in place and the place corresponding to the potential characteristic vector of the user of each user, each place
The potential characteristic vector in place corresponding to neighbouring place, establish personalized user destination preference pattern.
3. according to the method for claim 2, it is characterised in that described to pass through personalized user destination preference pattern
Prediction obtains each user and the preference in each place is further comprised：
Personalized user destination preference pattern is optimized, obtains the user characteristics vector sum of each user each
Site characterization vector corresponding to point；
According to Site characterization vector corresponding to each place of user characteristics vector sum of each user, each user couple is calculated
The preference in each place.
4. according to the method for claim 3, it is characterised in that described excellent to the progress of personalized user destination preference pattern
Change is specially to minimize following majorized function L (U, V) to optimize personalized user destination preference pattern：
<mrow>
<mi>L</mi>
<mrow>
<mo>(</mo>
<mi>U</mi>
<mo>,</mo>
<mi>V</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munder>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</munder>
<msub>
<mi>I</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</msub>
<msup>
<mrow>
<mo>(</mo>
<msub>
<mi>A</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</msub>
<mo>-</mo>
<msubsup>
<mi>U</mi>
<mi>i</mi>
<mi>T</mi>
</msubsup>
<mo>(</mo>
<mrow>
<msub>
<mi>&alpha;V</mi>
<mi>j</mi>
</msub>
<mo>+</mo>
<mrow>
<mo>(</mo>
<mrow>
<mn>1</mn>
<mo>-</mo>
<mi>&alpha;</mi>
</mrow>
<mo>)</mo>
</mrow>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>t</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mrow>
<mi>n</mi>
<mrow>
<mo>(</mo>
<mi>j</mi>
<mo>)</mo>
</mrow>
</mrow>
</munderover>
<mi>s</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>d</mi>
<mi>t</mi>
</msub>
<mo>)</mo>
</mrow>
<msub>
<mi>V</mi>
<mi>t</mi>
</msub>
</mrow>
<mo>)</mo>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
<mo>+</mo>
<mo>|</mo>
<mo>|</mo>
<mi>U</mi>
<mo>|</mo>
<msubsup>
<mo>|</mo>
<mi>F</mi>
<mn>2</mn>
</msubsup>
<mo>+</mo>
<mo>|</mo>
<mo>|</mo>
<mi>V</mi>
<mo>|</mo>
<msubsup>
<mo>|</mo>
<mi>F</mi>
<mn>2</mn>
</msubsup>
</mrow>
Wherein, U represents user characteristics vector matrix, and V represents Site characterization vector matrix, AijRepresent that i-th of user goes to j-th
The actual value of the number in place, Iij() is indicative function, represents that i-th of user goes to the number in j-th of place to be more than 0, formula
In difference terms represent error between actual value and predicted value,U mould square is represented,Represent V mould square.
5. according to the method for claim 4, it is characterised in that described excellent to the progress of personalized user destination preference pattern
Change, obtain each place of user characteristics vector sum of each user corresponding to Site characterization vector be specially：
L (U, V) minimum value is asked for using gradient descent method, obtains each place pair of user characteristics vector sum of each user
The Site characterization vector answered.
6. the method according to claim 4 or 5, it is characterised in that each according to the user characteristics vector sum of each user
Site characterization vector corresponding to place, each user is calculated is specially to the preference in each place：
Site characterization vector corresponding to each place of user characteristics vector sum of each user is substituted into the personalized user mesh
Ground preference pattern in, obtain preference of each user to each place.
7. according to the method for claim 1, it is characterised in that described that list is escaped generally based on the first user according to the driver
Rate, detect whether the driver produces and escape single act and specifically include：
Escape single probability of the driver based on the first user is input to SVM points as characteristic information together with further feature information
In class device, the SVM classifier assigns different weighted values to every kind of characteristic information of input in the training process, to conduct
Detect whether the driver produces the foundation for escaping single act.
8. according to the method for claim 7, it is characterised in that the one of further feature packet information containing following characteristics
Kind is a variety of：Driver it is total refuse single rate, driver refuses the translational speed information after list, the daily order number of driver.
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 CN104331747A (en) | 2015-02-04 |
| CN104331747B true 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) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106372674B (en) * | 2016-09-30 | 2020-01-21 | 百度在线网络技术（北京）有限公司 | Driver classification method and device in online taxi 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 |
| CN110992044B (en) * | 2019-11-12 | 2022-04-12 | 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 |
|---|
| A system for destination and future route prediction based on trajectory mining;Ling Chen 等;《Pervasive and Mobile Computing》;20100630;第657-676页 * |
| 基于概率的移动用户移动模型;苏孝强 等;《微计算机信息》;20110615;第27卷(第6期);第1-3页 * |
| 基于运动趋势的移动对象位置预测;李雯 等;《通信学报》;20140225;第35卷(第2期);第46-49页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104331747A (en) | 2015-02-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11625755B1 (en) | Determining targeting information based on a predictive targeting model | |
| Xiao et al. | Detecting trip purposes from smartphone-based travel surveys with artificial neural networks and particle swarm optimization | |
| Cui et al. | Forecasting current and next trip purpose with social media data and Google places | |
| Wang et al. | Crowdsensing-based consensus incident report for road traffic acquisition | |
| US10430736B2 (en) | System and method for estimating a dynamic origin-destination matrix | |
| CN104331747B (en) | Malice escapes single detection method | |
| Cang | A comparative analysis of three types of tourism demand forecasting models: Individual, linear combination and non‐linear combination | |
| US20110125691A1 (en) | System and method for output of comparison of physical entities of a received selection and associated with a social network | |
| CN104766215B (en) | A kind of comprehensive, various dimensions owner of cargo selects quantization method | |
| CN105303418A (en) | Site selection method and device | |
| CN106503868A (en) | A kind of fleet's layout of roads method and relevant device | |
| Guan et al. | Early detection of COVID-19 outbreaks using human mobility data | |
| EP3014493A1 (en) | Determining demographic data | |
| Ulusan et al. | Approximate dynamic programming for network recovery problems with stochastic demand | |
| CN107122440A (en) | Power network goods and materials air control early warning system based on public sentiment big data | |
| Chang et al. | Road safety performance across local governments: a data envelopment analysis approach | |
| CN109919675A (en) | Communication user upshift prediction probability recognition methods neural network based and system | |
| JP2003288459A (en) | Method and system for estimating commercial zone demand | |
| Graves et al. | Customer journey time metrics for New York city bus service using big data | |
| WO2021042541A1 (en) | Shopping guide method and apparatus in new retail model, device and storage medium | |
| Baxter | Peer reviewed: making better use of the policies and funding we already have | |
| Gonzalez-Calderon et al. | Service trip generation modeling in urban areas | |
| Sung et al. | Mode-specific impacts of social distancing measures on the intra-and inter-urban mobility of public transit in Seoul during the COVID-19 pandemic | |
| Jiang et al. | Disruption management recovery model of distribution delay with service priority | |
| Sartzetaki | Airport enterprises management performance evaluation towards innovation and sustainable development |
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 |