CN104462187A

CN104462187A - Maximum likelihood ratio-based crowdsourcing data effectiveness verifying method

Info

Publication number: CN104462187A
Application number: CN201410568300.XA
Authority: CN
Inventors: 闻于天; 张奇; 田晓华; 杨峰; 王新兵
Original assignee: Shanghai Jiaotong University
Current assignee: Shanghai Jiaotong University
Priority date: 2014-10-22
Filing date: 2014-10-22
Publication date: 2015-03-25
Anticipated expiration: 2034-10-22
Also published as: CN104462187B

Abstract

The invention provides a maximum likelihood ratio-based crowdsourcing data effectiveness verifying method. The method comprises the steps that the prior probability that an ordinary person not trained misjudges a certain observation component is obtained by an experiment; a server classifies all accumulated data according to the observation value; a probability density function of all data of the same measurement value is estimated and calculated through the nuclear density, and the fiducial probability is calcuatled; the server waits a user to upload new data; a measurer carries out measurement many times through a mobile terminal to obtain a set of data, and the data and the observation components observed by the measurer are uploaded to the server together; the server compares the data provided by the user with a database, and the likelihood reliability of the data is calculated by using the maximum likelihood ratio-based crowdsourcing data effectiveness verifying method; the server decides whether to receive the data or not, remuneration is paid according to the reliability, the database of the measurement value is updated, and the probability density function and the fiducial probability are calculated again.

Description

Crowd sourcing data validity verification method based on maximum likelihood ratio

Technical Field

The invention relates to the technical field of communication, in particular to a crowd sourcing data validity verification method based on a maximum likelihood ratio.

Background

Crowd sourcing (crowdsourceing) has a very broad prospect in smart phone applications. With the rapid development of internet technology, the number of individuals in a network is rapidly increased, and the individuals are more and more closely associated with each other. Under such a large environment, crowd sourcing services are in operation. How to effectively construct a crowd-sourcing service platform and promote resource sharing in the society is an important problem to be solved in the next generation of internet research.

Nowadays, information providers often adopt a crowd sourcing incentive mechanism (crowdsource incorporated mechanism) to deliver the work of collecting information to scattered users and give a certain return for the information or service they provide. For example, if a person wants to know the congestion of a certain road section, the information provided by the user on the road section is faster and more accurate than the information obtained by the provider sending a person to survey. Nowadays, Mobile Phone Sensing technology (Mobile Phone Sensing) is being developed vigorously, and various Sensing devices are being installed on smart phones, such as acceleration sensors, GPS, distance sensors, cameras, etc. It is now a popular approach to acquire the required information and upload it to the provider using the smart phone sensing technology of these distributed users.

Despite the numerous advantages of crowd sourcing, its drawbacks are inevitable. Since the surveyor of the data is not trained professionally, the observation error of the measured data is larger overall, and since the surveyor is not trained, the difference in validity of different data is larger than that of the data obtained by the conventional method. In extreme cases, if the measurer is strange to the tested object and even operates by mistake, the data is seriously deviated from the normal level, and the effectiveness of the sample is damaged by adopting the data.

This is an error that is characteristic in a crowd-sourcing scene, hereinafter referred to as observation error; the remainder are referred to as measurement errors. Both errors can be compensated by a larger sample size, but we aim to quantitatively evaluate and compare crowd-sourcing data by probabilistic methods. Further, the object is to be able to select a portion of the image that is relatively more effective, i.e., a portion of the image that has less observation error.

Through the search of the prior art documents, m.ramadan et al, 2008, proposed a video sharing mechanism based on cooperative downloading in international symposium on Personal, inour and Mobile Radio Communications, "which requires all the participating users to recognize each other and actively form a wireless local area network, and thus the application scenarios are greatly limited. "MicroCast: a video collaboration downloading acceleration mechanism realized by wireless communication between mobile phones is provided in cooperative video streaming on phones. However, this mechanism requires that all participating users wish to download the same video, which is largely unmet in most cases.

Disclosure of Invention

Aiming at the defects in the prior art, the invention aims to provide a maximum likelihood ratio-based crowd sourcing data validity verification method, which can be used for better screening valid data by utilizing a large amount of data contents accumulated in a server database and reducing judgment deviation caused by logging of wrong data.

The invention provides a crowd sourcing data validity verification method based on a maximum likelihood ratio, which comprises the following steps:

step 1: obtaining prior probability p by experiment_ljWherein p is_ljRepresenting the probability that an untrained measurer judges an observed component j as l for the observed component j;

step 2: the server classifies all the accumulated data according to the observation values; for all data of the same measured value j, a probability density function is calculated by using the kernel density estimation, and a confidence probability alpha is calculated_j；

And step 3: the server waits for the user to upload new data;

and 4, step 4: a measurer i uses a mobile terminal thereof to carry out measurement for multiple times to obtain a group of data, and the group of data and an observed component observed by the measurer are uploaded to a server;

and 5: the server compares the data provided by the user with the database and calculates the likelihood reliability of the group of data;

step 6: the server determines whether to accept the group of data and pays a reward according to the reliability; if the server accepts the data, it returns to step 2, updates the database of the measured value j, and reuses the method in step 2 to calculate the probability density function and the confidence probability alpha_j。

Preferably, the step 1 comprises the steps of:

step 1.1: in the training process of indoor positioning based on Wi-Fi signal strength, a measurer needs to determine the indoor position of the measurer and generates an observation error; the observational error of a measurer is abstracted as the estimated error of the distance of the two closest walls of a room when it is at one point in the room;

step 1.2: determining the prior probability p by a preliminary experiment_ljAnd will have a priori probability p_ljThe method is applied to all indoor positioning activities, and particularly comprises the steps of enabling a plurality of measurers to judge the positions l of the measurers at certain fixed points j in a room without a distance reference object, and collecting the judgment result distribution conditions of the plurality of measurers as p_lj；

Step 1.3: for p that cannot be determined by one experiment in advance_ljThe kronecker function may be taken:

<math> <mrow> <msub> <mi>p</mi> <mrow> <mtable> </mtable> <mi>lj</mi> </mrow> </msub> <mo>=</mo> <msub> <mi>δ</mi> <mi>lj</mi> </msub> <mo>=</mo> <mfenced open='{' close=''> <mtable> <mtr> <mtd> <mn>0</mn> </mtd> <mtd> <mi>if</mi> </mtd> <mtd> <mi>l</mi> <mo>&NotEqual;</mo> <mi>j</mi> </mtd> </mtr> <mtr> <mtd> <mn>1</mn> </mtd> <mtd> <mi>if</mi> </mtd> <mtd> <mi>l</mi> <mo>=</mo> <mi>j</mi> </mtd> </mtr> </mtable> </mfenced> </mrow> </math>

wherein,_ljrepresenting the kronecker function.

Preferably, the step 2 comprises the steps of:

step 2.1: each observed component in the server's database corresponds to an accumulated dataset D_jJ 1, 2, 3, N denotes the total number of observed components, D_jEach element D in (1)_j ^kT, obey f ═ 1, 2, 3^j(x) Distribution, T denotes the total number of data per observed component, f^j(x) Representing the probability density function to which the observed component j is subjected; t ═ D_jIf | is > M, M represents the total number of data uploaded by the measurer at one time, then

Wherein, K_hRepresenting a kernel density function, x representing a data variable;

step 2.2: is provided withI.e. n_s(x) Represents [ x-h, x + h]Has been stored in the internal databaseIn the number of data, h represents the kernel density function K_hThe bandwidth of (d);

n_s(x) There may be T +1 values, subject to distribution:

P (n_{s} (x) = n_{s}) = C_{T}^{n_{s}} P {(| x - {D_{j}}^{k} | < h)}^{n_{s}} {(1 - P (| x - {D_{j}}^{k} | < h))}^{T - n_{s}}

wherein P (. cndot.) represents n_s(x) Of the probability mass function, n_s(x) Represents [ x-h, x + h)]The number of data n stored in the internal database_sRepresents a possible value, preferablyAny one of 0, 1, T, T +1,means that n is taken out of T different elements_sThe number of combinations of n, h represents the kernel density function K_hThe bandwidth of (d);

step 2.3: determining r by database size_ilTaking this expectation as the confidence probability α, where r_ilRepresenting the probability density of the data uploaded by the observer i belonging to the observation component l; obviously, the amount of accumulated data for different observations is different, and therefore there is a different confidence probability α for different observations_j。

Preferably, the step 4 comprises the steps of:

step 4.1: the measurer obtains a set of M data as follows

<math> <mrow> <mover> <msub> <mi>x</mi> <mi>ij</mi> </msub> <mo>&RightArrow;</mo> </mover> <mo>=</mo> <mo>{</mo> <msub> <msup> <mi>x</mi> <mn>1</mn> </msup> <mi>i</mi> </msub> <mo>,</mo> <msub> <msup> <mi>x</mi> <mn>2</mn> </msup> <mi>i</mi> </msub> <mo>,</mo> <msub> <msup> <mi>x</mi> <mn>3</mn> </msup> <mi>i</mi> </msub> <mo>,</mo> <mo>.</mo> <mo>.</mo> <mo>.</mo> <mo>,</mo> <msub> <msup> <mi>x</mi> <mi>M</mi> </msup> <mi>i</mi> </msub> <mo>;</mo> <mi>j</mi> <mo>}</mo> <mo>,</mo> </mrow> </math>

Wherein,represents a set of data obtained by measuring the same observed component by a measurer i for a plurality of times, and j representsThe actual value of a component to be observed of the set of M data, j ∈ {1, 2, 3.., N }, where N represents the total number of observed components; x is the number of^t _iDistribution f corresponding to component j^j(x)，x^t _iRepresents the t-th data uploaded by the measurer i;

step 4.2: the observation error is represented by that the measurer reports j as j' to the server, namely

Preferably, the step 5 comprises the steps of:

step 5.1: server obtaining dataPost-computing all { r_il}：

Wherein M represents the total number of data uploaded by a measurer at one time, f (-) represents a probability density function obeyed by the observed component, l represents a possible observed component number, and x^t _ij′Representing the t-th data uploaded by the observer i and judging the t-th data as an observation component j', N representing the total number of the observation components, r_ilHas the physical meaning ofProbability density belonging to the observed component l; obviously, maximum when l ═ j;

step 5.2: defining parameters

<math> <mrow> <mi>L</mi> <mrow> <mo>(</mo> <mover> <msub> <mi>x</mi> <mrow> <mi>i</mi> <msup> <mi>j</mi> <mo>′</mo> </msup> </mrow> </msub> <mo>&RightArrow;</mo> </mover> <mo>)</mo> </mrow> <mo>=</mo> <munder> <mi>max</mi> <mi>l</mi> </munder> <mrow> <mo>(</mo> <mi>log</mi> <mrow> <mo>(</mo> <mfrac> <mrow> <msub> <mi>r</mi> <mi>il</mi> </msub> <msub> <mi>p</mi> <msup> <mi>lj</mi> <mo>′</mo> </msup> </msub> </mrow> <msub> <mi>α</mi> <mi>j</mi> </msub> </mfrac> <mo>)</mo> </mrow> <mo>)</mo> </mrow> </mrow> </math>

Wherein alpha is_jCalled confidence probability, p_lj′Representing the probability that the measurer judges the observed component j 'as the observed component l for the observed component j'; when alpha is_jWhen 1 is trueIn the sense of a logarithm of the maximum possible probability density of the measurement data; it is apparent that for a set of data of the same length,larger ones are more reliable;

step 5.3: by passingThe effectiveness of all the crowd's intelligence data can be sequenced, and the first few of the crowd's intelligence data can be selected as required.

Preferably, the first and second electrodes are formed of a metal,

in step 2.1, the kernel density function is taken as the uniform kernel function:h is small enough that the probability P that the data falls within this region is approximately evenly distributed over the bandwidth_s＝P(|x-D_j ^k|＜h)＝f(x)2h；

In step 2.3, allThe data of (a) have utility values, as follows is a calculation of r_ilDesired E r_ilThe method of (1):

wherein f is^l(x^t) Representing the value of the observed component l as x^tL represents the ith observed component, t represents the tth data uploaded by the observer, M represents the total number of data uploaded by the measurer at one time! Denotes factorial, e denotes natural base, P_s＝P(|x-D_j ^k|＜h)＝f(x_i)2h，f(x_i) Estimating by using the nuclear density; in the above formula, since there is no variable other than T, the confidence probability α is determined_jThe relation to the database size T.

Compared with the prior art, the invention has the following beneficial effects:

1. the invention can correct the judgment error of the crowd data observer through the preliminary experiment;

2. the method can evaluate the effectiveness of the new crowd sourcing data based on the existing reliable data set, thereby reasonably making effective choices for the new crowd sourcing data.

Drawings

Other features, objects and advantages of the invention will become more apparent upon reading of the detailed description of non-limiting embodiments with reference to the following drawings:

FIG. 1 is a flow chart of the steps of the present invention.

Detailed Description

The present invention will be described in detail with reference to specific examples. The following examples will assist those skilled in the art in further understanding the invention, but are not intended to limit the invention in any way. It should be noted that variations and modifications can be made by persons skilled in the art without departing from the spirit of the invention. All falling within the scope of the present invention.

The invention provides a crowd sourcing data validity verification method based on a maximum likelihood ratio, which comprises the following steps: obtaining the prior probability of a untrained ordinary person judging a certain observation component wrongly by an experiment; the server classifies all the accumulated data according to the observation values; calculating a probability density function by using nuclear density estimation on all data of the same measured value, and calculating a confidence probability; the server waits for the user to upload new data; a measurer uses a mobile terminal to carry out measurement for multiple times to obtain a group of data, and the data and an observed component observed by the measurer are uploaded to a server; the server compares the data provided by the user with the database, and calculates the likelihood reliability of the group of data by using a maximum likelihood ratio-based crowd-sourcing data validity verification method; the server determines whether to accept the set of data, pays a reward based on the reliability, updates the database of measurements, and recalculates the probability density function and the confidence probability.

The invention provides a maximum likelihood ratio-based crowd-sourcing data validity verification method, which can be used for better screening valid data by utilizing a large amount of data contents accumulated in a server database and reducing judgment deviation caused by logging error data.

Referring to the attached figure 1, the invention is realized by the following technical scheme, and comprises the following steps:

the first step is as follows: obtaining prior probability p by experiment_ljAnd represents the probability that an untrained average person judges it as l for a certain observed component j.

The second step is that: the server classifies all data that has been accumulated according to observations. For all data of the same measured value j, a probability density function is calculated by using the kernel density estimation, and a confidence probability alpha is calculated_j。

The third step: the server waits for the user to upload new data.

The fourth step: the measurer i uses the mobile terminal to carry out measurement for multiple times to obtain a group of data, and the data is uploaded to the server together with the observed component observed by the measurer.

The fifth step: the server compares the user-provided data with the database and calculates the likelihood reliability of the set of data using a maximum likelihood ratio-based crowd-sourcing data validity verification method.

And a sixth step: the server determines whether to accept the group of data and pays a reward according to the reliability; if the server accepts the data, it returns to step 2, updates the database of the measured value j, and reuses the method in step 2 to calculate the probability density function and the confidence probability alpha_j。

The practice of the present invention is explained in more detail below.

Step one, it is assumed that the server needs to measure a measurement value through crowd sourcing data, and the measurement value comprises a plurality of observation components. Subject to observation errors, the measurer takes a probability p_ljA certain observed component j is misjudged as another observed component l. The experiment first obtains the prior probability p_lj。

For example, in the training process of indoor positioning based on Wi-Fi signal strength, a measurer needs to determine the position of the measurer in a room, and observation errors are generated. The observational error of a measurer can be abstracted as the estimated error of the distance to the two closest walls of a room when it is at a point in the room. This distribution p can be determined by a preliminary experiment_ljAnd applies it to all indoor positioning activities. A large number of volunteers are recruited to judge the position l of the volunteers at certain fixed points j in a room without a significant distance reference object, and the distribution condition of the judgment results is collected to be regarded as p_lj。

P if not determinable by a previous experiment_ljA Kronecker Delta function can be taken.

Step two, each observation component in the database of the server corresponds to an accumulated data set D_j1, 2, 3, N, wherein each element D is a member of the group_j ^kT, obey f ═ 1, 2, 3^j(x) Distribution, T ═ D_jAnd | is the size of the data set. It is assumed that f can be recovered with sufficient accuracy by nuclear density estimation^j(x) In that respect Then

The kernel density function may take any other formVarious changes or modifications may be made by those skilled in the art within the scope of the claims without departing from the spirit of the invention. For example, take the kernel density function as a uniform kernel function:h is small enough that the probability P that the data falls within this region is approximately evenly distributed over the bandwidth_s＝P(|x-D_j ^k|＜h)＝f^j(x)2h。

Is provided withI.e., [ x-h, x + h]The number of data already existing in the internal database. n is_s(x) There may be T values whose distribution satisfies

P (n_{i} (x) = n_{s}) = C_{T}^{n_{s}} P {(| x - {D_{j}}^{k} | < h)}^{n_{s}} {(1 - P (| x - {D_{j}}^{k} | < h))}^{T - n_{s}}

Since different observation components accumulate different amounts of data, different observation components have different confidence probabilities α_j. Confidence probability alpha_jMethod for measuring adoption value of data uploaded by userWhereinIndicating that user i uploads data and that user judges it as an observed component j'. If use r_ilTo representProbability density of the component l under observation, then r_ilThe expectation of (a) can be taken as the confidence probability a. The following is a calculation E { r }_ilThe method of (1).

Wherein P is_s＝P(|x-D_j ^k|＜h)＝f(x_i)2h，f(x_i) Estimated by nuclear density. In the formula, there is no variable other than T, so that the confidence probability alpha is determined_jThe relation to the database size T.

And step three, the server waits for the user to upload new data.

Step four, the measurer i obtains a group of M data for a certain measurement component and records the following formula

j represents the true value of the set of data measurement components, j ∈ {1, 2, 3. x is the number of^t _iDistribution f corresponding to component j^j(x) In that respect The observation error is embodied as that the measurer judges the observation component j as j 'and reports the j' to the server, namely

Step five, the server obtains the dataPost-computing all { r_il}：

Obviously, it is maximum when l ═ j. Defining parameters

By passingThe system can rank the effectiveness of all the crowd's wisdom data, and take the first few of them as required.

The environmental parameters of this embodiment are:

a mobile terminal device: six Android smart phones are all Nexus 4, and each smart phone is provided with a 1.5GHz Snapdragon APQ8064 CPU and 2G RAM, and the operating systems of the six smart phones are all Android JellyBean (4.2). The six smart phones are used as test phones in parallel for indoor positioning.

A server: a macro 4930G notebook computer, a core processor, a 2G memory and a 2G main frequency.

The foregoing description of specific embodiments of the present invention has been presented. It is to be understood that the present invention is not limited to the specific embodiments described above, and that various changes and modifications may be made by one skilled in the art within the scope of the appended claims without departing from the spirit of the invention.

Claims

1. A crowd sourcing data validity verification method based on maximum likelihood ratio is characterized by comprising the following steps:

And step 3: the server waits for the user to upload new data;

2. The method for verifying the validity of the crowd sourcing data based on the maximum likelihood ratio as claimed in claim 1, wherein the step 1 comprises the steps of:

step 1.2: determining the prior probability p by a preliminary experiment_ljAnd will have a priori probability p_ljApplied to all indoor positioning activities; specifically, a plurality of measuring persons judge their own positions l at certain fixed points j in a room without a distance reference object, and the distribution of the judgment results of the plurality of measuring persons is collected as p_lj；

Step 1.3: for p that cannot be determined by one experiment in advance_ljTaking the kronecker function:

<math> <mrow> <msub> <mi>p</mi> <mi>lj</mi> </msub> <mo>=</mo> <msub> <mi>δ</mi> <mi>lj</mi> </msub> <mo>=</mo> <mfenced open='{' close=''> <mtable> <mtr> <mtd> <mn>0</mn> <mi>if l</mi> <mo>&NotEqual;</mo> <mi>j</mi> </mtd> </mtr> <mtr> <mtd> <mn>1</mn> <mi>if l</mi> <mo>=</mo> <mi>j</mi> </mtd> </mtr> </mtable> </mfenced> </mrow> </math>

wherein,_ljrepresenting the kronecker function.

3. The method for verifying the validity of the crowd sourcing data based on the maximum likelihood ratio as claimed in claim 1, wherein the step 2 comprises the steps of:

step 2.2: is provided withI.e. n_s(x) Represents [ x-h, x + h]The number of data in the internal database is stored, h represents a kernel density function K_hThe bandwidth of (d);

n_s(x) There may be T +1 values, subject to distribution:

P (n_{s} (x) = n_{s}) = C_{T}^{n_{s}} P {(| x - {D_{j}}^{k} | < h)}^{n_{s}} {(1 - P (| x - {D_{j}}^{k} | < h))}^{T - n_{s}}

whereinP (-) represents n_s(x) Of the probability mass function, n_s(x) Represents [ x-h, x + h)]The number of data n stored in the internal database_sTaking one value of 0, 1.. times, T, T +1,means that n is taken out of T different elements_sThe number of combinations of n, h represents the kernel density function K_hThe bandwidth of (d);

4. The method for verifying the validity of the crowd-sourcing data based on the maximum likelihood ratio according to any one of claims 1 to 3, wherein the step 4 comprises the steps of:

step 4.1: the measurer obtains a set of M data as follows

<math> <mrow> <msub> <mover> <mi>x</mi> <mo>&RightArrow;</mo> </mover> <mi>lj</mi> </msub> <mo>=</mo> <mo>{</mo> <msub> <msup> <mi>x</mi> <mn>1</mn> </msup> <mi>i</mi> </msub> <mo>,</mo> <msub> <msup> <mi>x</mi> <mn>2</mn> </msup> <mi>i</mi> </msub> <mo>,</mo> <msub> <msup> <mi>x</mi> <mn>3</mn> </msup> <mi>i</mi> </msub> <mo>,</mo> <mo>.</mo> <mo>.</mo> <mo>.</mo> <mo>,</mo> <msub> <msup> <mi>x</mi> <mi>M</mi> </msup> <mi>i</mi> </msub> <mo>;</mo> <mi>j</mi> <mo>}</mo> <mo>,</mo> </mrow> </math>

Wherein,representing a group of data obtained by a measurer i by measuring the same observation component for multiple times, wherein j represents the real value of a component needing to be observed of the group of M data, j belongs to {1, 2, 3., N }, and N represents the total number of the observation components; x is the number of^t _iDistribution f corresponding to component j^j(x)，x^t _iRepresents the t-th data uploaded by the measurer i;

5. The method for verifying the validity of the crowd sourcing data based on the maximum likelihood ratio as claimed in claim 4, wherein the step 5 comprises the steps of:

step 5.1: server obtaining dataPost-computing all { r_il}：

Wherein M represents one measurementTotal number of data uploaded, f (-) represents probability density function obeyed by observation component, l represents observation component number, x^t _ij′Representing the t-th data uploaded by the observer i and judging the t-th data as an observation component j', N representing the total number of the observation components, r_ilHas the physical meaning ofProbability density belonging to the observed component l; obviously, maximum when l ═ j;

step 5.2: defining parameters

<math> <mrow> <mi>L</mi> <mrow> <mo>(</mo> <msub> <mover> <mi>x</mi> <mo>&RightArrow;</mo> </mover> <msup> <mi>lj</mi> <mo>′</mo> </msup> </msub> <mo>)</mo> </mrow> <mo>=</mo> <munder> <mi>max</mi> <mi>l</mi> </munder> <mrow> <mo>(</mo> <mi>log</mi> <mrow> <mo>(</mo> <mfrac> <mrow> <msub> <mi>r</mi> <mi>il</mi> </msub> <msub> <mi>p</mi> <msup> <mi>lj</mi> <mo>′</mo> </msup> </msub> </mrow> <msub> <mi>α</mi> <mi>j</mi> </msub> </mfrac> <mo>)</mo> </mrow> <mo>)</mo> </mrow> </mrow> </math>

6. The maximum likelihood ratio-based crowd-sourcing data validity verification method of claim 5,