CN102693335A - User interest model classification method based on Gene coefficient measurement - Google Patents

Abstract

The invention discloses a user interest model classification method based on Gene coefficient measurement and relates to the technical field of computer modeling. The method comprises the steps of: S1, constructing user interest models based on a vector space model (VSM); S2, sorting the user interest models in ascending order of user interest degree to obtain the sorted interest degree U<Sorted><User>; S3, converting the interest degree U<Sorted><User> in S2 into percentage of overall interest of users themselves, U<%><User>; S4, calculating U<Gini><User> of each user according to the U<%><User>, wherein the U<Gini><User> is a vector for generating a Lorenz curve; and S5, obtaining a Lorenz curve of the user interest model by taking value of the U<Gini><User> as ordinate and interest degree of fields ranked from low to high as abscissa, calculating Gene coefficient, and classifying user interest models based on the Gene coefficient. The user interest model classification method provided by the invention enables accurate classification of the user interest models to be realized.

Description

The user interest mode division method of estimating based on Gini coefficient

Technical field

The present invention relates to the microcomputer modelling technical field, particularly a kind of user interest mode division method of estimating based on Gini coefficient.

Background technology

The user interest pattern is diversified, and some user belongs to the extensive type of interest, its to each field to like degree to distribute relatively even; Some user then is the single-minded things of liking few classification, and it likes the distribution of degree relatively also inhomogeneous to each field.Many times; Even if different user Cup of tea thing is different, but have identical interest mode, be audiophile's delight in music such as a user; Another user is that military fan only likes military affairs; Although music differs widely with military affiliated field, these two users are single interest pattern users, and they have the same interest pattern.And the research of at present relevant user interest lacks the measure to the user interest pattern.Therefore be necessary to find a kind of method of estimating the user interest pattern.

Pattern classification has very big reference to Gini coefficient for the research user interest in the economics.Gini coefficient is a kind of measure of the rich or poor difference of assessment general in the world in economics society; The distribution situation that is social gross income in all population of tolerance, this with the research tolerance user interest of user interest pattern between each field relatively the problem of distribution very big similarity is arranged.

Distribute unequal problem in order to study the wealth of society, 1905 U.S. statistician M.O. long-range navigation thatch (Max Otto Lorenz) famous lorenz curve (lurenz curve) has been proposed.Lorenz curve in the economics (solid wire among Fig. 1) is a kind of wealth distribution accumulative total graph of function shape method for expressing.

(x%, y%), its implication is that the poor ratio that accounts for social gross income to rich accumulative total gross income of arranging the population of preceding x% is y% for any point on the lorenz curve among Fig. 1.The social gross income of " absolute fair line " (curve of absolute equality) expression distribution of earnings curve during absolute mean allocation in the city is the straight line of " y=x " among the figure; " fair anything but line " (Curve of absolute inequality) is the distribution of earnings curves of all incomes of society during by unique the occupying of people, is a straight line perpendicular to the x axle.Lorenz curve is generally between absolute fair line and fair anything but line.

1912, Italian economist's Geordie proposed Gini coefficient (Gini coefficient) according to lorenz curve.Gini coefficient is as the index of estimating a variable distribution concentration degree (unequal character), and commonly used its measured gap between the rich and the poor in the modern economics.As shown in Figure 1, the area of establishing between lorenz curve and the absolute fair line is A, and lorenz curve is B with the graphics area that fair line anything but and x axle surround.And be Gini coefficient divided by the unequal degree of the quotient representation of A+B with A.Be expressed as with following formula (1):

$\mathrm{Gini\; coefficient}=\frac{A}{A+B}---\left(1\right)$

This numerical value is called as Gini coefficient or claims long-range navigation thatch coefficient, and size is between 0 to 1.Area A between lorenz curve and the absolute fair line is more little, and distribution of earnings tends to equality, and the radian of lorenz curve is also just more little, and Gini coefficient is also more little; Otherwise it is unequal that distribution of earnings tends to, and the radian of lorenz curve is big more, and Gini coefficient is also big more so.

Gini coefficient is when the tolerance gap between the rich and the poor, and the gross income that its essence hypothesis is a society is a homogeneity, measures its distribution situation at all population.

Prior art concentrates on according to user interest similarity research user preference.Shortage can't be divided the user interest pattern from user interest pattern angle research user interest exactly.

Summary of the invention

The technical matters that (one) will solve

The technical matters that the present invention will solve is: how exactly the user interest pattern to be divided.

(2) technical scheme

For solving the problems of the technologies described above, the invention provides and a kind ofly estimate the user interest division methods based on Gini coefficient, may further comprise the steps:

S1: based on vector space model VSM framework user interest model, the set of user interest field is T={ interest 1, interest 2 ...; Interest N}, for any one user, its user interest model can be expressed as U={ < interest 1, interest-degree 1 >; ..., < interest N, interest-degree N>};

S2: user interest model is sorted according to user interest degree ascending order, and the user interest degree after obtaining sorting is:

$U_{\mathrm{User}}^{\mathrm{Sorted}}=\{w_1^{\mathrm{sorted}},w_2^{\mathrm{sorted}},...,w_N^{\mathrm{Sorted}}\};$

S3: the step S2

interestingness

into account the degree of interest the users themselves overall percentage:

$U_{\mathrm{User}}^{\%}=\{w_1^{\%},w_2^{\%},...,w_N^{\%}\};$

S4:

that calculates each user according to

is for generating the vector of lorenz curve, wherein:

$w_i=\underset{j=1}{\overset{i}{\mathrm{\&Sigma;}}}{w}_j^{\%}(1\&le;i\&le;N);$

S5: the value with

is an ordinate; Interest-degree with the field is arranged as horizontal ordinate from low to high; Draw the lorenz curve of user interest pattern; And the calculating Gini coefficient, by Gini coefficient division user's interest mode.

Wherein, among the said step S5, calculate Gini coefficient by following formula:

$G_{\mathrm{User}}=1-\frac{\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}[\frac{1}{2}\&times;(w_i+w_{i-1})\&times;\frac{1}{N}]}{\frac{1}{2}}=1-\frac{1}{N}\&times;\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}(w_i+w_{i+1}).$

Wherein, the more little user of said Gini coefficient difference, interest mode is similar more.

(3) beneficial effect

The present invention is through utilizing the qualitative and quantitative method of estimating the user interest pattern of lorenz curve and Gini coefficient, and it is more accurate to the division of user interest pattern to make.

Description of drawings

Fig. 1 is the lorenz curve synoptic diagram;

Fig. 2 is a kind of user interest mode division method flow diagram of estimating based on Gini coefficient of the embodiment of the invention;

Fig. 3 is the lorenz curve synoptic diagram of expression user A, B, C interest among the embodiment;

Fig. 4 is the ladder approximation decomposing schematic representation of the lorenz curve of expression user A interest among the embodiment;

Fig. 5 is 943 user's lorenz curves of movielens data centralization synoptic diagram;

Fig. 6 is 943 user's Gini coefficients of movielens data centralization frequency distribution schematic diagram;

Fig. 7 is the Gini coefficient distribution situation that the Movielens data centralization is divided according to reference user.

Embodiment

Below in conjunction with accompanying drawing and embodiment, specific embodiments of the invention describes in further detail.Following examples are used to explain the present invention, but are not used for limiting scope of the present invention.

The user interest mode division method flow of estimating based on Gini coefficient of the present invention is as shown in Figure 1, comprising:

Step S201, (set of user interest field is T={ interest 1 for Vector Space Model, VSM) framework user interest model based on vector space model; Interest 2 ..., interest N}; For any one user, his user interest can be expressed as U={ < interest 1, interest-degree 1 >; ..., < interest N, interest-degree N>}.

U _user＝{<theme ₁，weight ₁>，...，<theme _N，weight _N>}

Theme is the theme among the corresponding set T, and weight is that the user is big or small to corresponding field interest-degree, is to set the weight of user to certain field interest level in advance, and the user is to the interested degree in certain field in expression, but also direct representation becomes:

U _user＝{weight ₁，weight ₂，...，weight _N}。

Step S202 sorts according to user interest degree ascending order to user interest model, and the user interest after obtaining sorting is:

$U_{\mathrm{User}}^{\mathrm{Sorted}}=\{w_1^{\mathrm{sorted}},w_2^{\mathrm{sorted}},...,w_N^{\mathrm{Sorted}}\}$

Wherein $w_1^{\mathrm{Sorted}}\&le;w_2^{\mathrm{Sorted}}\&le;,...,\&le;w_N^{\mathrm{Sorted}},$

Step S203; It is obvious then

interest-degree in second step to be changed into the number percent that takies the total interest degree in family own, and the interior numerical value sum of each user vector all is 100% after the number percentization:

$U_{\mathrm{User}}^{\%}=\{w_1^{\%},w_2^{\%},...,w_N^{\%}\}$

Wherein:

$\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{w}_i^{\mathrm{Sorted}}=100\%$

Step S204 is according to what calculate among the S203

Calculate each user's

Be with the vector that generates lorenz curve, w _i, (the 1st group number percent is added to the number percent that i group interest-degree sum accounts for all interest-degrees among the expression of 1≤i≤n) the expression step S202.That is:

$w_i=\underset{j=1}{\overset{i}{\mathrm{\&Sigma;}}}{w}_j^{\mathrm{Sorted}}(1\&le;i\&le;N)$

Step S204; Value with is an ordinate; Be arranged as horizontal ordinate by interested hanging down to height with the field; Can described point draw the lorenz curve of user interest pattern, Gini coefficient is:

$G_{\mathrm{User}}=1-\frac{\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}[\frac{1}{2}\&times;(w_i+w_{i-1})\&times;\frac{1}{N}]}{\frac{1}{2}}=1-\frac{1}{N}\&times;\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}(w_i+w_{i+1})$

Concrete reasoning process to Gini coefficient is exemplified below:

In the application Geordie systematic measure user interest mode process of deriving, simulate the user of three known types earlier, provide its Gini coefficient computation process.

As shown in table 1, there is one and is named as user's note of " just rich " to make user A, can find out intuitively that from table 1 his characteristics are that every field is all preferred; Another one user is " specialized personnel one ", and note is made user B, and it is concerned about physical culture, finance and economics and military field, the news of never reading other field; Also have a user " specialized personnel two ", note is made user C, user C basic trend-conscious, health and education sector.Numerical value in the table 1 is its number of reading the association area article, and as previously mentioned, the number of news that the user reads is equal to the interest-degree of user to this field.

Three typical users of table 1 interest-degree distribution situation

After carrying out the user interest modeling according to vector space model, three users' interest can be expressed as with vector form respectively:

U _A＝{5，3，4，7，6，2，8，4}

U _B＝{8，10，0，3，0，0，0，0}

U _C＝{0，0，1，0，0，5，12，7}

Can find out the difference of three user interest patterns according to these three vectors intuitively, obviously the interest-degree of user A distributes average more than user B, C.Next adopt Gini coefficient to come quantitative measurement user's interest mode, and can carry out qualitative observation through lorenz curve.

The computation process of user interest pattern Gini coefficient is:

The first step: the vector model of A, B and user C after obtaining successively sorting by each user interest degree ascending sort:

$U_A^{\mathrm{Sorted}}=\left\{\mathrm{2,3,4,4,5,6,7,8}\right\}$

$U_B^{\mathrm{Sorted}}=\left\{\mathrm{0,0,0,0,0,3,8,10}\right\}$

$U_C^{\mathrm{Sorted}}=\{\mathrm{0,0,0,0},\mathrm{1,5,7,12}\}$

Second step: the interest-degree in the vector in the first step is changed into the number percent that takies the total interest degree in family own, and obviously, the interior numerical value sum of each user vector all is 100% after the number percentization, that is:

$U_A^{\%}=\{5.2\%,7.7\%,10.3\%,10.3\%,12.8\%,15.4\%,17.9\%,20.5\%\}$

$U_B^{\%}=\{\mathrm{0,0,0,0,0,14.3}\%,30.1\%,55.6\%\}$

$U_C^{\%}=\{\mathrm{0,0,0,0,4}\%,20\%,28\%,48\%\}$

The 3rd step:, also need to calculate each user's according to second step for calculating Gini coefficient

Be that n representes the sum (n=8 here) in field, w with the vector that generates lorenz curve _i, (1≤i≤n) promptly:

$U_A^{\mathrm{Gini}}=\{5.2\%,12.9\%,23.1\%,33.4\%,46.2\%,61.6\%,79.5\%,100\%\}$

$U_B^{\mathrm{Gini}}=\{\mathrm{0,0,0,0,0,14.3}\%,44.4\%,100\%\}$

$U_C^{\mathrm{Gini}}=\{\mathrm{0,0,0,0,4}\%,24\%\%,52\%\%,100\%\}$

Then with vectorial U ^GiniValue be ordinate, be horizontal ordinate with all fields, the curve that described point generates is exactly the lorenz curve of user A, B and user C.As shown in Figure 3, the some bar line near " absolute fair line " is the lorenz curve of user A, and dotted lines is the lorenz curve of user B, and the stripline runs between user A, the B is the lorenz curve of user C.What should pay special attention to is that for these three users, the field of horizontal ordinate is just to arrange by the unique user fancy grade here; Increase progressively successively from left to right, might not be identical for the different user specific field, field 8 in Fig. 3 horizontal ordinate for example; Only representing the favorite field of unique user, is its favorite health for user A field 8, and for user B; Field 8 is its favorite finance and economics, and concerning these two users, the particular content in field 8 is also different; But identical is healthy and finance and economics is respectively user A, the favorite field of B, and promptly two users are identical to the field rank of 8 fancy grades in all spectra.

For calculating three users' Gini coefficient, be example with user A.As shown in Figure 4, area is defined as S between the lorenz curve of user A and absolute average line _A, and the area that forms between horizontal ordinate and the fair anything but line is S _B, can get according to formula (1), the Gini coefficient of user A is formula (2):

$G_A=\frac{S_A}{S_A+S_B}=\frac{(S_A+S_B)-S_B}{S_A+S_B}=1-\frac{S_B}{S_A+S_B}---\left(2\right)$

Interest worlds totally gets 100%, obvious S in the horizontal ordinate _A+ S _BArea equal 0.5.

Calculating S _BThe time, because actual lorenz curve is the line of a bending, directly reference area can only adopt certain methods to be similar to.As shown in Figure 4, among this paper with S _BBeing approximately n the width with shared total field, field between i group and the i-1 group is the end, with the accumulative total interest-degree w of i group _iWith i-1 group accumulative total interest-degree w _I-1Trapezoidal area sum for last bottom.In this general hypothesis n field arranged,

For the user: $U_{\mathrm{User}}^{\mathrm{Gini}}=\{{\mathrm{<figure-callout\; id="1"\; label="w"\; filenames="HDA0000159056190000011.png,HDA0000159056190000031.png"\; state="\{\{state\}\}">w</figure-callout>}}_1,{\mathrm{<figure-callout\; id="2"\; label="w"\; filenames="HDA0000159056190000031.png"\; state="\{\{state\}\}">w</figure-callout>}}_2,...,w_i,...,w_n\}(1\&le;i\&le;n;{\mathrm{<figure-callout\; id="0"\; label="w"\; filenames="HDA0000159056190000031.png,HDA0000159056190000051.png"\; state="\{\{state\}\}">w</figure-callout>}}_0=0),$ Its Gini coefficient computing formula is:

$G_{\mathrm{User}}=1-\frac{\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}[\frac{1}{2}\&times;(w_i+w_{i-1})\&times;\frac{1}{N}]}{\frac{1}{2}}=1-\frac{1}{N}\&times;\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}(w_i+w_{i+1})---\left(3\right)$

The Gini coefficient that can calculate user A, B and user C according to formula (3) is respectively: G _A=0.22, G _B=0.73, G _C=0.69.

This shows that the interest Gini coefficient of user A " just rich " is 0.22, is similar to 0.2, this user interest of the smaller explanation of Gini coefficient distributes even relatively, belongs to hobby one type of people widely; And the Gini coefficient of user B " specialized personnel one " is 0.73, and numeric ratio is bigger, and it is also inhomogeneous to explain that this user distributes, and belongs to the single-minded one type of narrow people of hobby; User C interest distributes also inhomogeneous, belongs to the narrow crowd of hobby equally.Can also find out intuitively on the lorenz curve that from Fig. 3 the lorenz curve of user A more near " absolute average line ", explain that user A is more even than user B in the distribution in all fields in interest than user B and user C.And the lorenz curve of user B and user C is very approaching, shows that it has similar interest mode.

To method of the present invention experiment simulation checking, proved feasibility, detailed process and simulation result are following:

For the effect of actual verification Gini coefficient in tolerance user interest pattern, chosen the checking that experimentizes of Movielens data set.The Movielens data set is user's film score data that the Grouplens tissue is gathered from the movielens website.The data centralization that adopts comprises 100,000 scorings (1～5 minute) of 943 users to 1682 films, and each user had scoring to 20 films at least.

Every film on the Movielens data set all belongs to a type or several types in 18 fields (comedy, action movie, romance movie etc.); Type with these 18 kinds of films is a user interest field overall space, investigates the distribution situation of user interest in these 18 types of film hobbies.For the calculating of user interest degree, adopt the user is added to the field under this film to the scoring of having seen a film, try to achieve the interest-degree of user respectively to these 18 types of films.

Find in the actual experiment; When investigating the user preferences distribution with these whole 18 types; Nearly all user's Gini coefficient is all near 0.8, and discrimination is very low, and this shows that almost not have the user all interested in 18 types film; This also is consistent with real daily life because general interest widely people's class of dabbling also seldom can cover all.

Confirm to find the solution the overall quantity in Gini coefficient field according to 80/20 principle for this reason.It is 80% that the main interest of supposing the user all accounts for its interest-degree population proportion; The mean value of obtaining 943 users preceding 80% the shared field of quantity, field, main interest place quantity then is 7.14, therefore when asking lorenz curve and calculating the user interest Gini coefficient, only investigates the hobby situation of user to the highest preceding 7 types of films of its interest.Fig. 5 is 943 users' drawing in view of the above a lorenz curve (horizontal ordinate is similar to Fig. 3, the field that the rightmost side is most interested in for the user, from left to right, level of interest increases progressively successively), and Fig. 6 is that corresponding user's Gini coefficient frequency distributes.

The frequency of 943 user's Gini coefficients distribution match normal distribution substantially among Fig. 6.In order to understand the implication of these numerical value more accurately, the Gini coefficient that provides 5 reference user is as reference.Suppose that the user meets 80/20 principle in the interest distribution in 7 kinds of most interested fields, the user's interest field takies 80% of family total interest, and all the other fields account for 20% altogether.For example for a two interest user, this user has only two main interest, and the user accounts for the half the of total amount 80% proportion respectively to these two field interest-degrees, and promptly 40%, remain five fields of loseing interest in and divide equally and remain 20% interest-degree, promptly 4%.

Can obtain following five types of reference user and corresponding fancy grade distribution condition thereof so respectively, as shown in table 2:

Table 2 five types of reference user interest-degrees distribution situation and Gini coefficient

Reference user	Interest-degree distributes	Gini coefficient
			Single interest user	{33％，33％，33％，33％，33％，33％，80％}	014
Two interest users	{4％，4％，4％，4％，4％，40％，40％}	024
			Three interest users	{5％，5％，5％，5％，266％，266％，266％}	038
Four interest users	{66％，66％，66％，20％，20％，20％，20％}	052
			Five interest users	{10％，10％，16％，16％，16％，16％，16％}	063

User's Gini coefficient frequency among Fig. 6 is repartitioned according to above five types of reference user standards respectively, and statistics draws user's Gini coefficient distribution situation, and is as shown in Figure 7.

As can be seen from the figure, Gini coefficient is very effective to the results of user interest pattern, and in 943 users that investigated, maximum is, and about 44.5% user has 3 to 4 interest worlds; 24.4% user has 4 to 5 main interest, and 22.6% user has 2 to 4 interest, and 6.5% user only has 1 to 2 main interest; 1.2% artificial absolute single interest mode type has only the people of only a few 0.8% to like interest worlds and reaches more than 5.

Above experimental data shows, uses Gini coefficient tolerance user interest pattern and has good discrimination, can observe qualitatively through lorenz curve, also can carry out quantitative analysis through Gini coefficient.

Above embodiment only is used to explain the present invention; And be not limitation of the present invention; The those of ordinary skill in relevant technologies field under the situation that does not break away from the spirit and scope of the present invention, can also be made various variations and modification; Therefore all technical schemes that are equal to also belong to category of the present invention, and scope of patent protection of the present invention should be defined by the claims.

Claims (3)

1. a user interest mode division method of estimating based on Gini coefficient is characterized in that, may further comprise the steps:

S1: based on vector space model VSM framework user interest model, the set of user interest field is T={ interest 1, interest 2 ...; Interest N}, for any one user, its user interest model can be expressed as U={ < interest 1, interest-degree 1 >; ..., < interest N, interest-degree N>};

S2: user interest model is sorted according to user interest degree ascending order, and the user interest degree after obtaining sorting is:

$U_{\mathrm{User}}^{\mathrm{Sorted}}=\{w_1^{\mathrm{sorted}},w_2^{\mathrm{sorted}},...,w_N^{\mathrm{Sorted}}\};$

S3: the step S2

interestingness

into account the degree of interest the users themselves overall percentage:

$U_{\mathrm{User}}^{\%}=\{w_1^{\%},w_2^{\%},...,w_N^{\%}\};$

S4:

that calculates each user according to is for generating the vector of lorenz curve, wherein:

$w_i=\underset{j=1}{\overset{i}{\mathrm{\&Sigma;}}}{w}_j^{\%}(1\&le;i\&le;N);$

S5: the value with

is an ordinate; Interest-degree with the field is arranged as horizontal ordinate from low to high; Draw the lorenz curve of user interest pattern; And the calculating Gini coefficient, by Gini coefficient division user's interest mode.

2. the user interest mode division method of estimating based on Gini coefficient as claimed in claim 1 is characterized in that, among the said step S5, calculates Gini coefficient by following formula:

$G_{\mathrm{User}}=1-\frac{\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}[\frac{1}{2}\&times;(w_i+w_{i-1})\&times;\frac{1}{N}]}{\frac{1}{2}}=1-\frac{1}{N}\&times;\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}(w_i+w_{i+1}).$

3. the user interest mode division method of estimating based on Gini coefficient as claimed in claim 1 is characterized in that, the more little user of said Gini coefficient difference, and interest mode is similar more.

Images