CN104462385A - Personalized movie similarity calculation method based on user interest model - Google Patents

Abstract

The invention discloses a personalized movie similarity calculation method based on a user interest model. The method includes: according to users' historical behaviors, namely through a personalized movie recommendation system platform, mining users' various behaviors of searching a movie resource library and users' watching and collecting behaviors; fully mining and analyzing different preference degrees of different users, upon six basic attributes including performers, directors, types, regions, times and content introductions of movies, thus acquiring a first-layer six-dimensional spatial vector representation of a user model; according to the users' behaviors above, by means of keyword extraction or semantic analysis, analyzing the different users' weights of characteristic values in the six dimensions, thus acquiring a second-layer six-dimensional spatial vector representation of the user model; using a two-layer multi-dimensional spatial vector to represent the user interest model, generating different movie similarity lists for the different users on the basis of the user interest model and basic content characteristics of movies. Therefore, recommending is more effective.

Description

The personalized similarity calculating method of a kind of film based on user interest model

Technical field

The present invention relates to the personalized similarity calculating method of a kind of film based on old user's interest model and the personalized similarity calculating method of a kind of film for new user.

Background technology

Along with the develop rapidly of internet, in the face of the magnanimity movie resource day by day upgraded, personalized recommendation application is also just random to produce.

The proposed algorithm of current Major Epidemic has: based on the recommendation of correlation rule, Knowledge based engineering recommendation, content-based recommendation, collaborative filtering recommending and combined recommendation etc.Above proposed algorithm all relates to a gordian technique, that is: calculate similarity between article, find the nearest-neighbors of article according to similarity.The content characteristic of method all only with film self of traditional calculating film similarity is relevant, does not consider the impact of interest preference on film of different user.Can not embody personalized feature according to existing computing method, it is satisfied that such Consumer's Experience is difficult to obtain user.

In real personalized recommendation system, user interest model is basis, personalized recommendation algorithm is core.Personalized film recommends the information source that take into full account three aspects: 1) user information database, the i.e. essential information of user, comprise age, sex, occupation etc.; 2) commodity information database, the i.e. attribute information of film, comprise performer, director, type, brief introduction, area, issuing time etc.; 3) user history information storehouse, the historical record namely in user's use procedure, comprises the preference degree different to each side such as performer, director, type, area, issuing time, brief introductions.Above information source is very crucial, only makes full use of this information bank and sets up user interest model, just can accomplish better recommendation.

Summary of the invention

For the above-mentioned technical matters existed in prior art, the present invention proposes the personalized similarity calculating method of a kind of film based on old user's interest model, by fully excavating and utilize the historical data of user, set up each user user interest model separately, then calculate film similarity based on user interest model, form personalized film similarity list.

To achieve these goals, the present invention adopts following technical scheme:

The personalized similarity calculating method of film based on old user's interest model, comprises step:

S1, evaluate the highest N portion film in viewing record based on the user behavior data in certain section of time T with in this period, set up user's dynamic behaviour information bank;

S2, data mining is carried out to above-mentioned user's dynamic behaviour information bank, obtain user to the preference of each dimension of film and user to the preference of eigenwert in dimension each in film, build user interest model; Wherein,

(1) user adopts space vector can be expressed as one hexa-atomic group to the preference of each dimension of film, that is:

V＝{V ₁:w ₁,V ₂:w ₂,V ₃:w ₃,V ₄:w ₄,V ₅:w ₅，,V ₆:w ₆} (1.1)

V _irepresent the dimension of film, w _irepresent that user is to the weight of each dimension of film, i ∈ [1,6], and

By the statistical study of the search behavior to user, condition query behavior, and the feature extraction to the N portion film of user's height scoring in the T section time, digging user, at performer, director, type, area, preference that time and brief introduction six dimensions are different, tries to achieve each dimension weight w _i;

(2) user can be expressed as the preference of eigenwert in dimension each in film: V _i={ T _ij: W _ij(1.2)

In formula, T _ijbe the jth eigenwert in the i-th dimension, W _ijbe the weight of eigenwert j in the i-th dimension, and

By the search behavior of user, the interested performer of digging user and director information, checked type, screening under time, from different places condition by user, digging user is to interested characteristic value information in these three dimensions; By analyzing N bar viewing record and the collection record of the scoring of user's height, extract each dimensional characteristics value information of film;

Comprehensive above analysis, calculates the weight of each eigenwert of each dimension, that is:

V _ij＝w _i*W _ij(1.3)

In formula, V _ijrepresent the interest-degree of a jth eigenwert in the i-th dimension, and i ∈ [1,6], j ∈ [1, n);

S3, by the analysis of user interest model to film six dimensions, calculate the similarity of each dimension between film A and film B according to formula (1.4), generate personalized film similarity table;

$\mathrm{sim}({M_i}^A,{M_i}^B)=\frac{\underset{j\∈(M_A\∩M_B)}{\mathrm{\Σ}}(V_{\mathrm{ij}}^A+V_{\mathrm{ij}}^B)}{2}---\left(1.4\right)$

In formula, i ∈ [Isosorbide-5-Nitrae], represent the performer of film, director, type, regional four dimensions respectively, j represents eigenwert in dimension, V _ijrepresent the weight of film jth eigenwert in i-th dimension;

Film A and the calculating formula of similarity of film B on time dimension are:

$\mathrm{sim}({M_5}^A,{M_5}^B)=\frac{\min \left(\right|\mathrm{Date}-D_A|,|\mathrm{Date}-D_B\left|\right)}{\max \left(\right|\mathrm{Date}-D_A|,|\mathrm{Date}-D_B\left|\right)}---\left(1.5\right)$

In formula, Date represents the current time, D _arepresent the issuing time of film A, D _brepresent the issuing time of film B, minimum value is got in min () expression, and maximal value is got in max () expression;

Film M _awith film M _bcalculating formula of similarity in brief introduction dimension is sim (M ₆ ^a, M ₆ ^b), utilize Simhash algorithm to calculate;

S4, according to each dimension weight obtained in s2 and s3 and each dimension similarity thereof, formula (1.6) is utilized to calculate film similarity:

$\mathrm{Sim}=\underset{i=1}{\overset{i=6}{\mathrm{\Σ}}}{w}_i\·\mathrm{sim}({M_i}^A,{m_i}^B)---\left(1.6\right)$

In formula, w _irepresent the weight of corresponding each dimension, sim (M _i ^a, M _i ^b) represent the similarity of corresponding each dimension, i ∈ [1,6].

In addition, the invention allows for the personalized similarity calculating method of a kind of film for new user, according to the log-on message of user, comprise age, sex codominance information, and with the interest preference of other masses of the age bracket same sex, based on film own content feature, the method for each dimension weighted sum is adopted to calculate film similarity.

To achieve these goals, the present invention adopts following technical scheme:

The personalized similarity calculating method of film for new user, comprises the steps:

S1, the actor information extracting every portion film, director information, type information and regional information, temporal information and brief introduction information, form six-vector space;

S2, based on user's dominant information, this user to be classified, to find bunch group the most similar to this user, use statistical method to analyze the average preference of this crowd in performer, director, type, area, issuing time and brief introduction six dimensions;

S3, based on film own content feature, the method for each dimension weighted sum is adopted to calculate film similarity, that is:

Sim＝w ₁x ₁+w ₂x ₂+w ₃x ₃+w ₄x ₄+w ₅x ₅+w ₆x ₆(2.1)

In formula, w _irepresent the weight of corresponding each dimension, i ∈ [1,6], this value is calculated by the method in step s2; x _irepresent the similarity of performer, director, type, area, issuing time five dimensions respectively, i ∈ [1,5], utilize cosine similarity formula (2.2) to calculate:

$x_i=\frac{{\mathrm{\Σ}}_{i=1}^n(A_i\×B_i)}{\sqrt{{\mathrm{\Σ}}_{i=1}^n{\left(A_i\right)}^2\×\sqrt{{\mathrm{\Σ}}_{i=1}^n{\left(B_i\right)}^2}}}=\frac{A^T\·B}{\left|\right|A\left|\right|\×\left|\right|B\left|\right|}---\left(2.2\right)$

Similarity value x in brief introduction dimension ₆, then Simhash algorithm is utilized to calculate.

Tool of the present invention has the following advantages:

The present invention is based on the performer of film, director, type, area, time, these six base attributes of brief introduction, set up the user interest model of corresponding two-layer 6 DOF; When calculating film similarity, customer-centric, goes out from user interest angle, sets up user interest model, in conjunction with film own content feature, and then is formed and user's personalized film similarity recommendation list one to one; For new user, first adopt sorting technique, find this user most suitable bunch of group, the average interest preference of this customer group of Statistics Application methods analyst, as the initial preference of this user, obtain corresponding film similarity initial list; For old user, the click of abundant digging user, search, condition query, viewing, collection behavior, set up two-layer sextuple user interest model, by the different preference degree of user to each dimension and each eigenwert thereof, calculate the list of personalized film similarity; According to the dynamic interests change of user, timing more New cinema similarity list.The customer-centric that the inventive method is real, can improve the effect of recommendation, let user experiencing better personalized ventilation system.

Accompanying drawing explanation

Fig. 1 is the FB(flow block) based on the personalized similarity calculating method of the film of old user's interest model in the embodiment of the present invention 1;

Fig. 2 is the user interest model structural representation set up in the embodiment of the present invention 1;

Fig. 3 is the FB(flow block) based on the personalized similarity calculating method of the film of new user in the embodiment of the present invention 2.

Embodiment

Below in conjunction with accompanying drawing and embodiment, the present invention is described in further detail:

Embodiment 1

Shown in composition graphs 1, the personalized similarity calculating method of a kind of film based on old user's interest model, comprises step:

The collection of s1, user interest

Choose the user behavior data in certain section of time T and evaluate the highest N portion film in viewing record in this period, setting up user's dynamic behaviour information bank;

Herein, user behavior data is mainly the dynamic interest of user, comprises click, search, viewing and collection behavior etc.

The formalization representation of s2, user interest model

By the analysis to this special medium of film, the present invention adopts a two-layer sextuple space vector to represent user interest model, and two-layer sextuple space vector as shown in Figure 2.

This user's film interest model comprises two-layer, that is: user to the preference of each dimension of film and user to the preference of eigenwert in dimension each in film;

(1) user adopts space vector can be expressed as one hexa-atomic group to the preference of each dimension of film, that is:

V＝{V ₁:w ₁,V ₂:w ₂,V ₃:w ₃,V ₄:w ₄,V ₅:w ₅，,V ₆:w ₆} (1.1)

V _irepresent the dimension of film, w _irepresent that user is to the weight of each dimension of film, i ∈ [1,6], and

By the statistical study of the search behavior to user, condition query behavior, and the feature extraction to the N portion film of user's height scoring in the T section time, digging user, at performer, director, type, area, preference that time and brief introduction six dimensions are different, tries to achieve each dimension weight w _i;

(2) user is to the preference of eigenwert in dimension each in film

In this layer, major embodiment user is interested in which eigenwert in this dimension.Such as: user misses potter Cheng Long in performer's dimension, so the weight of this eigenwert is higher.

V _i＝{T _ij:W _ij} (1.2)

In formula, T _ijbe the jth eigenwert in the i-th dimension, W _ijbe the weight of eigenwert j in the i-th dimension, and

By the search behavior of user, the interested performer of digging user and director information, checked type, screening under time, from different places condition by user, digging user is to interested characteristic value information in these three dimensions; By analyzing N bar viewing record and the collection record of the scoring of user's height, extract each dimensional characteristics value information of film;

Comprehensive above analysis, calculates the weight of each eigenwert of each dimension.

User is to the interest of certain eigenwert, first to know which dimension is this eigenwert belong to, and know the interest weighted value of user to this dimension, then under this dimension, user to the computing formula of the interest weighted value of this eigenwert as shown in (1.1):

V _ij＝w _i*W _ij(1.3)

In formula, V _ijrepresent the interest-degree of a jth eigenwert in the i-th dimension, and i ∈ [1,6], j ∈ [1, n);

S3, by the analysis of user interest model to film six dimensions, calculate film M according to formula (1.4) _awith film M _bbetween the similarity of each dimension, generate personalized film similarity table;

$\mathrm{sim}({M_i}^A,{M_i}^B)=\frac{\underset{j\∈(M_A\∩M_B)}{\mathrm{\Σ}}(V_{\mathrm{ij}}^A+V_{\mathrm{ij}}^B)}{2}---\left(1.4\right)$

In formula, i represents the performer of film, director, type, regional four dimensions, and j represents eigenwert in dimension, V _ijrepresent the weight of film jth eigenwert in i-th dimension;

Film M _awith film M _bcalculating formula of similarity on time dimension is:

$\mathrm{sim}({M_5}^A,{M_5}^B)=\frac{\min \left(\right|\mathrm{Date}-D_A|,|\mathrm{Date}-D_B\left|\right)}{\max \left(\right|\mathrm{Date}-D_A|,|\mathrm{Date}-D_B\left|\right)}---\left(1.5\right)$

In formula, Date represents the current time, D _arepresent the issuing time of film A, D _brepresent the issuing time of film B, minimum value is got in min () expression, and maximal value is got in max () expression;

Film M _awith film M _bcalculating formula of similarity in brief introduction dimension is sim (M ₆ ^a, M ₆ ^b), utilize Simhash algorithm to calculate;

S4, according to each dimension weight obtained in s2 and s3 and each dimension similarity thereof, formula (1.6) is utilized to calculate film similarity:

$\mathrm{Sim}=\underset{i=1}{\overset{i=6}{\mathrm{\Σ}}}{w}_i\·\mathrm{sim}({M_i}^A,{m_i}^B)---\left(1.6\right)$

In formula, w _irepresent the weight of corresponding each dimension, sim (M _i ^a, M _i ^b) represent the similarity of corresponding each dimension, i ∈ [1,6]

The whether change of s5, quantitative check every day user dynamic behaviour information bank, if change, off-line is New cinema similarity table more.

The main thought of Simhash algorithm is dimensionality reduction, the maps feature vectors of higher-dimension is become the fingerprint (fingerprint) of a f-bit, determine whether article repeats or highly approximate by the Hamming Distance of the f-bit fingerprint comparing two sections of documents.Simhash algorithm is very exquisite, but very easy understand and realization, concrete Simhash process is as follows:

1) first based on traditional IR method, the vector that eigenwert article being converted to one group of weighting is formed.

2) the vectorial V of an initialization f dimension, wherein each element initial value is 0.

3) for each feature that the proper vector of article is concentrated, following calculating is done:

Traditional hash algorithm is utilized to be mapped to the signature of a f-bit.For the signature of this f-bit, if i-th of signature is 1, then the weights of this feature are added to the i-th dimension in vectorial V, otherwise the weights of this feature are deducted to the i-th dimension of vector.

4) to after the above-mentioned computing of whole proper vector set iteration, determine the value of the f-bit fingerprint generated according to the symbol of one-dimensional vector every in V, if i-th dimension of V is positive number, then the i-th dimension generating f-bit fingerprint is 1, otherwise is 0.

To sum up, this embodiment 1 has following features:

1, according to the historical behavior of user, namely by personalized film commending system platform, digging user is to the various search behavior in movie resource storehouse and viewing and collection behavior; Abundant excavation and analyze the different preference degree of different user in the performer of film, director, type, area, time, these six base attributes of brief introduction, namely obtains the ground floor sextuple space vector representation of user model; 2, according to the above-mentioned behavior of user, by keyword extraction or semantic analysis, analyze the weight of different user in above-mentioned six dimensions shared by each eigenwert, namely obtain the second layer sextuple space vector representation of user model; 3, with the interest model of a two-layer hyperspace vector representation user, based on the interest model of user and the substance feature of film, for different user, different film similarity lists is generated.

Embodiment 2

The present embodiment 2, for new user, because user is also without historical behavior, now according to the log-on message of user, comprises age, sex codominance information, and with the interest preference of other masses of the age bracket same sex.Based on film own content feature, adopt the method for each dimension weighted sum to calculate film similarity, idiographic flow signal as shown in Figure 3.

The personalized similarity calculating method of film for new user, comprises the steps:

S1, the actor information extracting every portion film, director information, type information and regional information, temporal information and brief introduction information, form six-vector space, the Similarity value x of each dimension of calculated off-line _i;

S2, based on user's dominant information, this user to be classified, to find bunch group the most similar to this user, use statistical method to analyze the average preference of this crowd in performer, director, type, area, issuing time and brief introduction six dimensions;

S3, based on film own content feature, the method for each dimension weighted sum is adopted to calculate film similarity, that is:

Sim＝w ₁x ₁+w ₂x ₂+w ₃x ₃+w ₄x ₄+w ₅x ₅+w ₆x ₆(2.1)

In formula, w _irepresent the weight of corresponding each dimension, i ∈ [1,6], this value is calculated by the method in step s2;

In formula, x _irepresent the similarity of performer, director, type, area, issuing time five dimensions respectively, i ∈ [1,5], utilize cosine similarity formula (2.2) to calculate:

$x_i=\frac{{\mathrm{\Σ}}_{i=1}^n(A_i\×B_i)}{\sqrt{{\mathrm{\Σ}}_{i=1}^n{\left(A_i\right)}^2\×\sqrt{{\mathrm{\Σ}}_{i=1}^n{\left(B_i\right)}^2}}}=\frac{A^T\·B}{\left|\right|A\left|\right|\×\left|\right|B\left|\right|}---\left(2.2\right)$

Similarity value x in brief introduction dimension ₆, then Simhash algorithm is utilized to calculate.The Simhash algorithm that the present embodiment 2 is addressed can refer to the Simhash algorithm in embodiment 1.

Certainly; more than illustrate and be only preferred embodiment of the present invention; the present invention is not limited to enumerate above-described embodiment; should be noted that; any those of ordinary skill in the art are under the instruction of this instructions; made all equivalently to substitute, obvious form of distortion, within the essential scope all dropping on this instructions, protection of the present invention ought to be subject to.

Claims (2)

1., based on the personalized similarity calculating method of film of old user's interest model, it is characterized in that, comprise step:

S1, evaluate the highest N portion film in viewing record based on the user behavior data in certain section of time T with in this period, set up user's dynamic behaviour information bank;

S2, data mining is carried out to above-mentioned user's dynamic behaviour information bank, obtain user to the preference of each dimension of film and user to the preference of eigenwert in dimension each in film, build user interest model; Wherein,

(1) user adopts space vector can be expressed as one hexa-atomic group to the preference of each dimension of film, that is:

V＝{V ₁:w ₁,V ₂:w ₂,V ₃:w ₃,V ₄:w ₄,V ₅:w ₅，,V ₆:w ₆} (1.1)

V _irepresent the dimension of film, w _irepresent that user is to the weight of each dimension of film, i ∈ [1,6], and

By the statistical study of the search behavior to user, condition query behavior, and the feature extraction to the N portion film of user's height scoring in the T section time, digging user, at performer, director, type, area, preference that time and brief introduction six dimensions are different, tries to achieve each dimension weight w _i;

(2) user can be expressed as the preference of eigenwert in dimension each in film: V _i={ T _ij: W _ij(1.2)

In formula, T _ijbe the jth eigenwert in the i-th dimension, W _ijbe the weight of eigenwert j in the i-th dimension, and

By the search behavior of user, the interested performer of digging user and director information, checked type, screening under time, from different places condition by user, digging user is to interested characteristic value information in these three dimensions; By analyzing N bar viewing record and the collection record of the scoring of user's height, extract each dimensional characteristics value information of film;

Comprehensive above analysis, calculates the weight of each eigenwert of each dimension, that is:

V _ij＝w _i*W _ij(1.3)

In formula, V _ijrepresent the interest-degree of a jth eigenwert in the i-th dimension, and i ∈ [1,6], j ∈ [1, n);

S3, by the analysis of user interest model to film six dimensions, calculate the similarity of each dimension between film A and film B according to formula (1.4), generate personalized film similarity table;

$\mathrm{sim}({M_i}^A,{M_i}^B)=\frac{\underset{j\∈(M_A\∩M_B)}{\mathrm{\Σ}}(V_{\mathrm{ij}}^A+V_{\mathrm{ij}}^B)}{2}---\left(1.4\right)$

In formula, i ∈ [Isosorbide-5-Nitrae], represent the performer of film, director, type, regional four dimensions respectively, j represents eigenwert in dimension, V _ijrepresent the weight of film jth eigenwert in i-th dimension;

Film A and the calculating formula of similarity of film B on time dimension are:

$\mathrm{sim}({M_5}^A,{M_5}^B)=\frac{\min \left(\right|\mathrm{Date}-D_A|,|\mathrm{Date}-D_B\left|\right)}{\max \left(\right|\mathrm{Date}-D_A|,|\mathrm{Date}-D_B\left|\right)}---\left(1.5\right)$

In formula, Date represents the current time, D _arepresent the issuing time of film A, D _brepresent the issuing time of film B, minimum value is got in min () expression, and maximal value is got in max () expression;

Film A and the calculating formula of similarity of film B in brief introduction dimension are sim (M ₆ ^a, M ₆ ^b), utilize Simhash algorithm to calculate;

S4, according to each dimension weight obtained in s2 and s3 and each dimension similarity thereof, formula (1.6) is utilized to calculate film similarity:

$\mathrm{Sim}=\underset{i=1}{\overset{i=6}{\mathrm{\Σ}}}{w}_i\·\mathrm{sim}({M_i}^A,{M_i}^B)---\left(1.6\right)$

In formula, w _irepresent the weight of corresponding each dimension, sim (M _i ^a, M _i ^b) represent the similarity of corresponding each dimension, i ∈ [1,6].

2., for the new user's personalized similarity calculating method of film, it is characterized in that, comprise the steps:

S1, the actor information extracting every portion film, director information, type information and regional information, temporal information and brief introduction information, form six-vector space;

S2, based on user's dominant information, this user to be classified, to find bunch group the most similar to this user, use statistical method to analyze the average preference of this crowd in performer, director, type, area, issuing time and brief introduction six dimensions;

S3, based on film own content feature, the method for each dimension weighted sum is adopted to calculate film similarity, that is:

Sim＝w ₁x ₁+w ₂x ₂+w ₃x ₃+w ₄x ₄+w ₅x ₅+w ₆x ₆(2.1)

In formula, w _irepresent the weight of corresponding each dimension, i ∈ [1,6], this value is calculated by the method in step s2; x _irepresent the similarity of performer, director, type, area, issuing time five dimensions respectively, i ∈ [1,5], utilize cosine similarity formula (2.2) to calculate:

$x_i=\frac{{\mathrm{\Σ}}_{i=1}^n(A_i\×B_i)}{\sqrt{{\mathrm{\Σ}}_{i=1}^n{\left(A_i\right)}^2}\×\sqrt{{\mathrm{\Σ}}_{i=-1}^n{\left(B_i\right)}^2}}=\frac{A^T\·B}{\left|\right|A\left|\right|\×\left|\right|B\left|\right|}---\left(2.2\right)$

Similarity value x in brief introduction dimension ₆, then Simhash algorithm is utilized to calculate.