CN105808680A - Tensor decomposition based context-dependent position recommendation method - Google Patents

Abstract

The invention discloses a tensor decomposition based context-dependent place recommendation method and mainly solves the problem of poor place recommendation quality in the prior art. The method is implemented by the steps of 1, constructing a three-dimensional sign-in tensor A and a user similarity matrix B by utilizing sign-in data of all users of a to-be-recommended city; 2, obtaining a three-dimensional tensor A by using a high-order singular value decomposition algorithm; 3, obtaining the current position of a to-be-recommended user c; and 4, according to the three-dimensional tensor A, performing place recommendation on the to-be-recommended user c. According to the method, the communication traffic between the user and a recommendation system is reduced by utilizing tensor decomposition, and the effectiveness and reliability of a place recommendation result in a data sparsity scene is ensured in combination with a time context and historical user data; and the method can be applied to position based place recommendation services in a social network.

Description

Carry out context-sensitive position based on resolution of tensor and recommend method

Technical field

The invention belongs to technical field of data processing, particularly to a kind of and time context-sensitive personalized recommendation technology, can be used for the place Push Service in position-based social networks.

Background technology:

Along with the development of information technology and mobile terminal technology, the exponentially rule of the quantity of information in the Internet increases rapidly.In the face of the data increased sharply, it is desirable to excavate the many important information being hidden in behind, service for people such that it is able to better profit from these data.Service provider has the personal information relevant with user and the historical record of magnanimity, these data are utilized actively to recommend related service user can be helped to make a choice on the one hand to user, find service under their interested, valuable article and line, improve Consumer's Experience；Article and service is allowed to be presented in face of user interested in them by social media platform on the other hand, thus realizing user and the doulbe-sides' victory of article provider.

Traditional commending system algorithm mainly uses the proposed algorithm based on collaborative filtering, content-based recommendation algorithm.

The appearance of wireless network and location aware technology and progress, promote mobile Internet, being based especially on social networks flourish of position, the geographic position data of user starts to be obtained in a large number by client GPS so that some positions recommend task to be possibly realized.After collecting the geographic position data of user, it is necessary to first user is divided on geographical position, division methods is based on user and registers the division methods based on the on-site longitude in position and latitude.Then, all historical records of position history record and user to be recommended by scanning all users in this area, calculate user and the area prediction mark to position by preset model, and then obtain recommended location.Currently existing based in the recommendation numerical procedure of collaborative filtering, a kind of similarity degree being based between user itself, a kind of similarity degree being based between project itself.Collaborative filtering based on user is applicable to the less occasion of user, there is the cold start-up problem of new user, it is difficult to provide the recommendation making user convince to explain.Project-based collaborative filtering can utilize the historical behavior of user to make compellent recommendation of comparison and explain there is the cold start-up problem of new projects.

In said method, owing to all not accounting for time context in the impact carrying out place recommendation for user, thus in Sparse scene, performance is very poor, and for the cold start-up problem of new user and new projects, also cannot effectively solve.

Summary of the invention

Present invention aims to the deficiency of above method, it is proposed to one carries out context-sensitive position based on resolution of tensor and recommends method, to improve place recommendation results performance in sparse scene, it is to avoid the cold start-up problem in new user and Xin place.

For achieving the above object, technical scheme includes:

(1) data of registering of all users in city to be recommended are obtained, according to the user i of data of registering, place j, time t and user i at the time t scoring A to place j_ijk, construct the tensor A of registering of three-dimensional；

(2) according to registering the user record in data, weighting Pearson came Similarity Measure is utilized to go out user similarity matrix B, B ∈ R^m×m, wherein, m represents the quantity of user；

(3) combine three-dimensional register tensor A and user similarity matrix B, use Higher-order Singular value decomposition algorithm to calculate all users scoring to all places, obtain tensor

(4) according to tensorUser c to be recommended is carried out place recommendation:

(4a) current location of user c to be recommended is obtained；

(4b) the user c to be recommended scoring set A to all places, city to be recommended is obtained_c；

(4c) size of the place quantity comprised according to city to be recommended obtains recommendation results R:

If the place quantity that city to be recommended comprises is less than or equal to 500, then according to the user c to be recommended scoring set A to all places in city to be recommended_c, select the highest front K the place of scoring to determine by service provider as the quantity of recommendation results R, K；

If the place quantity that city to be recommended comprises is more than 500, then the algorithm based on threshold value is used to obtain the recommendation results R of user c to be recommended.

Present invention have the advantage that

Due to the fact that and consider time context, improve the performance of recommendation results under Sparse scene；

Due to the fact that the on-line off-line mode combined of use, reduce the amount of calculation on line and the traffic, improve response speed；

Accompanying drawing explanation

Fig. 1 is the flowchart of the present invention；

Fig. 2 is the recommendation system framework built by the present invention；

Specific embodiments

The core concept of the present invention is to propose one to carry out context-sensitive position recommendation method based on resolution of tensor, it is provided that high performance place recommendation results.

With reference to Fig. 1, it is as follows that the present invention realizes step:

Step 1, constructs the tensor A of registering of three-dimensional.

According to user i, place j, time t and user i in all data of registering in city to be recommended at the time t scoring A to place j_ijkRecord, constructs the tensor A of registering of user-place-time three-dimensional, wherein A_ijkFor tensor A the i-th row, jth row, element that t degree is corresponding.

Step 2, according to the user record registered in data, utilizes weighting Pearson came Similarity Measure to go out user similarity matrix B.

(2a) weight coefficient w it is calculated as follows_uv:

$w_{uv}=\frac{{\mathrm{\Σ}}_{j\∈N\left(u\right)\∩N\left(v\right)}\frac{1}{log1+\left|N\left(j\right)\right|}}{\sqrt{\left|N\left(u\right)\right|\left|N\left(v\right)\right|}},$

Wherein, N (u) represents the set of the user u position accessed, | N (u) | represents the quantity of the user u position accessed, and N (v) represents the set of the user v position accessed, and | N (v) | represents the quantity of the user v position accessed；

(2b) the similarity Sim of Pearson came Similarity Measure user u and user v is used_u,v:

${\mathrm{Sim}}_{u,v}=\frac{\underset{j}{\Σ}(r_{uj}-r_u)(r_{vj}-r_v)}{\sqrt{\underset{j}{\Σ}{(r_{uj}-r_u)}^2\underset{j}{\Σ}{(r_{vj}-r_v)}^2}},$

Wherein, r_uRepresent the meansigma methods of user u scoring, r_ujRepresent the user u score value to position j, r_vRepresent the meansigma methods of user v scoring, r_vjRepresent the user v score value to position j；

(2c) according to weight coefficient w_uvAnd the Pearson came similarity Sim between user_u,v, calculate the weighting Pearson came coefficient of similarity S between user_uv,

S_uv=w_uv·Sim_u,v

(2d) according to the weighting Pearson came coefficient of similarity S between user_uv, build similarity matrix B:

Wherein, m represents the quantity of user, B ∈ R^m×m。

Step 3, builds three-dimensional tensor

Use tensor A and the user similarity matrix B input as Higher-order Singular value decomposition algorithm of registering of three-dimensional, calculate all users the free scoring to all placesConstruct three-dimensional tensorWhereinFor tensorI-th row, jth row, the element that t degree is corresponding.

Step 4, according to three-dimensional tensorUser c to be recommended is carried out place recommendation.

(4a) with reference to Fig. 2, holding the user of terminal unit, use terminal unit and auxiliary positioning equipment to position, service provider obtains the current location of user c to be recommended by 2G/3G/4G or Wifi；

(4b) the user c to be recommended scoring set A to all places, city to be recommended is obtained_c；

(4c) size of the place quantity comprised according to city to be recommended obtains recommendation results R:

If the place quantity that (4c1) city to be recommended comprises is less than or equal to 500, then according to the user c to be recommended scoring set A to all places in city to be recommended_c, select the highest front K the place of scoring to determine by service provider as the quantity of recommendation results R, K；

If the place quantity that (4c2) city to be recommended comprises is more than 500, then the algorithm based on threshold value is used to obtain the recommendation results R of user c to be recommended:

(4c21) city to be recommended is divided into p sub regions, each subregion is built respectively the three-dimensional tensor of user, place and timeWherein, x=1,2 ..., p, p is the number of subregion；

(4c22) for the three-dimensional tensor of every sub regionsObtain user c to be recommended to the scoring set in place in subregionRightSort from high to low according to scoring, obtain p sorted lists sl_x；

(4c23) according to p the sorted lists sl obtained_x, use Priority Queues, select to mark the highest front K place as recommendation results R.

Above description is only to example of the present invention, does not constitute any limitation of the invention.Obviously for those skilled in the art; after having understood present invention and principle; all it is likely to when without departing substantially from the principle of the invention, structure; carry out the various corrections in form and details and change, but these based on the correction of inventive concept and change still within the claims of the present invention.

Claims (3)

1. carry out context-sensitive position based on resolution of tensor and recommend a method, including:

(1) data of registering of all users in city to be recommended are obtained, according to the user i of data of registering, place j, time t and user i at the time t scoring A to place j_ijk, construct the tensor A of registering of three-dimensional；

(2) according to registering the user record in data, weighting Pearson came Similarity Measure is utilized to go out user similarity matrix B, B ∈ R^m×m, wherein, m represents the quantity of user；

(3) combine three-dimensional register tensor A and user similarity matrix B, use Higher-order Singular value decomposition algorithm to calculate all users scoring to all places, obtain tensor

(4) according to tensorUser c to be recommended is carried out place recommendation:

(4a) current location of user c to be recommended is obtained；

(4b) the user c to be recommended scoring set A to all places, city to be recommended is obtained_c；

(4c) size of the place quantity comprised according to city to be recommended obtains recommendation results R:

If the place quantity that city to be recommended comprises is less than or equal to 500, then according to the user c to be recommended scoring set A to all places in city to be recommended_c, select the highest front K the place of scoring to determine by service provider as the quantity of recommendation results R, K；

If the place quantity that city to be recommended comprises is more than 500, then the algorithm based on threshold value is used to obtain the recommendation results R of user c to be recommended.

2. method according to claim 1, wherein utilizes the similarity matrix B of weighting Pearson came Similarity Measure user, carries out as follows in step (2):

(2a) weight coefficient w it is calculated as follows_uv:

$w_{uv}=\frac{{\mathrm{\Σ}}_{j\∈N\left(u\right)\∩N\left(v\right)}\frac{1}{log1+\left|N\left(j\right)\right|}}{\sqrt{\left|N\left(u\right)\right|\left|N\left(v\right)\right|}},$

Wherein, N (u) represents the set of the user u position accessed, | N (u) | represents the quantity of the user u position accessed, and N (v) represents the set of the user v position accessed, and | N (v) | represents the quantity of the user v position accessed；

(2b) the similarity Sim of Pearson came Similarity Measure user u and user v is used_u,v:

${\mathrm{Sim}}_{u,v}=\frac{\underset{j}{\Σ}(r_{uj}-r_u)(r_{vj}-r_v)}{\sqrt{\underset{j}{\Σ}{(r_{uj}-r_u)}^2\underset{j}{\Σ}{(r_{vj}-r_v)}^2}},$

Wherein, r_uRepresent the meansigma methods of user u scoring, r_ujRepresent the user u score value to position j, r_vRepresent the meansigma methods of user v scoring, r_vjRepresent the user v score value to position j；

(2c) according to weight coefficient w_uvAnd the Pearson came similarity Sim between user_u,v, calculate the weighting Pearson came coefficient of similarity S between user_uv, i.e. the element of similarity matrix B u row, v row correspondence:

S_uv=w_uv·Sim_u,v。

3. method according to claim 1, wherein obtains the recommendation results R of user c to be recommended, carries out as follows based on the algorithm of threshold value in step (4c):

(4c1) city to be recommended is divided into p sub regions, each subregion is built respectively the three-dimensional tensor of user, place and timeWherein, x=1,2 ..., p, p is the number of subregion；

(4c2) for the three-dimensional tensor of every sub regionsObtain user c to be recommended to the scoring set in place in subregionRightSorted lists sl is obtained from high to low according to scoring_x；

(4c3) according to p the sorted lists sl obtained_x, use Priority Queues, select to mark the highest front K place as recommendation results R.