CN107122407A - Multi-Domain Recommendation Method Based on Feature Selection - Google Patents

Abstract

The invention discloses a multi-field recommendation method based on feature selection, which includes: acquiring rating matrices of multiple users on multiple items and domain information of each item; calculating similarity matrix between users and similarity between domains degree matrix; optimize the complete model of the preset multi-domain recommendation system to obtain the partial derivatives of the objective function for the similarity matrix between users, the similarity matrix between domains and the feature selection vector of domains until the convergence condition to obtain the final user preference matrix, final item feature matrix, and final feature selection vector; rating prediction is performed based on the end user preference matrix, final item feature matrix, and final feature selection vector, so as to recommend information based on the rating results. The present invention has the following advantages: Alleviate the impact of the sparse scoring matrix on the performance of the recommendation system, effectively eliminate the original multi-field recommendation composed of two independent steps and ignore the combined effect of field segmentation and recommendation, and effectively improve the accuracy of recommendation sex.

Description

Multi-Domain Recommendation Method Based on Feature Selection

technical field

The invention relates to the field of recommendation methods, in particular to a multi-field recommendation method based on feature selection.

Background technique

Nowadays, with the rapid increase of information on the Internet, the problem of information overload is becoming more and more serious. Both consumers and information producers have encountered great challenges: as information consumers, how to find the content they are interested in from a large amount of information has become a very difficult task, especially when users do not have clear needs. As an information producer, how to make the information you produce stand out and attract the attention of the majority of users is also a very difficult thing. The recommended method is an important tool to solve this contradiction.

Among the various recommendation algorithms, collaborative filtering algorithm is one of the most successful algorithms, whose basic assumption is that users with similar rating behaviors also have similar preferences in the selection of other items. Although collaborative filtering-based recommendation methods have achieved great success in the real world, they still have certain shortcomings and limitations. The biggest challenge is the problem of data sparsity, that is, in the huge rating matrix, the rating data of users is extremely sparse. In recent years, to alleviate such problems, multi-domain based recommendation methods have been proposed. Here, a domain (aka subgroup) refers to a group of users who have similar preferences for a set of items. By dividing users and items into overlapping subgroups, and then independently generating recommendations on each subgroup, such methods improve the performance of collaborative filtering-based recommendation algorithms to a certain extent. Traditional multi-domain-based recommendation techniques need to classify users and items into different categories, and then combine the results from each area. However, these two steps ignore their interaction in the recommendation process.

Contents of the invention

The present invention aims to solve at least one of the above-mentioned technical problems.

Therefore, the purpose of the present invention is to propose a multi-domain recommendation method based on feature selection to solve the problem of data sparsity.

In order to achieve the above purpose, the embodiment of the present invention discloses a multi-domain recommendation method based on feature selection, including the following steps: S1: Obtain the rating matrix of multiple users for multiple items and the domain information of each item; S2: Calculate the similarity matrix between users according to the scoring matrix, and calculate the similarity matrix between domains according to the domain information of each item; S3: use the stochastic gradient descent algorithm to optimize the complete model of the preset multi-domain recommendation system to obtain the target The partial derivatives of the function for the similarity matrix between users, the similarity matrix between domains and the feature selection vector of domains until the convergence conditions are met to obtain the final user preference matrix, the final item feature matrix and the final feature selection vector; S4: According to Score prediction is performed on the final user preference matrix, the final item feature matrix and the final feature selection vector, so as to recommend information to the plurality of users according to the scoring results.

Further, the similarity matrix between the users is calculated by the following formula:

Among them, s _ij represents the element in row i and column j in the similarity matrix between users, r _i =[r _i1 ,...,r _ic ] is the score distribution, and r _ij is the normalized user The number of ratings for i in the jth domain.

Further, the domain information of each item is calculated by the following formula to calculate the similarity matrix between domains:

in, is the similarity between domains k and l for user u, where Indicates that user u has scored in k domains.

Further, the complete model of the preset multi-domain recommendation system is:

Among them, P and Q are the d-dimensional hidden variable expressions of users and items, p _i and q _i represent the feature vectors of each user i or item j, m _k represents the feature selection vector of domain k, α, β, γ and λ is the parameter to balance each item;

The partial derivatives of the objective function for the similarity matrix between users, the similarity matrix between domains and the feature selection vector of domains are respectively:

Among them, D is a diagonal matrix dependent on m _k , and the value of the i-th element on the diagonal is calculated by the following formula:

where ε is a positive number that smoothes the target;

Initialize all variables P, Q and m to random numbers of [0,1], and update the variables according to the step size parameters ω _p , ω _q and ω _m until the algorithm converges to obtain the final user preference matrix and the final item feature matrix and the final feature selection vector.

Further, in step S1, each of the plurality of users has at least a first threshold number of scoring data, and each of the plurality of items has at least a second threshold number of ratings.

According to the feature selection-based multi-domain recommendation method of the embodiment of the present invention, the feature selection vector is introduced on the basis of matrix decomposition by using multi-domain information of commodities, so as to alleviate the impact of the sparse rating matrix on the performance of the recommendation system. And it effectively eliminates the fact that the original multi-domain recommendation consists of two independent steps and ignores the joint effect of domain segmentation and recommendation. Using this framework to recommend items for users can effectively improve the accuracy of recommendation.

Additional aspects and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.

Description of drawings

The above and/or additional aspects and advantages of the present invention will become apparent and comprehensible from the description of the embodiments in conjunction with the following drawings, wherein:

Fig. 1 is the flow chart of the multi-domain recommendation method based on feature selection in an embodiment of the present invention;

Fig. 2 is a model schematic diagram of a multi-domain recommendation method based on feature selection according to an embodiment of the present invention.

detailed description

Embodiments of the present invention are described in detail below, examples of which are shown in the drawings, wherein the same or similar reference numerals designate the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the figures are exemplary only for explaining the present invention and should not be construed as limiting the present invention.

These and other aspects of embodiments of the invention will become apparent with reference to the following description and drawings. In these descriptions and drawings, some specific implementation manners in the embodiments of the present invention are specifically disclosed to represent some ways of implementing the principles of the embodiments of the present invention, but it should be understood that the scope of the embodiments of the present invention is not limited by This restriction. On the contrary, the embodiments of the present invention include all changes, modifications and equivalents coming within the spirit and scope of the appended claims.

The present invention is described below in conjunction with accompanying drawing.

FIG. 1 is a flowchart of a multi-domain recommendation method based on feature selection according to an embodiment of the present invention. The multi-domain recommendation method based on feature selection according to an embodiment of the present invention includes the following steps:

S1: Obtain the rating matrix of multiple users on multiple items and the domain information of each item.

Specifically, in step S1, each of the plurality of users has at least a first threshold number of scoring data, and each of the plurality of items has at least a second threshold number of ratings, so as to reduce data sparsity . In an example of the present invention, the first threshold is 10, and the second threshold is 5.

As shown in FIG. 2 , a rating matrix R is formed according to ratings of multiple users on multiple items. P is the user feature matrix, Q is the item feature matrix, and M is the feature selection vector. Among them, the first score 4 in the scoring matrix R represents the result of multiplying the first row of the user feature matrix P with the diagonal matrix generated by the feature selection vector and then multiplying the first column of the item feature matrix Q.

S2: Calculate the similarity matrix between users according to the scoring matrix, and calculate the similarity matrix between domains according to the domain information of each item.

Specifically, the embodiment of the present invention proposes a complete model of the preset multi-domain recommendation system as follows:

Among them, P and Q are the d-dimensional hidden variable expressions of users and items, p _i and q _i represent the feature vectors of each user i or item j, and m _k represents the feature selection vector of domain k. α, β, γ, and λ are parameters to balance the terms, and their values can be determined by five-fold cross-validation. S and T refer to the similarity matrix between users and the similarity matrix between domains, and S _ij and T _ij are the ijth elements in the similarity matrix, respectively.

S _ij and T _ij are calculated as follows:

Wherein, r _i =[r _i1 ,...,r _ic ] is the rating distribution, and r _ij is the number of normalized ratings of user i in the jth domain. Once two users have similar rating distributions, they have a relatively high similarity.

in, is the similarity between domains k and l for user u, where Indicates that user u has scored in k domains. It should be noted that when hour, is worthless.

S3: Use the stochastic gradient descent algorithm to optimize the complete model of the preset multi-domain recommendation system to obtain the partial derivatives of the objective function for the similarity matrix between users, the similarity matrix between domains and the feature selection vector of domains until convergence is satisfied condition to obtain the final user preference matrix, the final item feature matrix and the final feature selection vector;

Specifically, the stochastic gradient descent method is used to optimize the complete model of the preset multi-domain recommendation system. The objective function of the complete model of the preset multi-domain recommendation system is denoted as L, then L is about p _i , q _j and m _k The partial derivatives of are as follows:

Among them, D is a diagonal matrix dependent on m _k , and the value of the i-th element on the diagonal is calculated by the following formula:

where ε is a positive number that smoothes the target;

Initialize all variables P, Q and m to random numbers of [0,1], and update the variables according to the step size parameters ω _p , ω _q and ω _m until the algorithm converges to obtain the final user preference matrix and the final item feature matrix and the final feature selection vector. ω _p , ω _q and ω _m are determined by a linear search algorithm.

S4: Perform score prediction according to the final user preference matrix, the final item feature matrix, and the final feature selection vector, so as to recommend information to the plurality of users according to the score results.

According to the feature selection-based multi-domain recommendation method of the embodiment of the present invention, the feature selection vector is introduced on the basis of matrix decomposition by using multi-domain information of commodities, so as to alleviate the impact of the sparse rating matrix on the performance of the recommendation system. And it effectively eliminates the fact that the original multi-domain recommendation consists of two independent steps and ignores the combined effect of domain segmentation and recommendation. Using this framework to recommend items for users can effectively improve the accuracy of recommendation.

In addition, other components and functions of the multi-domain recommendation method based on feature selection in the embodiment of the present invention are known to those skilled in the art, and will not be repeated in order to reduce redundancy.

In the description of the present invention, it should be understood that the terms "first" and "second" are used for description purposes only, and should not be understood as indicating or implying relative importance.

In the description of this specification, descriptions referring to the terms "one embodiment", "some embodiments", "example", "specific examples", or "some examples" mean that specific features described in connection with the embodiment or example , structure, material or characteristic is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

Although the embodiments of the present invention have been shown and described, those skilled in the art can understand that various changes, modifications, substitutions and modifications can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the invention is defined by the claims and their equivalents.

Claims (5)

1. A multi-field recommendation method based on feature selection is characterized by comprising the following steps:

s1: acquiring a scoring matrix of a plurality of items by a plurality of users and domain information of each item;

s2: calculating a similarity matrix between users according to the grading matrix, and calculating a similarity matrix between domains according to the domain information of each article;

s3: optimizing a complete model of a preset multi-field recommendation system by adopting a random gradient descent algorithm to obtain partial derivatives of a target function respectively corresponding to a similarity matrix among users, a similarity matrix among fields and a feature selection vector of the fields until convergence conditions are met to obtain a final user preference matrix, a final article feature matrix and a final feature selection vector;

s4: and performing grading prediction according to the final user preference matrix, the final article feature matrix and the final feature selection vector so as to recommend information to the users according to a grading result.

2. The multi-domain recommendation method based on feature selection according to claim 1, wherein the similarity matrix between the users is calculated by the following formula:

<mrow> <msub> <mi>S</mi> <mrow> <mi>i</mi> <mi>j</mi> </mrow> </msub> <mo>=</mo> <mfrac> <mrow> <msubsup> <mi>r</mi> <mi>i</mi> <mi>T</mi> </msubsup> <msub> <mi>r</mi> <mi>j</mi> </msub> </mrow> <mrow> <mo>|</mo> <mo>|</mo> <msub> <mi>r</mi> <mi>i</mi> </msub> <mo>|</mo> <msup> <mo>|</mo> <mn>2</mn> </msup> <mo>|</mo> <mo>|</mo> <msub> <mi>r</mi> <mi>j</mi> </msub> <mo>|</mo> <msup> <mo>|</mo> <mn>2</mn> </msup> </mrow> </mfrac> </mrow>

wherein s is_ijElements, r, representing the ith row and jth column in the similarity matrix between said users_i＝[r_i1,...,r_ic]Is the score distribution, r_ijIs the normalized number of user i's scores in the jth domain.

3. The multi-domain recommendation method based on feature selection according to claim 1, wherein the similarity matrix between the domain information calculation domains of each item is calculated by the following formula:

<mrow> <msubsup> <mi>T</mi> <mrow> <mi>k</mi> <mi>l</mi> </mrow> <mi>u</mi> </msubsup> <mo>=</mo> <mi>exp</mi> <mrow> <mo>(</mo> <mo>-</mo> <mo>(</mo> <mrow> <mi>m</mi> <mi>a</mi> <mi>x</mi> <mrow> <mo>(</mo> <mrow> <msubsup> <mi>N</mi> <mi>k</mi> <mi>u</mi> </msubsup> <mo>,</mo> <msubsup> <mi>N</mi> <mi>l</mi> <mi>u</mi> </msubsup> </mrow> <mo>)</mo> </mrow> <mo>/</mo> <mi>m</mi> <mi>i</mi> <mi>n</mi> <mrow> <mo>(</mo> <mrow> <msubsup> <mi>N</mi> <mi>k</mi> <mi>u</mi> </msubsup> <mo>,</mo> <msubsup> <mi>N</mi> <mi>l</mi> <mi>u</mi> </msubsup> </mrow> <mo>)</mo> </mrow> <mo>-</mo> <mn>1</mn> </mrow> <mo>)</mo> <mo>)</mo> </mrow> </mrow>

wherein,is the similarity for user u, fields k and l, whereIndicating that user u has over-scored behavior in k domains.

4. The multi-domain recommendation method based on feature selection according to claim 1, wherein in step S3:

the complete model of the preset multi-domain recommendation system is as follows:

<mfenced open = "" close = ""> <mtable> <mtr> <mtd> <mrow> <munder> <mi>min</mi> <mrow> <mi>P</mi> <mo>,</mo> <mi>Q</mi> <mo>,</mo> <mi>m</mi> </mrow> </munder> <munder> <mi>&amp;Sigma;</mi> <mrow> <mo>{</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>}</mo> <mo>&amp;Element;</mo> <mi>O</mi> </mrow> </munder> <msup> <mrow> <mo>(</mo> <msub> <mi>r</mi> <mrow> <mi>i</mi> <mi>j</mi> </mrow> </msub> <mo>-</mo> <msubsup> <mi>p</mi> <mi>i</mi> <mi>T</mi> </msubsup> <mi>d</mi> <mi>i</mi> <mi>a</mi> <mi>g</mi> <mo>(</mo> <msub> <mi>m</mi> <mi>k</mi> </msub> <mo>)</mo> <msub> <mi>q</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mo>+</mo> <mi>&amp;alpha;</mi> <munder> <mi>&amp;Sigma;</mi> <mi>k</mi> </munder> <mo>|</mo> <mo>|</mo> <msub> <mi>m</mi> <mi>k</mi> </msub> <mo>|</mo> <msub> <mo>|</mo> <mn>1</mn> </msub> <mo>+</mo> <mi>&amp;beta;</mi> <munder> <mi>&amp;Sigma;</mi> <mi>i</mi> </munder> <munder> <mi>&amp;Sigma;</mi> <mi>j</mi> </munder> <msub> <mi>S</mi> <mrow> <mi>i</mi> <mi>j</mi> </mrow> </msub> <mo>|</mo> <mo>|</mo> <msub> <mi>p</mi> <mi>i</mi> </msub> <mo>-</mo> <msub> <mi>p</mi> <mi>j</mi> </msub> <mo>|</mo> <msup> <mo>|</mo> <mn>2</mn> </msup> </mrow> </mtd> </mtr> <mtr> <mtd> <mrow> <mo>+</mo> <mi>&amp;gamma;</mi> <munder> <mi>&amp;Sigma;</mi> <mi>i</mi> </munder> <munder> <mi>&amp;Sigma;</mi> <mi>j</mi> </munder> <msub> <mi>T</mi> <mrow> <mi>i</mi> <mi>j</mi> </mrow> </msub> <mo>|</mo> <mo>|</mo> <msub> <mi>m</mi> <mi>i</mi> </msub> <mo>-</mo> <msub> <mi>m</mi> <mi>j</mi> </msub> <mo>|</mo> <msup> <mo>|</mo> <mn>2</mn> </msup> <mo>+</mo> <mi>&amp;lambda;</mi> <mrow> <mo>(</mo> <mo>|</mo> <mo>|</mo> <mi>P</mi> <mo>|</mo> <msubsup> <mo>|</mo> <mi>F</mi> <mn>2</mn> </msubsup> <mo>+</mo> <mo>|</mo> <mo>|</mo> <mi>Q</mi> <mo>|</mo> <msubsup> <mo>|</mo> <mi>F</mi> <mn>2</mn> </msubsup> <mo>)</mo> </mrow> </mrow> </mtd> </mtr> </mtable> </mfenced>

where P and Q are d-dimensional steganography of users and itemsExpression of variables, p_iAnd q is_iFeature vector, m, representing each user i or item j_kA feature selection vector representing the domain k, α, β, γ, and λ are parameters that balance the terms;

the partial derivatives of the objective function with respect to the similarity matrix between users, the similarity matrix between domains, and the feature selection vector of the domain are respectively:

<mrow> <mfrac> <mrow> <mo>&amp;part;</mo> <mi>L</mi> </mrow> <mrow> <mo>&amp;part;</mo> <msub> <mi>p</mi> <mi>i</mi> </msub> </mrow> </mfrac> <mo>=</mo> <mo>-</mo> <mn>2</mn> <mrow> <mo>(</mo> <msub> <mi>r</mi> <mrow> <mi>i</mi> <mi>j</mi> </mrow> </msub> <mo>-</mo> <msubsup> <mi>p</mi> <mi>i</mi> <mi>T</mi> </msubsup> <mi>d</mi> <mi>i</mi> <mi>a</mi> <mi>g</mi> <mo>(</mo> <msub> <mi>m</mi> <mi>k</mi> </msub> <mo>)</mo> <msub> <mi>q</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> <mi>d</mi> <mi>i</mi> <mi>a</mi> <mi>g</mi> <mrow> <mo>(</mo> <msub> <mi>m</mi> <mi>k</mi> </msub> <mo>)</mo> </mrow> <msub> <mi>q</mi> <mi>j</mi> </msub> <mo>+</mo> <mn>2</mn> <msub> <mi>&amp;lambda;p</mi> <mi>i</mi> </msub> <mo>+</mo> <mn>2</mn> <mi>&amp;beta;</mi> <munder> <mi>&amp;Sigma;</mi> <mi>i</mi> </munder> <msub> <mi>S</mi> <mrow> <mi>i</mi> <mi>j</mi> </mrow> </msub> <mrow> <mo>(</mo> <msub> <mi>p</mi> <mi>i</mi> </msub> <mo>-</mo> <msub> <mi>p</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> </mrow>

<mrow> <mfrac> <mrow> <mo>&amp;part;</mo> <mi>L</mi> </mrow> <mrow> <mo>&amp;part;</mo> <msub> <mi>q</mi> <mi>j</mi> </msub> </mrow> </mfrac> <mo>=</mo> <mo>-</mo> <mn>2</mn> <mrow> <mo>(</mo> <msub> <mi>r</mi> <mrow> <mi>i</mi> <mi>j</mi> </mrow> </msub> <mo>-</mo> <msubsup> <mi>p</mi> <mi>i</mi> <mi>T</mi> </msubsup> <mi>d</mi> <mi>i</mi> <mi>a</mi> <mi>g</mi> <mo>(</mo> <msub> <mi>m</mi> <mi>k</mi> </msub> <mo>)</mo> <msub> <mi>q</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> <mi>d</mi> <mi>i</mi> <mi>a</mi> <mi>g</mi> <mrow> <mo>(</mo> <msub> <mi>m</mi> <mi>k</mi> </msub> <mo>)</mo> </mrow> <msub> <mi>p</mi> <mi>i</mi> </msub> <mo>+</mo> <mn>2</mn> <msub> <mi>&amp;lambda;q</mi> <mi>j</mi> </msub> </mrow>

<mrow> <mfrac> <mrow> <mo>&amp;part;</mo> <mi>L</mi> </mrow> <mrow> <mo>&amp;part;</mo> <msub> <mi>m</mi> <mi>k</mi> </msub> </mrow> </mfrac> <mo>=</mo> <mo>-</mo> <mn>2</mn> <mrow> <mo>(</mo> <msub> <mi>r</mi> <mrow> <mi>i</mi> <mi>j</mi> </mrow> </msub> <mo>-</mo> <msubsup> <mi>p</mi> <mi>i</mi> <mi>T</mi> </msubsup> <mi>d</mi> <mi>i</mi> <mi>a</mi> <mi>g</mi> <mo>(</mo> <msub> <mi>m</mi> <mi>k</mi> </msub> <mo>)</mo> <msub> <mi>q</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> <msub> <mi>p</mi> <mi>i</mi> </msub> <msub> <mi>q</mi> <mi>j</mi> </msub> <mo>+</mo> <mn>2</mn> <msub> <mi>&amp;alpha;Dm</mi> <mi>k</mi> </msub> <mo>+</mo> <mn>2</mn> <mi>&amp;gamma;</mi> <munder> <mi>&amp;Sigma;</mi> <mi>i</mi> </munder> <msub> <mi>T</mi> <mrow> <mi>k</mi> <mi>l</mi> </mrow> </msub> <mrow> <mo>(</mo> <msub> <mi>m</mi> <mi>k</mi> </msub> <mo>-</mo> <msub> <mi>m</mi> <mi>l</mi> </msub> <mo>)</mo> </mrow> </mrow>

wherein D is dependent on m_kThe value of the ith element on the diagonal is calculated by the following formula:

<mrow> <msub> <mi>D</mi> <mrow> <mi>i</mi> <mi>i</mi> </mrow> </msub> <mo>=</mo> <mfrac> <mn>1</mn> <mrow> <mn>2</mn> <msqrt> <mrow> <msubsup> <mi>m</mi> <mrow> <mi>k</mi> <mi>i</mi> </mrow> <mn>2</mn> </msubsup> <mo>+</mo> <mi>&amp;epsiv;</mi> </mrow> </msqrt> </mrow> </mfrac> </mrow>

wherein, is a positive number for smoothing the target;

all variables P, Q and m are initialized to [0,1 ]]According to the step size parameter omega_p、ω_qAnd ω_mAnd updating the variables until the algorithm converges to obtain a final user preference matrix, a final article feature matrix and a final feature selection vector.

5. The multi-domain recommendation method based on feature selection according to claim 1, wherein in step S1, each of said plurality of users has at least a first threshold number of scores and each of said plurality of items has at least a second threshold number of scores.