CN103886073B - Coal information commending system based on collaborative filtering - Google Patents

Abstract

The present invention proposes a kind of coal information commending system based on collaborative filtering, and the system includes coal information collaborative filtering module, community users colony access log module and recommending module；Coal information collaborative filtering module is responsible for the access log for analyzing user, finds out user neighbour using the algorithm of collaborative filtering；Community users colony access log module is responsible for the management of daily record；Recommending module is responsible for calculating the information of recommendation and recommending user.After using the system, can recommend to meet the project or information of its requirement, realize personalized service according to the hobby of user.Interest preference of the coal information commending system based on each user, there is provided meet the targetedly recommendation of specific user, the convenience that this recommendation enhances the experience of user, the user for improving uses also improves the work efficiency of user.

Description

Coal information commending system based on collaborative filtering

Technical field

The invention belongs to Coal Information System.

Background technology

Due to developing rapidly for Internet, the information capacity of every field is very huge, including colliery field.At this Under background, the search engine system in colliery field has obtained extensive research and application, finds useful information for people and provides It is convenient.However, with the continuous improvement of people's Search Requirement, the deep processing of retrieval result is increasingly becoming one of the area research Emphasis.

The search modes of traditional " One-Size-Fit-All " can not allow user to be satisfied with, the substitute is retrieval knot The personalization of fruit, and " people looks for information " is replaced with the pattern of " information looks for people ".Then, coal information commending system just seems ten Divide necessity.

Commending system refers to be liked personalized recommendation according to personal as the system of output or in extensive optional object Guide well user's system alternatively.Coal information commending system proposed by the present invention is according to the daily record for having been browsed before user Content, by new, not browsed, user may coal information interested be automatically pushed to user.

Collaborative filtering is one of relatively early method being used in Technologies of Recommendation System in E-Commerce, and its core concept is that user inclines To the commodity or content bought in customer group of the purchase with similar interests hobby, basic starting point is：User can be Classify by interest；Evaluation of the user to different information contains the interest-degree of user；Evaluation of the user to a unknown message Will be similar with the evaluation of its similar interests user.

In the environment of community, often possess identical interest, demand, expectation between user and go selection similar with motivation Information, such case are to be provided the foundation based on the recommendation of collaborative filtering.

The content of the invention

In order to realize the present invention, it is proposed that a kind of based on collaborative filtering, the coal information commending system of Community-oriented, The system includes coal information collaborative filtering module, community users colony access log module and recommending module；Coal information is assisted Be responsible for analyzing the access log of user with filtering module, user neighbour is found out using the algorithm of collaborative filtering；Community users colony Access log module is responsible for the management of daily record；Recommending module is responsible for calculating the information of recommendation and recommending user；Wherein in colliery In Collaborative Filtering module, filtered using following methods：

(1) system initialization：Based on the access log of user in the community of colliery, coal mine user-colliery webpage scoring is built Matrix A, matrix A include s user u₁,u₂,...,u_sTo t webpage p₁,p₂,...,p_tScoring, A for s × t ranks matrix, s Natural number is with t, the element a in matrix_ijScorings of the user i to webpage j is represented, i and j is natural number, and 1≤i≤s, 1 ≤ j≤t,

(2) rating matrix optimization：T webpage obtains k cluster after K-Means clustering algorithms, is designated as c₁,c₂,..., c_k, k is number of clusters mesh, and k is natural number；Cluster c_mComprising webpage number be N_m, this N_mSequence number of the individual webpage in original t webpage Respectively f (m, 1), f (m, 2) ..., f (m, N_m), f function is used for the original number of webpage in cluster after calculating is clustered, i.e., in original The sequence number come in t webpage, f (m, N_m) represent cluster c_mIn N_mThe original number of individual webpage, m is natural number, 1≤m≤k, 1≤ N_m≤ t, the rating matrix after being optimizedWherein b_imThe element of representing matrix B,

(3) neighbor searching：User u_vThe cluster sequence number collection for scoring is combined into M_v, user u_wThe cluster sequence number collection for scoring is combined into M_w, User u_vAnd u_wThe cluster sequence number collection for all scoring is combined into M, v and w and is natural number, 1≤v≤s, 1≤w≤s, then user u_vAnd u_w Between similarity beWhereinWithUser is represented respectively u_vAnd u_wAverage score to webpage, tries to achieve each user u_iIn the user space, by similarity sort from high to low it is near Adjacent user's set U_i；

(4) recommend：According to U_iDraw set U_iInterior all users, by scoring from high to low, the collections of web pages that browses P_i, user u_iThe collections of web pages for having browsed is P'_i, then gather (P_i-P'_i) in before L coal information webpage be user u_iIt is interested Content.

After using system above, can recommend to meet the project or information of its requirement according to the hobby of user, can be real Existing personalized service.After introducing commending system, can be believed according to the hobby of coal mine user, colliery that may be interested user Breath is formed and recommends user in the form of a list.User can be so allowed to search oneself in less range of information interested Content, whole information browse process are more quick, are changed into " information looks for people " from the mode of original " people looks for information ".Coal information Interest preference of the commending system based on each user, there is provided meet the targetedly recommendation of specific user, this recommendation strengthens The convenience that the experience of user, the user for improving use, also improves the work efficiency of user.

Description of the drawings

Fig. 1 is the system structure diagram of the present invention.

Specific embodiment

Based on collaborative filtering, the coal information commending system coal information collaborative filtering module of Community-oriented, community User group's access log module, recommending module.Wherein coal information collaborative filtering module is responsible for the access log for analyzing user, User neighbour is found out using the algorithm of collaborative filtering；Community users colony access log module is responsible for the management of daily record；Recommend mould Block is responsible for calculating the information of recommendation and recommending user.Coal information collaborative filtering module is the emphasis of the system, and which mainly walks Suddenly it is：

(1) system initialization：The step is based primarily upon the access log of user in the community of colliery, builds coal mine user-coal Ore deposit webpage rating matrix, is designated as A, it is assumed that the matrix includes s user u₁,u₂,...,u_sTo t webpage p₁,p₂,...,p_tComment Point, i.e. matrixes of the A for s × t ranks, s and t are natural number, the element a in matrix_ijRepresent scorings of the user i to webpage j, i and j It is natural number, and 1≤i≤s, 1≤j≤t.Matrix A can be expressed as：

(2) rating matrix optimization：Under normal circumstances, user's score data is fewer, and continuous with webpage quantity t Increase, cause rating matrix A extremely sparse, the final precision for affecting to recommend.In view of this, the present invention is first to rating matrix A Do and further optimize, the method for optimization is that webpage is clustered, cluster process adopts K-Means algorithms.

Assume that t webpage is obtained k cluster after using K-Means clustering algorithms, be designated as c₁,c₂,...,c_k, k is cluster Number, k are natural number.Assume cluster c_mComprising webpage number be N_m, this N_mSequence number of the individual webpage in original t webpage point Not Wei f (m, 1), f (m, 2) ..., f (m, N_m), f function is used for the original number of webpage in cluster after calculating is clustered, i.e., in original t Sequence number in individual webpage, f (m, N_m) represent cluster c_mIn N_mThe original number of individual webpage, m is natural number, 1≤m≤k, 1≤N_m≤ t。

Through above-mentioned optimization process, the rating matrix B after being optimized.

Wherein b_imThe element of representing matrix B,

(3) neighbor searching：Based on the rating matrix B after optimization, the user of same or similar interest can be found out, user it Between the calculating of similarity carried out using improved cosine similarity method.If user is u_vThe cluster sequence number collection for scoring is combined into M_v, user u_wThe cluster sequence number collection for scoring is combined into M_w, user u_vAnd u_wThe cluster sequence number collection for all scoring is combined into M, v and w and is natural number, 1≤v ≤ s, 1≤w≤s, then user u_vAnd u_wBetween similarity be：

WhereinWithUser u is represented respectively_vAnd u_wAverage score to webpage.

According to above-mentioned similarity calculating method, can be in the hope of each user u_iIn the user space, by similarity from The neighbour user set U of high to Low sequence_i。

(4) recommend：According to the user u for obtaining_iNeighbour user set U_i, it can be deduced that set U_iInterior all users, press Scoring from high to low, collections of web pages P that browses_i, it is assumed that user u_iThe collections of web pages for having browsed is P'_i, then gather (P_i-P '_i) in before L coal information webpage possibly user u_iContent interested, can recommend user u as content recommendation_i。

The system can be recommended to meet the project or information of its requirement according to the hobby of user, be a kind of personalized Service system.After introducing commending system, can be according to the hobby of coal mine user, user possible coal information shape interested Into recommending user in the form of a list.Can thus allow user search in less range of information oneself it is interested in Hold, whole information browse process is more quick, is changed into " information looks for people " from the mode of original " people looks for information ".Coal information is pushed away Recommend interest preference of the system based on each user, there is provided meet the targetedly recommendation of specific user, this recommendation is enhanced The convenience that the experience of user, the user for improving use, also improves the work efficiency of user.

Claims (1)

1. a kind of coal information commending system based on collaborative filtering, the system include coal information collaborative filtering module, community User group's access log module and recommending module；Coal information collaborative filtering module is responsible for the access log for analyzing user, profit User neighbour is found out with the algorithm of collaborative filtering；Community users colony access log module is responsible for the management of daily record；Recommending module It is responsible for calculating the information of recommendation and recommending user；Wherein in coal information collaborative filtering module, carried out using following methods Filter：

(1) system initialization：Based on the access log of user in the community of colliery, coal mine user-colliery webpage rating matrix is built A, matrix A include s user u₁,u₂,...,u_sTo t webpage p₁,p₂,...,p_tScoring, A for s × t ranks matrix, s and t Natural number is, the element a in matrix_ijScorings of the user i to webpage j is represented, i and j is natural number, and 1≤i≤s, 1≤j ≤ t,

(2) rating matrix optimization：T webpage obtains k cluster after K-Means clustering algorithms, is designated as c₁,c₂,...,c_k, k is Number of clusters mesh, k are natural number；Cluster c_mComprising webpage number be N_m, this N_mSequence number of the individual webpage in original t webpage is respectively f (m,1),f(m,2),...,f(m,N_m), f function is used for the original number of webpage in cluster after calculating is clustered, i.e., in original t net Sequence number in page, f (m, N_m) represent cluster c_mIn N_mThe original number of individual webpage, m is natural number, 1≤m≤k, 1≤N_m≤ t, obtains Rating matrix to after optimizationWherein b_imThe element of representing matrix B,

$b_{\mathrm{im}}=\frac{1}{N_m}\underset{n=1}{\overset{N_m}{\mathrm{\Σ}}}{a}_{\mathrm{if}(m,n)};$

(3) neighbor searching：User u_vThe cluster sequence number collection for scoring is combined into M_v, user u_wThe cluster sequence number collection for scoring is combined into M_w, user u_vAnd u_wThe cluster sequence number collection for all scoring is combined into M, v and w and is natural number, 1≤v≤s, 1≤w≤s, then user u_vAnd u_wBetween Similarity be $\mathrm{sim}(u_v,u_w)=\frac{{\mathrm{\Σ}}_{z\∈M}(b_{\mathrm{vz}}-{\stackrel{\‾}{b}}_v)({\stackrel{\‾}{b}}_{\mathrm{wz}}-{\stackrel{\‾}{b}}_w)}{\sqrt{{\mathrm{\Σ}}_{z\∈M_v}(b_{\mathrm{vz}}-{\stackrel{\‾}{b}}_v)}\sqrt{{\mathrm{\Σ}}_{z\∈M_w}(b_{\mathrm{wz}}-{\stackrel{\‾}{b}}_w)}},$ WhereinWithUser u is represented respectively_vAnd u_w Average score to webpage, tries to achieve each user u_iIn the user space, the neighbour that sorted by similarity from high to low uses Family set U_i；

(4) recommend：According to U_iDraw set U_iInterior all users, by scoring from high to low, collections of web pages P that browses_i, User u_iThe collections of web pages for having browsed is P'_i, then gather (P_i-P'_i) in before L coal information webpage be user u_iInterested is interior Hold.