CN103455555B - Recommendation method and recommendation apparatus based on mobile terminal similarity - Google Patents

Abstract

This application discloses a kind of recommendation method based on mobile terminal similarity and recommendation apparatus, relate to moving communicating field. Wherein the recommendation method based on mobile terminal similarity includes: obtain data, inquiry data, data process, generation terminal attribute data acquisition system, Data Dimensionality Reduction, recommendation calculates and output is recommended; Recommendation apparatus based on the recommendation method of mobile terminal similarity includes: obtaining data cell, inquiry data cell, data processing unit, generation terminal attribute data aggregation unit, Data Dimensionality Reduction unit, recommend computing unit and output recommendation unit, wherein data processing unit includes: inquires about matrix unit, add data cell, similarity calculated and generation terminal similar matrix unit. The application provides the benefit that and proposes a kind of recommendation method based on mobile terminal similarity and recommendation apparatus, it is achieved that the difference for mobile terminal attribute carries out personalized recommendation, reduces the mean error of conventional recommendation method.

Description

Recommendation method and recommendation apparatus based on mobile terminal similarity

Technical field

The application relates to moving communicating field, particularly relates to a kind of recommendation method based on mobile terminal similarity and recommendation apparatus.

Background technology

By now, quantity of information is very abundant in internet development, greatly exceed the scope that people can accept, process and utilize. The information of bulk redundancy is full of network, and severe jamming user's selection to useful information, information overload has become as a problem demanding prompt solution, it is recommended that system is the important means of information filtering, it is possible to effectively alleviate problem of information overload.

Mobile commending system is the commending system being applied in mobile Internet field, is the extension of conventional recommendation systems, at present the research of mobile commending system is in the stage at the early-stage. Mobile commending system is basically identical with the thought of conventional recommendation systems, the method and the algorithm that adopt can also be general, but, feature and limitation due to mobile terminal and mobile network, mobile recommendation to be also affected, and first, in mobile Internet, user can access the Internet whenever and wherever possible, acquisition information, and different customer locations is also not quite similar for demand; Second, mobile interchange network users compares conventional internet user, and the context environmental faced is more complicated changeable; 3rd, the disposal ability of mobile terminal, screen size, input and output convenience different. Therefore, move commending system and have higher real-time, accuracy, personalized requirement.

According to investigation display, the differentiation of mobile terminal and mobile platform makes user present different individualized features and use habit, the multipair position of commending system and environment currently for mobile Internet are recommended, and there is no the recommendation method specially different mobile terminal attribute recommended.

Summary of the invention

The application provides a kind of recommendation method based on mobile terminal similarity and recommendation apparatus.

First aspect according to the application, it is provided that a kind of recommendation method based on mobile terminal similarity, including:

Obtain data: obtain user data package, user data package be expressed as by user property, item attribute, terminal attribute and attributes preferred form;

Inquiry data: the terminal attribute data in the user data package obtained are inquired about in the terminal attribute data set pre-build; these terminal attribute data are the set of the data with more than one dimension; and the type of data is redefined for the data type determined in each dimension, this data type includes not having the classifying type data of numerical values recited relation and have the numeric type data of numerical values recited relation;

Data process: the result for inquiry data processes, if there are the terminal attribute data inquired about in terminal attribute data set, then inquires about matrix, and this inquiry matrix is the terminal similar matrix corresponding with these terminal attribute data that inquiry pre-builds;

If terminal attribute data set is absent from the terminal attribute data inquired about, then these terminal attribute data are added into the terminal attribute data set pre-build, the similarity of all terminals in the terminal attribute data set calculating these terminal attribute data and pre-build, according to terminal Similarity Measure result, generate inquiry terminal similar matrix, and preserve;

This terminal Similarity Measure carries out according to below equation:

$\mathrm{deviceSim}(a,b)=\frac{\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}S(a_i,b_i)}{\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}W(a_i,b_i)}$

Wherein (a, b) for the terminal similarity of terminal attribute example a and terminal attribute example b for deviceSim;

S(a_i,b_i) for dimensional attribute similarity, represent the similarity of terminal attribute example a and terminal attribute example b attribute in terminal attribute i+1 dimension;

W(ai, bi) weights in terminal attribute i+1 dimension for terminal attribute example a and terminal attribute example b;

The computational methods of this dimensional attribute similarity are:

If terminal attribute is classifying type data on certain dimension, then this dimensional attribute similarity takes 1 value in terminal attribute data in this dimension time identical, takes 0 value time different;

If terminal attribute is numeric type data on certain dimension, dimensional attribute Similarity Measure is undertaken by below equation:

$S(a_i,b_i)=1-\frac{|a_i-b_i|}{\mathrm{Max}\left(D_i\right)-\mathrm{Min}\left(D_i\right)}$

Wherein S (a_i,b_i) for the dimensional attribute similarity of terminal attribute example a and terminal attribute example b;

a_iRepresent terminal attribute example a attribute in i+1 dimension; b_iRepresent terminal attribute example b attribute in i+1 dimension;

Max(D_i) for terminal attribute maximum of attribute in i+1 dimension, Min (D_i) for terminal attribute in i+1 dimension attribute minima;

Generate terminal attribute data acquisition system: according to the similarity threshold set, inquiry terminal similar matrix is processed, extract the similarity terminal attribute data acquisition system more than similarity threshold, generate the first data;

Data Dimensionality Reduction: the terminal corresponding to the terminal attribute in these first data is considered as similar terminals, extracts the UAD of this similar terminals, item attribute data and attributes preferred data, generate the second data;

Recommend to calculate: the second data separate proposed algorithm is generated preference and predicts and store;

Output is recommended: according to preference prediction output recommendation results.

Second aspect according to the application, it is provided that a kind of recommendation apparatus recommending method based on mobile terminal similarity, including:

Obtain data cell: be used for obtaining user data package, user data package be expressed as by user property, item attribute, terminal attribute and attributes preferred form;

Inquiry data cell: for the terminal attribute data in the user data package obtained are inquired about in the terminal attribute data set pre-build; these terminal attribute data are the set of the data with more than one dimension; and the type of data is redefined for the data type determined in each dimension, this data type includes not having the classifying type data of numerical values recited relation and have the numeric type data of numerical values recited relation;

Data processing unit: for processing for the result of inquiry data, if terminal attribute data set exists the terminal attribute data inquired about, then inquiring about matrix, this inquiry matrix is the terminal similar matrix corresponding with these terminal attribute data that inquiry pre-builds;

If terminal attribute data set is absent from the terminal attribute data inquired about, then these terminal attribute data are added into the terminal attribute data set pre-build, the similarity of all terminals in the terminal attribute data set calculating these terminal attribute data and pre-build, according to terminal Similarity Measure result, generate inquiry terminal similar matrix, and preserve;

This terminal Similarity Measure carries out according to below equation:

$\mathrm{deviceSim}(a,b)=\frac{\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}S(a_i,b_i)}{\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}W(a_i,b_i)}$

Wherein (a, b) for the terminal similarity of terminal attribute example a and terminal attribute example b for deviceSim;

S(a_i,b_i) for dimensional attribute similarity, represent the similarity of terminal attribute example a and terminal attribute example b attribute in terminal attribute i+1 dimension;

W(ai, bi) weights in terminal attribute i+1 dimension for terminal attribute example a and terminal attribute example b;

The computational methods of this dimensional attribute similarity are:

If terminal attribute is classifying type data on certain dimension, then this dimensional attribute similarity takes 1 value in terminal attribute data in this dimension time identical, takes 0 value time different;

If terminal attribute is numeric type data on certain dimension, dimensional attribute Similarity Measure is undertaken by below equation:

$S(a_i,b_i)=1-\frac{|a_i-b_i|}{\mathrm{Max}\left(D_i\right)-\mathrm{Min}\left(D_i\right)}$

Wherein S (a_i,b_i) for the dimensional attribute similarity of terminal attribute example a and terminal attribute example b;

a_iRepresent terminal attribute example a attribute in i+1 dimension; b_iRepresent terminal attribute example b attribute in i+1 dimension;

Max(D_i) for terminal attribute maximum of attribute in i+1 dimension, Min (D_i) for terminal attribute in i+1 dimension attribute minima;

Generate terminal attribute data aggregation unit: for inquiry terminal similar matrix being processed according to the similarity threshold set, extract the similarity terminal attribute data acquisition system more than similarity threshold, generate the first data;

Data Dimensionality Reduction unit: for the terminal corresponding to the terminal attribute in these first data is considered as similar terminals, extracts the UAD of this similar terminals, item attribute data and attributes preferred data, generate the second data;

Recommend computing unit: predict for the second data separate proposed algorithm is generated preference and store;

Output recommendation unit: for according to preference prediction output recommendation results;

Wherein data processing unit includes:

Inquiry matrix unit: during for there are, in terminal attribute data set, the terminal attribute data inquired about, the terminal similar matrix corresponding with these terminal attribute data that inquiry pre-builds;

Add data cell: during for being absent from, in terminal attribute data set, the terminal attribute data inquired about, these terminal attribute data are added into the terminal attribute data set pre-build;

Similarity calculated: for the similarity of all terminals in the terminal attribute data calculating this interpolation and the terminal attribute data set pre-build;

Generate inquiry terminal similar matrix unit: for according to terminal Similarity Measure result, generating inquiry terminal similar matrix, and preserve.

The application provides the benefit that and proposes a kind of recommendation method based on mobile terminal similarity and recommendation apparatus, it is achieved that the difference for mobile terminal attribute carries out personalized recommendation, reduces the mean error of conventional recommendation method.

Accompanying drawing explanation

Fig. 1 is the data Establishing process figure of the recommendation method based on mobile terminal similarity;

Fig. 2 is the flow chart of the recommendation embodiment of the method based on mobile terminal similarity;

Fig. 3 is the structured flowchart of the recommendation apparatus embodiment recommending method based on mobile terminal similarity;

Fig. 4 is the structured flowchart of the recommendation apparatus data processing unit recommending method based on mobile terminal similarity.

Detailed description of the invention

The present invention is described in further detail in conjunction with accompanying drawing below by detailed description of the invention.

The application provides a kind of recommendation method based on mobile terminal similarity and recommendation apparatus.

Embodiment one:

The application carries out based on collaborative filtering based on the recommendation method of mobile terminal similarity, owing to collaborative filtering generally there are cold start-up problem, if system not existing initial data when namely recommending or initial data being less, the recommendation results precision that cannot export recommendation results or output is relatively low, in order to solve cold start-up problem, system need to be carried out the foundation of initial data, the foundation of initial data is adopted the mode of typing user data by the present embodiment, simultaneously in order to facilitate subsequent user to use the method to recommend, also user is generated in follow-up required data and stores, based on mobile terminal similarity recommendation method data Establishing process figure as shown in Figure 1, including:

Obtain data 10: obtain user data package, user data package be expressed as by user property, item attribute, terminal attribute and attributes preferred form;

Initial data when setting up in carrying out system, obtaining user data package can be a document, the form of document includes but not limited to txt, word and xls, the document comprises user property, item attribute, terminal attribute and attributes preferred four data row, is called UAD group, item attribute data set, terminal attribute data set and attributes preferred data set; If attributes preferred disappearance, available specific symbol is indicated, such as null etc. Every data line of document is mutual corresponding relation, for mutually corresponding user property, item attribute, terminal attribute and attributes preferred. User property, item attribute, terminal attribute and attributes preferred data are likely various structures, can pass through to create a rational protocol rule, be standardized data processing. In this application, user property is used for identifying different user, can directly be standardized as digital data, such as 1,2,3 etc.; Attributes preferred it is standardized as a range of numeral, such as the scoring of 1 to 5, the satisfaction etc. of 1 to 100; Terminal attribute has various structures, can the data structure of setting terminal attribute according to actual needs, item attribute, normally due to data scale is limited, can not process. It is standardized data processing, is conducive to the execution of follow-up flow process.

In order to different user is recommended, it is necessary to understand the preference value of a certain project of different user, for user, the hobby of project is recommended, be only the effective way of recommendation. User, for the preference data of a certain project, obtains typically via the comment of user under a certain project, third party evaluation mechanism or alternate manner. The data obtained by such mode are frequently not and complete comprise user property, item attribute, terminal attribute and attributes preferred data, the attributes preferred data of usual excalation. The application is based on the purpose of the recommendation method of mobile terminal similarity, it is simply that obtain the attributes preferred data of this excalation more accurately.

The user data package that the application obtains can be expressed as by user property, item attribute, terminal attribute and attributes preferred form, terminal attribute has been had more by user property, item attribute and the attributes preferred data formed than traditional being expressed as, namely a kind of recommendation method based on mobile terminal similarity of the application is not merely based on user property and item attribute is considered attributes preferred, has been additionally based upon terminal attribute.

Terminal Similarity Measure 11: the terminal attribute data comprising all terminals in user data package are carried out terminal Similarity Measure;

Terminal Similarity Measure 11 is carry out terminal Similarity Measure for all terminals of terminal attribute data set in user data package.

Terminal Similarity Measure carries out according to the terminal similarity algorithm set, and the calculating of terminal similarity will be illustrated later.

Generate terminal similar matrix 12: generate the terminal similar matrix corresponding with terminal attribute according to the result of terminal Similarity Measure 11, and store;

The step generating terminal similar matrix 12 also includes: the data of the terminal attribute data set carrying out terminal Similarity Measure are carried out deduplication process and updated;

The data of terminal similar matrix are carried out deduplication process and update.

Data are carried out deduplication process and update in order that reduce amount of calculation, save memory space. If there being n different terminals, then by n total after terminal Similarity Measure²Individual data, it is contemplated that some characteristics of terminal similar matrix, can carry out data compression to it. Particularly as follows: owing to the similarity of two terminal rooms is by the restriction of terminal sequencing when carrying out terminal Similarity Measure, namely regardless of terminal sequencing when carrying out terminal Similarity Measure, the terminal similarity drawn is same value, therefore the terminal similar matrix drawn by all terminal attributes carry out terminal Similarity Measure is symmetrical matrix, symmetric data can be removed, retain upper triangle or the lower triangular matrix of symmetrical matrix; The terminal similar matrix additionally generated necessarily includes the similarity of same terminal, and namely terminal similarity is 1, can be deleted by this record, namely removes the cornerwise data of terminal similar matrix. The data of terminal similar matrix not only can be carried out above-mentioned deduplication process, it is also possible to removing the data of entirely different terminal, namely terminal similarity is the data of 0. The data of the terminal similar matrix after processing above are stored, covers original terminal similar matrix. After processing above, although the order of magnitude of data does not reduce, but saves memory space. The data of the terminal similar matrix after processing above are stored, covers original terminal similar matrix.

The terminal similar matrix corresponding with terminal attribute generated contains the similarity between terminal, this terminal similar matrix is stored, can first passing through the mode of inquiry when user uses, in order to simple flow, inquiry mode when user uses will be illustrated later.

By performing step 10��12, not only typing initial data in systems, solve cold start-up problem, and original user data bag and the terminal similar matrix corresponding with terminal attribute can be set up in the application is based on the system of the recommendation method of mobile terminal similarity, when user uses the application based on the recommendation method of mobile terminal similarity, can inquire about first with terminal attribute data set in the user data package pre-build in system and the terminal similar matrix corresponding with terminal attribute, and utilize the data that the user data package pre-build in system calculates as collaborative filtering basic, as shown in Figure 2, Fig. 2 is the application flow chart based on the recommendation embodiment of the method for mobile terminal similarity, including:

Obtaining data 10, identical with obtaining data 10 in Fig. 1, this repeats no more.

The user data package got in Fig. 2 is by user property, item attribute, terminal attribute and attributes preferred forms, the data of the user property of one mutual correspondence, item attribute, terminal attribute and attributes preferred composition are designated as preference record, it this user data package can be a preference record, can also be a plurality of preference record, utilize the application to be in that to obtain attributes preferred data value unknown in preference record based on the purpose of the recommendation method of mobile terminal similarity.

Inquiry data 21: the terminal attribute data in the user data package obtained are inquired about in the terminal attribute data set pre-build; these terminal attribute data are the set of the data with more than one dimension; and the type of data is redefined for the data type determined in each dimension, this data type includes not having the classifying type data of numerical values recited relation and have the numeric type data of numerical values recited relation;

The terminal attribute data obtained in user data package are inquired about in the terminal attribute data set pre-build, the terminal attribute data set that this pre-builds contains the terminal attribute data of accumulation in subsequent user use procedure in the initial data process of system as shown in Figure 1 and as shown in Figure 2. If the user data package obtained comprises a plurality of preference record, adopt the mode that the terminal attribute of every preference record is inquired about respectively, the terminal attribute data set pre-build is inquired about, if there is the terminal attribute data message of required inquiry, the terminal similar matrix pre-build can be directly utilized, namely system comprises similarity between the terminal attribute of this inquiry and other terminal attribute, it is not necessary to terminal similarity is calculated.

The terminal attribute of user has multiple, such as brand, system, version etc., a n-dimensional vector model can be used to represent a terminal having n attribute, as follows:

Device={D₀,D₁,������,D_i,������,D_n-1}

Wherein D_iRepresent terminal attribute attribute in i+1 dimension.

Data process 22: the result for inquiry data processes, if terminal attribute data set exists the terminal attribute data inquired about, then inquiry matrix 221, this inquiry matrix 221 is the terminal similar matrix corresponding with these terminal attribute data that inquiry pre-builds;

If terminal attribute data set is absent from the terminal attribute data inquired about, then these terminal attribute data are added into the terminal attribute data set pre-build, the similarity of all terminals in the terminal attribute data set calculating these terminal attribute data and pre-build, according to terminal Similarity Measure result, generate inquiry terminal similar matrix, and preserve;

If there are the terminal attribute data inquired about in terminal attribute data set, then inquiry matrix 221, inquiry matrix is the similarity inquiring about this terminal with other all terminals, extracts Query Result from matrix. The preserving type of Query Result can be multiple, as matrix or the form identical with the flesh and blood that matrix represents can be adopted to preserve.

If as in figure 2 it is shown, terminal attribute data set is absent from the terminal attribute data inquired about, performing: add data 222, Similarity Measure 223 and generate terminal similar matrix 224;

Add data 222: terminal attribute data are added into the terminal attribute data set pre-build;

In the present embodiment, the mode user data package got being added in the user data package document pre-build in system can be adopted, be about in the user property in the user data package got, item attribute, terminal attribute and attributes preferred UAD group, item attribute data set, terminal attribute data set and the attributes preferred data set being added separately to pre-build. So, both terminal attribute data are added into the terminal attribute data set pre-build, save again the user data package of new acquisition, the data in this user data package can as the accumulation of data in system, as the basic data that subsequent user uses.

Similarity Measure 223: calculate newly added terminal attribute data and the similarity of all terminals in the terminal attribute data set pre-build;

Terminal Similarity Measure carries out according to below equation:

$\mathrm{deviceSim}(a,b)=\frac{\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}S(a_i,b_i)}{\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}W(a_i,b_i)}$

Wherein (a, b) for the terminal similarity of terminal attribute example a and terminal attribute example b for deviceSim;

S(a_i,b_i) for dimensional attribute similarity, represent the similarity of terminal attribute example a and terminal attribute example b attribute in terminal attribute i+1 dimension;

W(ai, bi) weights in terminal attribute i+1 dimension for terminal attribute example a and terminal attribute example b;

The computational methods of dimensional attribute similarity are:

If terminal attribute is not for having the classifying type data of numerical values recited relation on certain dimension, then dimensional attribute similarity takes 1 value in terminal attribute data in this dimension time identical, takes 0 value time different;

If terminal attribute is for there being the numeric type data of numerical values recited relation on certain dimension, dimensional attribute Similarity Measure is undertaken by below equation:

$S(a_i,b_i)=1-\frac{|a_i-b_i|}{\mathrm{Max}\left(D_i\right)-\mathrm{Min}\left(D_i\right)}$

Wherein S (a_i,b_i) for the dimensional attribute similarity of terminal attribute example a and terminal attribute example b;

a_iRepresent terminal attribute example a attribute in i+1 dimension; b_iRepresent terminal attribute example b attribute in i+1 dimension;

Max(D_i) for terminal attribute maximum of attribute in i+1 dimension, Min (D_i) for terminal attribute in i+1 dimension attribute minima.

Generate terminal similar matrix 224: according to Similarity Measure 223 result, generate inquiry terminal similar matrix, and preserve;

Similarity Measure 223 calculates newly added terminal attribute data and the similarity of all terminals in the terminal attribute data set pre-build, generates and comprises these newly added terminal attribute data and the inquiry terminal similar matrix of the similarity of all terminals in the terminal attribute data set pre-build.

This inquiry terminal similar matrix can be added into the terminal similar matrix pre-build, and preserve, it is achieved the terminal similar matrix of similarity data that containing after interpolation pre-builds and newly added terminal attribute and other all terminal similarity data.

Generate terminal attribute data acquisition system 23: according to the similarity threshold set, inquiry terminal similar matrix is processed, extract the similarity terminal attribute set more than similarity threshold, generate the first data;

These first data can adopt the form of matrix to represent, contains similarity more than the similarity information between end message and the terminal of similarity threshold in the first data.

Data Dimensionality Reduction 24: the terminal attribute in the first data is considered as similar terminals, extracts the UAD of described similar terminals, item attribute data and attributes preferred data, generates the second data;

Similarity in inquiry terminal similar matrix is considered as similar terminals more than the terminal of similarity threshold, owing to the terminal attributive information of similar terminals is likely to identical, therefore the user property of this terminal attribute cannot be extracted according to terminal attribute, item attribute and attributes preferred, can adopt and search corresponding user property according to terminal attribute, search corresponding item attribute and attributes preferred mode further according to user property. Corresponding user property is searched according to terminal attribute, the principle followed is: if the first data acquisition matrix or the form identical with the flesh and blood that matrix represents represent, this terminal attribute position in former participation terminal Similarity Measure terminal attribute data set can be found by the dimensional information at terminal attribute place in matrix, the user property of correspondence can be found according to this position; And first look for user property but not search item attribute and attributes preferred reason being in that, only having user property is absent from repeating, item attribute that certain user property is corresponding and the attributes preferred polyisomenism that would be likely to occur, it is impossible to one_to_one corresponding.

Terminal attribute in first data is considered as similar terminals, extract the UAD of described similar terminals, item attribute data and attributes preferred data, generate the second data, these second data be regenerate only comprise UAD group, the document of item attribute data set and attributes preferred data set.

By performing generation terminal attribute data acquisition system 23 step and Data Dimensionality Reduction 24 step, achieve, by what the user data package of input comprised, there is user property, item attribute, terminal attribute and attributes preferred data are converted to and comprise user property, item attribute and the second attributes preferred data, so both considered the terminal attribute of equipment, and be easy to again utilize traditional proposed algorithm to carry out data recommending to calculate.

Recommend calculating 25: the second described data separate proposed algorithm is generated preference prediction;

The proposed algorithm adopted is slopeone proposed algorithm, and slopeone proposed algorithm is project-based collaborative filtering, adopts this algorithm, according to existing subscriber, the attributes preferred value of project can be calculated the preference to the project not providing attributes preferred value and predict.

Output recommendation 26: according to preference prediction output recommendation results.

In the present embodiment, the recommendation results of output is the output user's attributes preferred data value to project.

By step 21��26 it can be seen that this step is calculated just for a preference record in the user data package obtained, if this user data package comprises M bar preference record, step 21��26 need to be performed M time. How flow process is performed repeatedly by initialization system automatically, is known to the skilled person general knowledge, repeats no more herein.

Embodiment two:

Second aspect according to the application, it is provided that a kind of recommendation apparatus recommending method based on mobile terminal similarity, recommends the structured flowchart of recommendation apparatus embodiment of method as it is shown on figure 3, include based on mobile terminal similarity:

Obtain data cell 30: be used for obtaining user data package, user data package be expressed as by user property, item attribute, terminal attribute and attributes preferred form;

Inquiry data cell 31: for the terminal attribute data in the user data package obtained are inquired about in the terminal attribute data set pre-build; these terminal attribute data are the set of the data with more than one dimension; and the type of data is redefined for the data type determined in each dimension, this data type includes not having the classifying type data of numerical values recited relation and have the numeric type data of numerical values recited relation;

Data processing unit 32: for processing for the result of inquiry data, if terminal attribute data set exists the terminal attribute data inquired about, then inquiring about matrix, this inquiry matrix is the terminal similar matrix corresponding with these terminal attribute data that inquiry pre-builds;

If terminal attribute data set is absent from the terminal attribute data inquired about, then these terminal attribute data are added into the terminal attribute data set pre-build, the similarity of all terminals in the terminal attribute data set calculating these terminal attribute data and pre-build, according to terminal Similarity Measure result, generate inquiry terminal similar matrix, and preserve;

This terminal Similarity Measure carries out according to below equation:

$\mathrm{deviceSim}(a,b)=\frac{\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}S(a_i,b_i)}{\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}W(a_i,b_i)}$

Wherein (a, b) for the terminal similarity of terminal attribute example a and terminal attribute example b for deviceSim;

S(a_i,b_i) for dimensional attribute similarity, represent the similarity of terminal attribute example a and terminal attribute example b attribute in terminal attribute i+1 dimension;

W(ai, bi) weights in terminal attribute i+1 dimension for terminal attribute example a and terminal attribute example b;

The computational methods of this dimensional attribute similarity are:

If terminal attribute is classifying type data on certain dimension, then this dimensional attribute similarity takes 1 value in terminal attribute data in this dimension time identical, takes 0 value time different;

If terminal attribute is numeric type data on certain dimension, dimensional attribute Similarity Measure is undertaken by below equation:

$S(a_i,b_i)=1-\frac{|a_i-b_i|}{\mathrm{Max}\left(D_i\right)-\mathrm{Min}\left(D_i\right)}$

Wherein S (a_i,b_i) for the dimensional attribute similarity of terminal attribute example a and terminal attribute example b;

a_iRepresent terminal attribute example a attribute in i+1 dimension; b_iRepresent terminal attribute example b attribute in i+1 dimension;

Max(D_i) for terminal attribute maximum of attribute in i+1 dimension, Min (D_i) for terminal attribute in i+1 dimension attribute minima;

Generate terminal attribute data aggregation unit 33: for inquiry terminal similar matrix being processed according to the similarity threshold set, extract the similarity terminal attribute data acquisition system more than similarity threshold, generate the first data;

Data Dimensionality Reduction unit 34: for the terminal corresponding to the terminal attribute in these first data is considered as similar terminals, extracts the UAD of this similar terminals, item attribute data and attributes preferred data, generate the second data;

Recommend computing unit 35: predict for the second data separate proposed algorithm is generated preference and store;

Output recommendation unit 36: for according to preference prediction output recommendation results;

The structured flowchart of recommendation apparatus data processing unit 32 of method is recommended as shown in Figure 4 based on mobile terminal similarity, including:

Inquiry matrix unit 321: during for there are, in terminal attribute data set, the terminal attribute data inquired about, the terminal similar matrix corresponding with these terminal attribute data that inquiry pre-builds;

Add data cell 322: during for being absent from, in terminal attribute data set, the terminal attribute data inquired about, these terminal attribute data are added into the terminal attribute data set pre-build;

Similarity calculated 323: for the similarity of all terminals in the terminal attribute data calculating this interpolation and the terminal attribute data set pre-build;

Generate terminal similar matrix unit 324: for according to terminal Similarity Measure result, generating inquiry terminal similar matrix, and preserve.

In sum, the application provides the benefit that and proposes a kind of recommendation method based on mobile terminal similarity and recommendation apparatus, achieve the difference for mobile terminal attribute and carry out personalized recommendation, owing to considering terminal attribute, therefore reduce the mean error of conventional recommendation method.

It will be appreciated by those skilled in the art that, in above-mentioned embodiment, all or part of step of various methods can be carried out instruction related hardware by program and completes, this program can be stored in a computer-readable recording medium, and storage medium may include that read only memory, random access memory, disk or CD etc.

Above content is in conjunction with specific embodiment further description made for the present invention, it is impossible to assert that specific embodiment of the invention is confined to these explanations. For general technical staff of the technical field of the invention, without departing from the inventive concept of the premise, it is also possible to make some simple deduction or replace.

Claims (4)

1. the recommendation method based on mobile terminal similarity, it is characterised in that including:

Obtain data: obtain user data package, user data package be expressed as by user property, item attribute, terminal attribute and attributes preferred form;

Inquiry data: the terminal attribute data in the user data package obtained are inquired about in the terminal attribute data set pre-build, described terminal attribute data are the set of the data with more than one dimension, and the type of data is redefined for the data type determined in each dimension, described data type includes not having the classifying type data of numerical values recited relation and have the numeric type data of numerical values recited relation;

Data process: the result for inquiry data processes, if there are the terminal attribute data inquired about in terminal attribute data set, then inquires about matrix, and described inquiry matrix is the terminal similar matrix corresponding with these terminal attribute data that inquiry pre-builds;

If terminal attribute data set is absent from the terminal attribute data inquired about, then described terminal attribute data are added into the terminal attribute data set pre-build, the similarity of all terminals in the terminal attribute data set calculating described terminal attribute data and pre-build, according to terminal Similarity Measure result, generate inquiry terminal similar matrix, and preserve;

Described terminal Similarity Measure carries out according to below equation:

$deviceSim(a,b)=\frac{\underset{i=0}{\overset{n-1}{\Σ}}S(a_i,b_i)}{\underset{i=0}{\overset{n-1}{\Σ}}W(a_i,b_i)}$

Wherein (a, b) for the terminal similarity of terminal attribute example a and terminal attribute example b for deviceSim;

S(a_i,b_i) for dimensional attribute similarity, represent the similarity of terminal attribute example a and terminal attribute example b attribute in terminal attribute i+1 dimension;

W (ai, bi) is the weights in terminal attribute i+1 dimension for terminal attribute example a and terminal attribute example b;

The computational methods of described dimensional attribute similarity are:

If terminal attribute is classifying type data on certain dimension, then described dimensional attribute similarity takes 1 value in terminal attribute data in this dimension time identical, takes 0 value time different;

If terminal attribute is numeric type data on certain dimension, then described dimensional attribute Similarity Measure is undertaken by below equation:

$S(a_i,b_i)=1-\frac{|a_i-b_i|}{Max\left(D_i\right)-Min\left(D_i\right)}$

Wherein S (a_i,b_i) for the dimensional attribute similarity of terminal attribute example a and terminal attribute example b;

a_iRepresent terminal attribute example a attribute in i+1 dimension; b_iRepresent terminal attribute example b attribute in i+1 dimension, D_iRepresent terminal attribute attribute in i+1 dimension;

Max(D_i) for terminal attribute maximum of attribute in i+1 dimension, Min (D_i) for terminal attribute in i+1 dimension attribute minima;

Generate terminal attribute data acquisition system: according to the similarity threshold set, inquiry terminal similar matrix is processed, extract the similarity terminal attribute data acquisition system more than similarity threshold, generate the first data;

Data Dimensionality Reduction: the terminal corresponding to the terminal attribute in described first data is considered as similar terminals, extracts the UAD of described similar terminals, item attribute data and attributes preferred data, generate the second data;

Recommend to calculate: described second data separate proposed algorithm is generated preference and predicts and store;

Output is recommended: according to preference prediction output recommendation results.

2. the recommendation method based on mobile terminal similarity as claimed in claim 1, it is characterised in that described proposed algorithm is SlopeOne proposed algorithm.

3. the recommendation apparatus recommending method based on mobile terminal similarity, it is characterised in that including:

Obtain data cell: be used for obtaining user data package, user data package be expressed as by user property, item attribute, terminal attribute and attributes preferred form;

Inquiry data cell: for the terminal attribute data in the user data package obtained are inquired about in the terminal attribute data set pre-build, described terminal attribute data are the set of the data with more than one dimension, and the type of data is redefined for the data type determined in each dimension, described data type includes not having the classifying type data of numerical values recited relation and have the numeric type data of numerical values recited relation;

Data processing unit: for processing for the result of inquiry data, if terminal attribute data set exists the terminal attribute data inquired about, then inquiring about matrix, described inquiry matrix is the terminal similar matrix corresponding with these terminal attribute data that inquiry pre-builds;

If terminal attribute data set is absent from the terminal attribute data inquired about, then these terminal attribute data are added into the terminal attribute data set pre-build, the similarity of all terminals in the terminal attribute data set calculating these terminal attribute data and pre-build, according to terminal Similarity Measure result, generate inquiry terminal similar matrix, and preserve;

Described terminal Similarity Measure carries out according to below equation:

$deviceSim(a,b)=\frac{\underset{i=0}{\overset{n-1}{\Σ}}S(a_i,b_i)}{\underset{i=0}{\overset{n-1}{\Σ}}W(a_i,b_i)}$

Wherein (a, b) for the terminal similarity of terminal attribute example a and terminal attribute example b for deviceSim;

S(a_i,b_i) for dimensional attribute similarity, represent the similarity of terminal attribute example a and terminal attribute example b attribute in terminal attribute i+1 dimension;

W (ai, bi) is the weights in terminal attribute i+1 dimension for terminal attribute example a and terminal attribute example b;

The computational methods of described dimensional attribute similarity are:

If terminal attribute is classifying type data on certain dimension, then described dimensional attribute similarity takes 1 value in terminal attribute data in this dimension time identical, takes 0 value time different;

If terminal attribute is numeric type data on certain dimension, dimensional attribute Similarity Measure is undertaken by below equation:

$S(a_i,b_i)=1-\frac{|a_i-b_i|}{Max\left(D_i\right)-Min\left(D_i\right)}$

Wherein S (a_i,b_i) for the dimensional attribute similarity of terminal attribute example a and terminal attribute example b;

a_iRepresent terminal attribute example a attribute in i+1 dimension; b_iRepresent terminal attribute example b attribute in i+1 dimension, D_iRepresent terminal attribute attribute in i+1 dimension;

Max(D_i) for terminal attribute maximum of attribute in i+1 dimension, Min (D_i) for terminal attribute in i+1 dimension attribute minima;

Generate terminal attribute data aggregation unit: for inquiry terminal similar matrix being processed according to the similarity threshold set, extract the similarity terminal attribute data acquisition system more than similarity threshold, generate the first data;

Data Dimensionality Reduction unit: for the terminal corresponding to the terminal attribute in described first data is considered as similar terminals, extracts the UAD of described similar terminals, item attribute data and attributes preferred data, generate the second data;

Recommend computing unit: predict for the second described data separate proposed algorithm is generated preference and store;

Output recommendation unit: for according to preference prediction output recommendation results.

4. as claimed in claim 3 a kind of based on mobile terminal similarity recommend method recommendation apparatus, it is characterised in that described data processing unit includes:

Inquiry matrix unit: during for there are, in terminal attribute data set, the terminal attribute data inquired about, the terminal similar matrix corresponding with these terminal attribute data that inquiry pre-builds;

Add data cell: during for being absent from, in terminal attribute data set, the terminal attribute data inquired about, these terminal attribute data are added into the terminal attribute data set pre-build;

Similarity calculated: for the similarity of all terminals in the terminal attribute data calculating described interpolation and the terminal attribute data set pre-build;

Generate inquiry terminal similar matrix unit: for according to terminal Similarity Measure result, generating inquiry terminal similar matrix, and preserve.