CN104331433A - Tobacco information recommending method based on mobile terminal user logs - Google Patents

Abstract

The invention relates to a tobacco information recommending method based on mobile terminal user logs. The method comprises three stages including user category determination, inter-user similarity model building and tobacco information recommending. The method concretely comprises the following steps of 1, user category determination, 2, inter-user similarity model building and 3, tobacco information recommending. The tobacco information recommending method has the advantages that 1, registration information of users in the tobacco information logs is analyzed, and the classification of different users is completed; 2, by aiming at the users, a user-user similarity model is built for meeting the specific tobacco information demands through combining the historically browsed tobacco information of the users in the recommending process, and the recommending method based on the historically browsed behaviors of the users is realized; 3, the method belongs to a completely automatic recommending method, and the users do not need any manual intervention in the information browsing process.

Description

A kind of Tobacco Reference recommend method based on mobile phone users daily record

Technical field

The present invention relates to technical field of information recommendation, particularly relate to the Tobacco Reference recommend method based on mobile phone users daily record.

Background technology

Internet has become the important channel of people's obtaining information.Because the information content on internet is huge, find its interested information fast for ease of user, information recommendation technology is achieving fast development in recent years.Information recommendation technology generally can be divided into the information recommendation based on temperature and the large class of the information recommendation based on user behavior two.Based on the number of times comformed information temperature that the information recommendation of temperature is accessed according to information, by information recommendation high for temperature to user, because the calculating of heatrate and particular user have nothing to do, so these class methods cannot realize personalized recommendation.Information recommendation based on user behavior can realize personalized recommendation, and it can be further divided into again content-based information recommendation and the large class of information recommendation two based on collaborative filtering.Content-based information filtering is according to the similarity determination recommendation information with user browsing information, but these class methods are difficult to the new interest finding user.Information recommendation based on collaborative filtering is pinpointed the problems by introducing the similar users calculating new interest solved in recommendation process, but all there is new customer problem in itself and content-based information filtering, cannot provide effective information recommendation when user just uses a certain website for it.

Along with the development of mobile Internet, increasing user brings into use obtaining information in mobile terminal.The log content of mobile terminal is enriched (behavior etc. that such as user uses mobile terminal), and there is clear and definite corresponding relation between log content and particular user, provides possibility for solving based on the new customer problem in the information recommendation of collaborative filtering.It is one of Core Feature of tobacco website that Tobacco Reference is recommended, and the user accessing this type of website mainly can be divided into smoker, smoker family members and tobacco practitioner three class, need consider the classification of user when carrying out information recommendation.

Summary of the invention

The problem to be solved in the present invention is how by mobile terminal Web log mining (the mobile terminal log recording of user is on Mobile solution middleware server), for user recommends its interested Tobacco Reference.First the present invention determines that mobile phone users is smoker, smoker family members or tobacco practitioner, the user of all mobile terminals is divided into three colonies (smoker colony, smoker family members colony and tobacco practitioner colony); Then analyze the historical viewings information of user in three types of populations, set up the Similarity Model between fellow users; Finally in conjunction with the Tobacco Reference browsing histories of collaborative filtering thought and similarity user, generating recommendations is to the Tobacco Reference of targeted customer.

In order to realize above-mentioned object, present invention employs following technical scheme:

Tobacco Reference recommend method the method based on mobile phone users daily record is divided into be determined class of subscriber, sets up Similarity Model between user, recommend a Tobacco Reference three phases, specifically comprises the following steps:

One, class of subscriber is determined:

Step 1.1, extracts the user type information that all users register when browsing tobacco web first;

Step 1.2, determines the classification of all users according to step 1;

Step 1.3, according to step 2, is divided into smoker colony, smoker family members colony and tobacco practitioner colony three part by all users;

Two, Similarity Model between user is set up:

Step 2.1, extracts the message reference record of all users;

Step 2.2, adds up the number of times of all users to message reference;

Step 2.3, according to step 2.1-2.2, sets up user-user Similarity Model;

Step 2.4, according to step 2.3, sets up user's similarity matrix.

Three, Tobacco Reference is recommended:

Step 3.1, extracts user ID and classification, finds the customer group at user place;

Step 3.2, according to the application that step 3.1 and user are selected, finds the user of the same app of application with its use to gather;

Step 3.3, according to step 3.2 and set up user's similarity matrix, this user set in find top n similarity user;

Step 3.4, carries out collaborative filtering calculating to the historical information of similarity user;

Step 3.5, the score calculated according to step 3.4 sorts, and selects top m information recommendation to user.

As preferably, the historical information Visitor Logs of described step 2.1 user u is expressed as: u=<ID_user, I>, and wherein ID_user is the unique identifying number of user, I is the message reference record that user ID _ user stays, I=(I ₁, I ₂..., I _n),

As preferably, the vector that described step 2.2 couple user sets up is: u=<ID_user, (I ₁, t ₁; I ₂, t ₂; I _n, t _n) >.

As preferably, in the model described in described step 2.3, user u _pwith user u _qbetween the computing formula of similarity degree as (1-3):

$\mathrm{Sim}(u_p,u_q)=\frac{{\mathrm{\&Sigma;}}_{i=1}^n{I}_{\mathrm{pi}}\&times;I_{\mathrm{qi}}}{\left|\right|u_p\left|\right|\&CenterDot;\left|\right|u_q\left|\right|}---\left(1\right)$

$\left|\right|u_p\left|\right|=\sqrt{{\mathrm{\&Sigma;}}_{i=1}^n{I_{\mathrm{pi}}}^2}---\left(2\right)$

$\left|\right|u_q\left|\right|=\sqrt{{\mathrm{\&Sigma;}}_{i=1}^n{I_{\mathrm{qi}}}^2}---\left(3\right)$

In this formula, represent I _piuser p browses the number of times of record i, and n represents that user browses the number of record.

As preferably, described step 2.4 user-user similarity matrix is M _uu, M _uu=[S _ij], S in this matrix _ijrepresent the similarity degree value of i-th user and a jth user.

As preferably, described step 3.3 compositional similarity user gathers U ^'; Step 3.4 is calculated by formula (4-7) for every bar information;

$\mathrm{Score}\left(I_i\right)=\stackrel{\&OverBar;}{R_p}+k{\mathrm{\&Sigma;}}_{u_q\&Element;U^{\&prime;}}\mathrm{Sim}(u_p,u_q)\&CenterDot;(r_{q_i}-\stackrel{\&OverBar;}{R_q})---\left(4\right)$

$k=\frac{1}{\left|U^{\&prime;}\right|}{\mathrm{\&Sigma;}}_{u_q\&Element;U^{\&prime;}}\mathrm{Sim}(u_p,u_q)---\left(5\right)$

$\stackrel{\&OverBar;}{R_p}=\frac{1}{|S\left(R_p\right)|}{\mathrm{\&Sigma;}}_{j\&Element;S\left(R_p\right)}{r}_{p_j}---\left(6\right)$

$\stackrel{\&OverBar;}{R_q}=\frac{1}{|S\left(R_q\right)|}{\mathrm{\&Sigma;}}_{j\&Element;S\left(R_q\right)}{r}_{q_j}---\left(7\right)$

In formula, R _puser u _phistorical information record R _p=<r _p0, r _p1..., r _pn>, r _piuser u _pto the access times of i bar record; S (R _p) be user u _pthe subset of historical information record, user u _pthe mean value of historical information record.

Tobacco user browse history analysis is one of the important method that understanding tobacco user pays close attention to angle, user interest, travel log have recorded the historical information that user browses tobacco in detail, the present invention is directed to the problem that suitable Tobacco Reference can not be recommended suitable user by traditional information recommendation engine, propose the Tobacco Reference recommend method based on mobile phone users daily record, the classification browsing user is determined according to user's registration information, excavate the historical information browsing tobacco of similar users, in conjunction with collaborative filtering thought, final generating recommendations gives the Tobacco Reference being applicable to user.Advantage of the present invention comprises:

1) log-on message of user in Tobacco Reference daily record is analyzed, complete the classification of different user;

2) for user for meeting specific Tobacco Reference demand, in recommendation process, combine user browsed Tobacco Reference in the past, establish the Similarity Model of user-user, achieve the recommend method based on the behavior of user's historical viewings tobacco;

3) a full automatic recommend method, user in browsing information process without the need to any manual intervention.

Accompanying drawing explanation

Fig. 1: a kind of Tobacco Reference recommend method process flow diagram based on mobile phone users daily record.

Fig. 2: determine class of subscriber process flow diagram.

Fig. 3: set up Similarity Model process flow diagram between user.

Fig. 4: recommend Tobacco Reference phase flow figure.

Embodiment

The present invention proposes a kind of Tobacco Reference recommend method based on mobile phone users daily record, process flow diagram as shown in Figure 1.The method is divided into be determined class of subscriber, sets up Similarity Model between user and recommend Tobacco Reference three phases.

The flow process determining the class of subscriber stage as shown in Figure 2, mainly comprises the following steps:

1) user's registration information is extracted.User is when browsing tobacco web first, and system requirements registered user selects type belonging to it (i.e. smoker, smoker family members and tobacco practitioner), extracts the classification that these information determine each user;

2) according to the classification information of the user's registration determined in step 1, all users are divided into smoker colony, smoker family members colony and tobacco practitioner colony;

3) Output rusults.

The flow process setting up the Similarity Model stage between user as shown in Figure 3, mainly comprises the following steps:

1) the message reference record of each user in Mobile solution middleware is extracted.The historical information Visitor Logs of user u is expressed as: u=<ID_user, I>, and wherein ID_user is the unique identifying number of user, and I is the message reference record that user ID _ user stays, I=(I ₁, I ₂..., I _n);

2) counting user is to the number of times of every bar message reference, sets up a vector to each user.According to step 1, to the vector that user sets up be: u=<ID_user, (I ₁, t ₁; I ₂, t ₂; I _n, t _n) >;

3) according to step 1-2, user-user Similarity Model is set up.In the model, user u _pwith user u _qbetween the computing formula of similarity degree as (1-3):

$\mathrm{Sim}(u_p,u_q)=\frac{{\mathrm{\&Sigma;}}_{i=1}^n{I}_{\mathrm{pi}}\&times;I_{\mathrm{qi}}}{\left|\right|u_p\left|\right|\&CenterDot;\left|\right|u_q\left|\right|}---\left(1\right)$

$\left|\right|u_p\left|\right|=\sqrt{{\mathrm{\&Sigma;}}_{i=1}^n{I_{\mathrm{pi}}}^2}---\left(2\right)$

$\left|\right|u_q\left|\right|=\sqrt{{\mathrm{\&Sigma;}}_{i=1}^n{I_{\mathrm{qi}}}^2}---\left(3\right)$

In this formula, represent I _piuser p browses the number of times of record i, and n represents that user browses the number of record;

4) according to step 1-3, user-user similarity matrix M is set up _uu.M _uu=[S _ij], S in this matrix _ijrepresent the similarity degree value of i-th user and a jth user.

User's Tobacco Reference recommends the flow process in stage as shown in Figure 4, mainly comprises the following steps:

1) according to user ID and classification, find the customer group at this user place, namely this user belongs to smoker colony, smoker family members colony or tobacco practitioner colony;

2) according to the application app that user selects, the user of the same application with its use is found to gather;

3) according to the user-user similarity matrix set up, find top n the user similar to this user, compositional similarity user gathers U ^';

4) utilize the historical information of collaborative filtering to similarity user to calculate, every bar information is calculated by formula (4-7), in formula, R _puser u _phistorical information record R _p=<r _p0, r _p1..., r _pn>, r _piuser u _pto the access times of i bar record.S (R _p) be user u _pthe subset of historical information record, user u _pthe mean value of historical information record.Such as, user u _pand u _qhistorical record be respectively R _p=<2,3,2,0,3,1,2,0,1> and R _q=<2,3,0,3,3,4,4,2,0>, then S (R _p)=<2,3,2,3,1,2,1>, S (R _q)=<2,3,3,3,4,4,2>, =2, =3;

$\mathrm{Score}\left(I_i\right)=\stackrel{\&OverBar;}{R_p}+k{\mathrm{\&Sigma;}}_{u_q\&Element;U^{\&prime;}}\mathrm{Sim}(u_p,u_q)\&CenterDot;(r_{q_i}-\stackrel{\&OverBar;}{R_q})---\left(4\right)$

$k=\frac{1}{\left|U^{\&prime;}\right|}{\mathrm{\&Sigma;}}_{u_q\&Element;U^{\&prime;}}\mathrm{Sim}(u_p,u_q)---\left(5\right)$

$\stackrel{\&OverBar;}{R_p}=\frac{1}{|S\left(R_p\right)|}{\mathrm{\&Sigma;}}_{j\&Element;S\left(R_p\right)}{r}_{p_j}---\left(6\right)$

$\stackrel{\&OverBar;}{R_q}=\frac{1}{|S\left(R_q\right)|}{\mathrm{\&Sigma;}}_{j\&Element;S\left(R_q\right)}{r}_{q_j}---\left(7\right)$

5) score calculated according to step 4 sorts, and selects top m information recommendation to user.

Claims (6)

1., based on a Tobacco Reference recommend method for mobile phone users daily record, it is characterized in that the method is divided into and determine class of subscriber, set up Similarity Model between user, recommend Tobacco Reference three phases, specifically comprise the following steps:

One, class of subscriber is determined:

Step 1.1, extracts the user type information that all users register when browsing tobacco web first;

Step 1.2, determines the classification of all users according to step 1;

Step 1.3, according to step 2, is divided into smoker colony, smoker family members colony and tobacco practitioner colony three part by all users;

Two, Similarity Model between user is set up:

Step 2.1, extracts the message reference record of all users;

Step 2.2, adds up the number of times of all users to message reference;

Step 2.3, according to step 2.1-2.2, sets up user-user Similarity Model;

Step 2.4, according to step 2.3, sets up user's similarity matrix.

Three, Tobacco Reference is recommended:

Step 3.1, extracts user ID and classification, finds the customer group at user place;

Step 3.2, according to the application that step 3.1 and user are selected, finds the user of the same app of application with its use to gather;

Step 3.3, according to step 3.2 and set up user's similarity matrix, this user set in find top n similarity user;

Step 3.4, carries out collaborative filtering calculating to the historical information of similarity user;

Step 3.5, the score calculated according to step 3.4 sorts, and selects top m information recommendation to user.

2. a kind of Tobacco Reference recommend method based on mobile phone users daily record according to claim 1, it is characterized in that: the historical information Visitor Logs of step 2.1 user u is expressed as: u=<ID_user, I>, wherein ID_user is the unique identifying number of user, I is the message reference record that user ID _ user stays, I=(I ₁, I ₂..., I _n).

3. a kind of Tobacco Reference recommend method based on mobile phone users daily record according to claim 2, is characterized in that: the vector that step 2.2 couple user sets up is: u=<ID_user, (I ₁, t ₁; I ₂, t ₂; I _n, t _n) >.

4. a kind of Tobacco Reference recommend method based on mobile phone users daily record according to claim 3, is characterized in that: in the model described in step 2.3, user u _pwith user u _qbetween the computing formula of similarity degree as (1-3):

In this formula, represent I _piuser p browses the number of times of record i, and n represents that user browses the number of record.

5. a kind of Tobacco Reference recommend method based on mobile phone users daily record according to claim 4, is characterized in that: step 2.4 user-user similarity matrix is M _uu, M _uu=[S _ij], S in this matrix _ijrepresent the similarity degree value of i-th user and a jth user.

6. a kind of Tobacco Reference recommend method based on mobile phone users daily record according to Claims 1 to 5 any one claim, is characterized in that: step 3.3 compositional similarity user gathers U '; Step 3.4 is calculated by formula (4-7) for every bar information;

In formula, R _puser u _phistorical information record R _p=<r _p0, r _p1..., r _pn>, r _piuser u _pto the access times of i bar record; S (R _p) be user u _pthe subset of historical information record, user u _pthe mean value of historical information record.