CN103778260A - Individualized microblog information recommending system and method - Google Patents

Abstract

The invention relates to an individualized microblog information recommending system and method, belongs to the technical field of social media information service, and solves the problem that user's information acquiring qualities and efficiencies are low and effective information transmission speeds are low in existing microblog platforms. The individualized microblog information recommending system comprises a microblog feature extracting module, a user interest model module, a microblog information recommending module, a microblog information display module and a data module. The individualized microblog information recommending method mainly includes the following several steps of a, acquiring a current real-time microblog listing, user preference and relevant setting information when a user accesses the system; b, extracting statistic and text features of each microblog in the current micoblog listing; c, sequencing microblogs acquired by the user according to a technique of the microblog information recommending module, and preferentially sequencing the microblogs with the highest relevance. The individualized microblog information recommending system and method are applicable to network interactive and share platforms.

Description

A kind of personalized micro-blog information commending system and method

Technical field

The present invention relates to Personalized Information Recommendation System and method, belong to social media information service technology field.

Background technology

In recent years, Social Media is rapid as the emerging mutual and shared platform development in internet.A large number of users carries out the consumption of information with shared by Social Media.For example, social networking website Facebook global registration user exceedes 500,000,000, and the registered user of Sina's microblogging exceedes 200,000,000.

Microblogging platform is one of most typical social media, as: Twitter and Sina's microblogging.Other users that user can pay close attention on microblogging platform simultaneously read the information that other users issue.At microblogging platform, user allows to issue length and is no more than the short text of 140 words.Because microblogging text is short and small, make microblogging release threshold very low with being connected of mobile phone, instant messaging service software, user can upgrade whenever and wherever possible.The packets of information that user issues, containing multiple theme, may comprise Daily hassles, mood record, Environmental Industry Information etc.Existing microblogging service, all information that the crowd who conventionally user is paid close attention to issues return to user according to time-sequencing and browse.Along with increasing of the number of concern, user will receive a large amount of micro-blog informations every day, exceed the limit that user can effectively process, be referred to as the problem of " information overload ", be that a large amount of microbloggings that user browses to not are own that really need, interested, this will seriously expend user's time and energy.Therefore, be extremely necessary to consider user individual factor, automatically identify the interest preference of different user, preferentially present user and want the information content of seeing most, improve quality and the efficiency of user's obtaining information, accelerate the effect spread that has of information.

Summary of the invention

The present invention will solve existing microblogging platform and have the interest preference difference of not considering well micro-blog information audient, cause the quality of user's obtaining information and the problem that efficiency is low, the effective propagation velocity of information is lower, and propose a kind of personalized micro-blog information commending system and method.

The personalized micro-blog information commending system of one in the present invention, comprise: microblogging feature extraction module 4, user interest model module 2, micro-blog information recommending module 5, micro-blog information display module 7 and data module, wherein: data module comprises that micro-blog information module 3, User Information Database 1 and user browse module 6 is set;

Microblogging feature extraction module 4: according to user history information and behavior, microblogging is carried out to semantic analysis and modeling, each piece of microblogging is represented as the form of a term vector, is designated as V _d, wherein each word has a weight, and the computing formula of weight is:

$\mathrm{weight}(w,d)=\mathrm{tf}(w,d)\log \frac{N}{\mathrm{df}\left(w\right)}$

Wherein weight (w, d) weight of word w in expression microblogging d, tf (w, d) represent the frequency that word w occurs in microblogging d, df (w) represents that word w appears in the different microblogging of how many pieces of writing, N is the sum of microblogging, and the value of df (w) and the N all microblogging data based on constantly obtaining is added up;

User interest model module 2: the feature calculate, stored energy embodying user interest preference, and in the given time according to the characteristic information of the behavior renewal relative users of user's access system, user model is a term vector with weight, is designated as V _u, wherein the weight of each word is calculated according to following formula:

$\mathrm{weight}(w,u)=\mathrm{tf}(w,u)\log \frac{M}{\mathrm{uf}\left(u\right)}$

Wherein weight (w, u) represent the weight for word w in user u, tf (w, u) frequency occurring in all microbloggings that expression word w delivered this user, uf (w) represents that word w was once used by how many different users, M is the sum of microblog users, and same, the value of uf (w) and the M all microblogging data based on not gathering is added up;

Micro-blog information recommending module 5: according to current real-time microblogging list and user's characteristic information, based on the output of microblogging feature extraction module and user interest model module, be microblogging model and user interest model, carry out the calculating of the degree of correlation, according to result of calculation by microblogging list ordering, this module can built-in multiple personal the way of recommendation, parallel computation result;

Mode 1: hot language in circle, in the user that user is paid close attention to, there is the microblogging priority ordering of larger transfer amount, each microblogging is endowed a scoring and represents its priority, and computing formula is:

$p\left(d\right|u)=\underset{f\∈F}{\mathrm{\Σ}}p\left(d\right|f)p\left(f\right|u)$

Wherein p (d|u) represents the marking for user u microblogging d, F is user's set that user u pays close attention to, the marking of the user f that p (d|f) expression user u pays close attention to microblogging d, if f delivers, forward or commented on microblogging d, be set to 1, otherwise be set to 0, p (f|u) is that user u calculates according to frequency of interaction between the two the degree of concern of user f, therefore, the meaning of this formula is what more user u paid close attention to, user in close relations is interested in a microblogging, this microblogging should be more relevant to user u, according to this formula, all microblogging candidates are sorted and recommend user,

Mode 2: close friend seeks track, the user that user is paid close attention to sorts according to frequency of interaction, and the microblogging that the user who comes is above delivered preferentially shows;

Mode 3: similar taste, calculate the similarity between user model Vu and microblogging model Vd, account form is as follows:

$\mathrm{Sim}(V_u,V_d)=\frac{V_u\·V_d}{\left|V_u\right|\left|V_d\right|}$

The inner product of two term vectors and they ratio of the product of mould separately, becomes cosine similarity, and hypothesis comprises word important in user model here microblogging and user interest are more pressed close to, and should preferentially be recommended.According to this formula, all microblogging candidates are sorted and recommended;

Micro-blog information display module 7: according to the ranking results calculating in recommending module, micro-blog information is presented to user, in considering user interest preference, for further reducing the time cost of user's obtaining information, this module will provide a kind of presentation mode of novelty: by extremely brief words such as microblogging keyword label, content of text summaries, make user within the extremely short time, understand the theme of current micro-blog information, while user selectively clickthrough checks detailed content;

Data module:

Be used for obtaining and storing of data, comprise:

Micro-blog information module 3: obtain in real time, store all micro-blog informations that user can receive;

User Information Database 1: the aspect such as text, the social activity information of storing statically microblog users;

User browses module 6 is set: the mode of independently selective reception microblogging personalized recommendation service of user;

A kind of personalized micro-blog information recommend method, comprises the following steps:

When a, user login commending system, obtain all micro-blog informations that its follower issues in the recent period, user preference information is obtained in microblogging list in real time simultaneously;

B, extract the feature of every microblogging in real-time microblogging list;

In c, compute user preferences information and microblogging list in real time, the degree of correlation of each microblogging, is that the most possible interested microblogging of user is preferentially shown by the microblogging large degree of correlation;

When d, user operate microblogging, its operation behavior of server record;

After e, user log off, in the given time, upgrade the preference of user in text, social activity, the micro-blog information of next time logining for it is recommended.

The present invention has taken into full account user individual factor, automatically identifies the interest preference of different user, preferentially presents user and wants the information content of seeing most, improves quality and the efficiency of user's obtaining information, accelerates the effect spread that has of information.Compared with prior art, the advantage of a kind of personalized micro-blog information recommend method in the present invention is: automatically identify the preference information of microblog users in text, social activity, realize personalized micro-blog information service, user priority is browsed to pay close attention to most, most interested microblogging content; Meanwhile, novel presentation mode, can further reduce the time cost of user's obtaining information.The present invention is applicable to internet alternately and shared platform.

Accompanying drawing explanation

Fig. 1 is the structured flowchart of the personalized micro-blog information commending system of the embodiment of the present invention;

Fig. 2 is the process flow diagram of the personalized micro-blog information recommend method of the embodiment of the present invention.

Embodiment

For above-mentioned purpose of the present invention, feature and advantage can be become apparent more, below in conjunction with the drawings and specific embodiments, the present invention is further detailed explanation.Figure 1 shows that the personalized micro-blog information commending system structured flowchart of the embodiment of the present invention.

A kind of personalized micro-blog information commending system in embodiment one, the present invention. comprise with lower module: microblogging feature extraction module 4, user interest model module 2, micro-blog information recommending module 5, micro-blog information display module 7 and data module, wherein: data module comprises that micro-blog information module 3, User Information Database 1 and user browse module 6 is set;

Microblogging feature extraction module 4: according to user history information and behavior, microblogging is carried out to semantic analysis and modeling, each piece of microblogging is represented as the form of a term vector, is designated as V _d, wherein each word has a weight, and the computing formula of weight is:

$\mathrm{weight}(w,d)=\mathrm{tf}(w,d)\log \frac{N}{\mathrm{df}\left(w\right)}$

Wherein weight (w, d) weight of word w in expression microblogging d, tf (w, d) represent the frequency that word w occurs in microblogging d, df (w) represents that word w appears in the different microblogging of how many pieces of writing, N is the sum of microblogging, and the value of df (w) and the N all microblogging data based on constantly obtaining is added up;

User interest model module 2: the feature calculate, stored energy embodying user interest preference, and in the given time according to the characteristic information of the behavior renewal relative users of user's access system, user model is a term vector with weight, is designated as V _u, wherein the weight of each word is calculated according to following formula:

$\mathrm{weight}(w,u)=\mathrm{tf}(w,u)\log \frac{M}{\mathrm{uf}\left(u\right)}$

Wherein weight (w, u) represent the weight for word w in user u, tf (w, u) frequency occurring in all microbloggings that expression word w delivered this user, uf (w) represents that word w was once used by how many different users, M is the sum of microblog users, and same, the value of uf (w) and the M all microblogging data based on not gathering is added up;

Micro-blog information recommending module 5: according to current real-time microblogging list and user's characteristic information, based on the output of microblogging feature extraction module and user interest model module, be microblogging model and user interest model, carry out the calculating of the degree of correlation, according to result of calculation by microblogging list ordering, this module can built-in multiple personal the way of recommendation, parallel computation result;

Mode 1: hot language in circle, in the user that user is paid close attention to, there is the microblogging priority ordering of larger transfer amount, each microblogging is endowed a scoring and represents its priority, and computing formula is:

$p\left(d\right|u)=\underset{f\∈F}{\mathrm{\Σ}}p\left(d\right|f)p\left(f\right|u)$

Wherein p (d|u) represents the marking for user u microblogging d, F is user's set that user u pays close attention to, the marking of the user f that p (d|f) expression user u pays close attention to microblogging d, if f delivers, forward or commented on microblogging d, be set to 1, otherwise be set to 0, p (f|u) is that user u calculates according to frequency of interaction between the two the degree of concern of user f, therefore, the meaning of this formula is what more user u paid close attention to, user in close relations is interested in a microblogging, this microblogging should be more relevant to user u, according to this formula, all microblogging candidates are sorted and recommend user,

Mode 2: close friend seeks track, the user that user is paid close attention to sorts according to frequency of interaction, and the microblogging that the user who comes is above delivered preferentially shows;

Mode 3: similar taste, calculate the similarity between user model Vu and microblogging model Vd, account form is as follows:

$\mathrm{Sim}(V_u,V_d)=\frac{V_u\·V_d}{\left|V_u\right|\left|V_d\right|}$

The inner product of two term vectors and they ratio of the product of mould separately, becomes cosine similarity, and hypothesis comprises word important in user model here microblogging and user interest are more pressed close to, and should preferentially be recommended.According to formula

all microblogging candidates are sorted and recommended;

Micro-blog information display module 7: according to the ranking results calculating in recommending module, micro-blog information is presented to user, in considering user interest preference, for further reducing the time cost of user's obtaining information, this module will provide a kind of presentation mode of novelty: by extremely brief words such as microblogging keyword label, content of text summaries, make user within the extremely short time, understand the theme of current micro-blog information, while user selectively clickthrough checks detailed content;

Data module:

Be used for obtaining and storing of data, comprise:

Micro-blog information module 3: obtain in real time, store all micro-blog informations that user can receive;

User Information Database 1: the aspect such as text, the social activity information of storing statically microblog users;

User browses module 6 is set: the mode of independently selective reception microblogging personalized recommendation service of user;

Embodiment two, present embodiment are to the further illustrating of User Information Database 1 in embodiment one, User Information Database 1, static attribute when this module stores user Accreditation System; Meanwhile, this module is also stored user and is used text that microblogging service accumulates afterwards and the information of two dimensions of social networks; In addition, this module has also been responsible for the network behavior operation since storage user uses commending system.

Embodiment three, present embodiment are that user in embodiment one is browsed further illustrating of module 6 is set, user can independently select proposed algorithm, thereby the microblogging content that browses to different sequences, in the time that user uses commending system first, system can arrange a default option.

In embodiment four, the present invention, a kind of personalized micro-blog information recommend method comprises the steps:

Step 101, when user logins commending system, obtains all micro-blog informations that its follower issues in the recent period, and user preference information is obtained in microblogging list in real time simultaneously.Wherein, comprise the preference in user version and social activity.Can, by the text analyzing to the passing microblogging of delivering, forward, collecting of user and the comment delivered, obtain the keyword vector with weight; By the interaction between bean vermicelli and user, user and concern, obtain crowd that user pays close attention to most, pay close attention to user's crowd and the people group-list the closest with user interaction most.

Step 102, extracts the feature of every microblogging in real-time microblogging list.Microblogging text is analyzed, calculated the text vector that represents this microblogging, extract this microblogging author and authorship's information simultaneously.

Step 103, in compute user preferences information and microblogging list in real time, the degree of correlation of each microblogging, is that the most possible interested microblogging of user is preferentially shown by the microblogging large degree of correlation.Can calculate the similarity between user version preference and microblogging text, and check that the author of this microblogging or the author of original microblogging are whether in the preference people's group-list on user social contact, if,, by corresponding this author weight and text similarity linear combination, draw final similarity.

Step 104, when user operates microblogging, its operation behavior of server record.User's interest may produce along with the time cheaply.And the click of browsing, linking of the click of user to microblogging list, picture, author's page the preference that can embody to a certain extent user such as browse.Therefore, recorded, for calculating more accurately user's interest model.

Step 105, after user log off, in the given time, upgrades the preference of user in text, social activity, and the micro-blog information of next time logining for it is recommended.

Claims (4)

1. a personalized micro-blog information commending system, it is characterized in that it comprises: microblogging feature extraction module (4), user interest model module (2), micro-blog information recommending module (5), micro-blog information display module (7) and data module, wherein: data module comprises that micro-blog information module (3), User Information Database (1) and user browse module (6) is set;

Microblogging feature extraction module (4): according to user history information and behavior, microblogging is carried out to semantic analysis and modeling, each piece of microblogging is represented as the form of a term vector, is designated as V _d, wherein each word has a weight, and the computing formula of weight is:

$\mathrm{weight}(w,d)=\mathrm{tf}(w,d)\log \frac{N}{\mathrm{df}\left(w\right)}$

Wherein, wherein weight (w, d) weight of word w in expression microblogging d, tf (w, d) represent the frequency that word w occurs in microblogging d, df (w) represents that word w appears in the different microblogging of how many pieces of writing, the sum that N is microblogging, and the value of df (w) and the N all microblogging data based on constantly crawling is added up;

User interest model module (2): the feature calculate, stored energy embodying user interest preference, and in the given time according to the characteristic information of the behavior renewal relative users of user's access system, user model is a term vector with weight, is designated as V _u, wherein the weight of each word is calculated according to following formula:

$\mathrm{weight}(w,u)=\mathrm{tf}(w,u)\log \frac{M}{\mathrm{uf}\left(u\right)}$

Wherein, wherein weight (w, u) represent the weight for word w in user u, tf (w, u) frequency occurring in all microbloggings that expression word w delivered this user, uf (w) represents that word w was once used by how many different users, the sum that M is microblog users, equally, the value of uf (w) and M all the microblogging data based on not gathering add up;

Micro-blog information recommending module (5): according to current real-time microblogging list and user's characteristic information, based on the output of microblogging feature extraction module and user interest model module, be microblogging model and user interest model, carry out the calculating of the degree of correlation, according to result of calculation by microblogging list ordering, the way of recommendation of the built-in multiple personal of this module, parallel computation result;

The way of recommendation of described multiple personal is respectively:

Mode 1: hot language in circle, in the user that user is paid close attention to, there is the microblogging priority ordering of larger transfer amount, each microblogging is endowed a scoring and represents its priority, and computing formula is:

$p\left(d\right|u)=\underset{f\∈F}{\mathrm{\Σ}}p\left(d\right|f)p\left(f\right|u)$

Wherein p (d|u) represents the marking for user u microblogging d, F is user's set that user u pays close attention to, the marking of the user f that p (d|f) expression user u pays close attention to microblogging d, if f delivers, forwards or commented on microblogging d, be set to 1, be that user u calculates according to frequency of interaction between the two the degree of concern of user f otherwise be set to 0, p (f|u), according to this formula, all microblogging candidates sorted from high to low and recommend user;

Mode 2: close friend seeks track, the user that user is paid close attention to sorts according to frequency of interaction, and the microblogging that the user who comes is above delivered preferentially shows;

Mode 3: similar taste, calculates user model V _uwith microblogging model V _dbetween similarity, account form is as follows:

$\mathrm{Sim}(V_u,V_d)=\frac{V_u\·V_d}{\left|V_u\right|\left|V_d\right|}$

The inner product of two term vectors and they ratio of the product of mould separately, becomes cosine similarity, according to formula

all microblogging candidates are sorted and recommended;

Micro-blog information display module (7): according to the ranking results calculating in recommending module, micro-blog information is presented to user;

Data module:

Be used for obtaining and storing of data, comprise:

Micro-blog information module (3): obtain in real time, store all micro-blog informations that user can receive;

User Information Database (1): store statically the text of microblog users, social aspect information;

User browses module (6) is set: the mode of independently selective reception microblogging personalized recommendation service of user.

2. the personalized micro-blog information commending system of one as claimed in claim 1, is characterized in that User Information Database (1), static attribute when this module stores user Accreditation System; Meanwhile, this module is also stored user and is used text that microblogging service accumulates afterwards and the information of two dimensions of social networks; In addition, this module has also been responsible for the network behavior operation since storage user uses commending system.

3. the personalized micro-blog information commending system of one as claimed in claim 1, it is characterized in that utilizing user to browse module (6) is set, user can independently select proposed algorithm, thereby browse to the microblogging content of different sequences, in the time that user uses commending system first, system can arrange a default option.

4. a personalized micro-blog information recommend method, is characterized in that it comprises the following steps:

When a, user login commending system, obtain all micro-blog informations that its follower issues in the recent period, user preference information is obtained in microblogging list in real time simultaneously;

B, extract the feature of every microblogging in real-time microblogging list;

In c, compute user preferences information and microblogging list in real time, the degree of correlation of each microblogging, is that the most possible interested microblogging of user is preferentially shown by the microblogging large degree of correlation;

When d, user operate microblogging, its operation behavior of server record;

After e, user log off, in the given time, upgrade the preference of user in text, social activity, the micro-blog information of next time logining for it is recommended.

Images