CN109767301A - Recommended method and system, computer installation, computer readable storage medium - Google Patents
Recommended method and system, computer installation, computer readable storage medium Download PDFInfo
- Publication number
- CN109767301A CN109767301A CN201910032524.1A CN201910032524A CN109767301A CN 109767301 A CN109767301 A CN 109767301A CN 201910032524 A CN201910032524 A CN 201910032524A CN 109767301 A CN109767301 A CN 109767301A
- Authority
- CN
- China
- Prior art keywords
- interest
- social networks
- target user
- user
- article
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention relates to a kind of recommended method and systems, wherein recommended method is the following steps are included: the article set that building target user is consumed corresponds to the social networks of target user；The dynamic personal interest model of target user is established according to article set；The short-term interest model of social networks is constructed according to article set；Construct the Long-term Interest model of social networks；Spliced according to short-term interest model and Long-term Interest model；The node for calculating target user indicates to indicate with the node of friend in social networks；Merge feature weight about the weight computing of target user according to friend in social networks；Nonlinear transformation is carried out to feature weight is merged；It is calculated according to dynamic personal interest model；Obtain recommending the probability of article according to the final interest of user；According to the probability calculation log-likelihood function value for recommending article；In the inventive solutions, it can consider the social networks of user and the dynamic hobby factor of user, simultaneously to promote the accuracy recommended.
Description
Technical field
The present invention relates to information recommendation field more particularly to a kind of recommended method, a kind of recommender system, a kind of computer dresses
It sets and a kind of computer readable storage medium.
Background technique
Hidasi et al., which proposes one, carries out the technology of dialogue-based recommendation using LSTM.This method mainly utilizes
LSTM mentioned above models according to the history consumer record of user the expression of each goods for consumption.It is next in order to recommend
A article, they indicate the current interest of user using the last one article of customer consumption, we are according to this interest
Mathematical notation, calculates the similarity degree of current interest and all items, it is final we to user recommend one with current interest most
Similar article.The technology is a mature technology, but it there are many problems.It does not model the long-term of user first
Interest only models his recent interest with the recent consumer record of user.Another question is that it does not account for user
Social influence, this makes the prediction of model have very big deviation.Chaney et al. proposes a social activity Bai Song decomposition model.
The recommendation influence degree that they have modeled a kind of trust-factor to portray friend to some user, when modeling user interest
The influence of friend is considered, but this model has proposed for a long time, and does not account for the serializing of consumption history record
Feature, therefore also suffer from certain drawbacks.
The prior art models the dynamic interest of user or divides the social influence in recommender system
Analysis, but as far as we know, still above-mentioned two factor is combined without a kind of technology.A nearest research is passed about using
Return the user behavior of neuron network simulation session level, but does not account for social influence.Other work sutdies social influence, example
Such as, Ma et al. has inquired into systematic influence of the social networks to recommendation of friend.But the influence from different user is all static
, they will not change according to the current interest of recommended user.
Summary of the invention
The present invention is directed to solve at least one of the technical problems existing in the prior art or related technologies.
For this purpose, it is an object of the present invention to provide a kind of dialogue-based and social influence recommended method, it can
The social networks of user and the dynamic hobby factor of user are considered simultaneously, to promote the accuracy recommended；And according to user
The hobby of oneself, dynamic infer the higher friend of influence power in social networks, make friend similar in current interest to recommendation
As a result it influences bigger.
It is another object of the present invention to provide a kind of dialogue-based and social influence recommender systems, can be in life
When at recommendation results, user's own interests and its friend's interest are comprehensively considered, joined and mould is inferred to the dynamic of friend's influence power
Block, to preferably grab the recent hobby of associated friends, the article for being allowed to serve target user is recommended.
It is yet a further object of the present invention to provide a kind of computer installations.
Yet another object of the invention is that providing a kind of computer readable storage medium.
To achieve the above object, the technical solution of first aspect present invention provides a kind of dialogue-based and social influence
Recommended method, comprising the following steps:
The article set being consumed in building target user's current sessionsThe social networks G of corresponding target user, enables:
G=(U, E)；
Wherein, u indicates that target user, U indicate the set of the friend of target user u in social networks, and E indicates to use with target
The social networks of family u, i expression are consumed commodity；
According to article setThe dynamic personal interest model of target user is established, is enabled:
Wherein, hnIndicate the newest interest of target user, hn-1Indicate the previous interest of newest interest, f expression will be newest
It is consumed nonlinear function of the commodity in conjunction with previous interest；
According to article setConstruct the short-term interest model of social networks GIt enables:
Wherein,Indicate the newest interest of friend k in social networks G,Indicate the previous interest of friend k；
Construct the Long-term Interest model of social networks GIt enables:
Wherein, in social networks G friend k Long-term Interest modelIt is that user is embedded in representing matrix WuRow k；
According to short-term interest modelWith Long-term Interest modelSpliced, obtains split-join model sk, enable
Wherein, ReLU (x)=max (0, x) is a nonlinear activation primitive, W1It is transformation matrix；
The node for calculating target user indicatesIt is indicated with the node of friend k in social networks GIt enables:
Wherein,Expression for target user u at l layers,To calculate the similar of two nodes
The function of degree,For weight of the user k about target user u in social networks G；
Merge feature weight about the weight computing of target user u according to friend k in social networks G, enable:
Wherein,It is fusion of the interest of the social networks G of target user u on l layer；
Nonlinear transformation is carried out to feature weight is merged, is obtained:
Wherein, W(l)It is l layers of a weight matrix that is shared and can learning；
The final interest of user is calculated according to dynamic personal interest model, enables:
Wherein, W2For the matrix of a linear transformation,For the final interest of target user；
The probability for recommending article to be y is obtained according to the final interest of user, it may be assumed that
Wherein, N (u) is the number of user in social networks G, zyIt is indicated for the insertion of article y, | I | it is the number of article；
It is the log-likelihood function value of the probability calculation article of y according to recommendation article:
In the technical scheme, this method is intended to model the dynamic interest and real-time social influence of user simultaneously.It is specific and
Speech, we are modeled using the nerual network technique session current to user, extract the interest preference of user, and count in real time
The influence that the friend of user generates it under current scene is calculated, influences to carry out article recommendation in conjunction with itself hobby and friend.
In the above-mentioned technical solutions, it is preferable that by the newest nonlinear function f for being consumed commodity in conjunction with previous interest
Are as follows:
Wherein, σ is sigmoid function, σ (x)=(1+exp (- x))-1。
The technical solution of second aspect of the present invention provides a kind of dialogue-based and social influence recommender system, comprising:
Module is constructed, the article set being consumed in building target user's current sessions is arranged to be used forCorresponding mesh
The social networks G of user is marked, is enabled:
G=(U, E)；
Wherein, u indicates that target user, U indicate the set of the friend of target user u in social networks, and E indicates to use with target
The social networks of family u, i expression are consumed commodity；
Dynamic personal interest model building module, is arranged to be used for according to article setEstablish the dynamic of target user
State personal interest model enables:
Wherein, hnIndicate the newest interest of target user, hn-1Indicate the previous interest of newest interest, f expression will be newest
It is consumed nonlinear function of the commodity in conjunction with previous interest；
Short-term interest model construction module is arranged to be used for according to article setConstruct the short-term emerging of social networks G
Interesting modelIt enables:
Wherein,Indicate the newest interest of friend k in social networks G,Indicate the previous interest of friend k；
Long-term Interest model construction module is arranged to be used for the Long-term Interest model of building social networks GIt enables:
Wherein, in social networks G friend k Long-term Interest modelIt is that user is embedded in representing matrix WuRow k；
Splicing module is arranged to be used for according to short-term interest modelWith Long-term Interest modelSpliced, is obtained
Split-join model sk, enable
Wherein, ReLU (x)=max (0, x) is a nonlinear activation primitive, W1It is transformation matrix；
Computing module, the node for being arranged to be used for calculating target user indicateWith the section of friend k in social networks G
Point indicatesIt enables:
Wherein,Expression for target user u at l layers,To calculate the similar of two nodes
The function of degree,For weight of the user k about target user u in social networks G；
Merge feature weight computing module, is arranged to be used for according to friend k in social networks G about target user u's
Weight computing merges feature weight, enables:
Wherein,It is fusion of the interest of the social networks G of target user u on l layer；
Nonlinear transformation module is arranged to be used for obtaining to feature weight progress nonlinear transformation is merged:
Wherein, W(l)It is l layers of a weight matrix that is shared and can learning；
Final interest computing module, is arranged to be used for that the final emerging of user is calculated according to dynamic personal interest model
Interest enables:
Wherein, W2For the matrix of a linear transformation,For the final interest of target user；
Probability evaluation entity is arranged to be used for the probability for obtaining that article is recommended to be y according to the final interest of user, it may be assumed that
Wherein, N (u) is the number of user in social networks G, zyIt is indicated for the insertion of article y, | I | it is the number of article；
Log-likelihood function value computing module is arranged to be used for according to article is recommended being pair of the probability calculation article of y
Number likelihood function value:
In the technical scheme, in recommender system, while considering the social networks of user and the dynamic interest love of user
Good factor, to promote the accuracy recommended；And the hobby according to user oneself, dynamic infer social networks in influence power compared with
High friend influences friend similar in current interest on recommendation results bigger.In order to be lifted at the standard of line platform recommender system
True property, the present invention in propose to model the dynamic hobby and dynamic social networks of user.When generating recommendation results,
Comprehensively consider user's own interests and its friend's interest, joined the dynamic inference module to friend's influence power, thus preferably
The recent hobby for grabbing associated friends, the article for being allowed to serve target user are recommended.
In the above-mentioned technical solutions, it is preferable that by the newest nonlinear function f for being consumed commodity in conjunction with previous interest
Are as follows:
Wherein, σ is sigmoid function, σ (x)=(1+exp (- x))-1。
The technical solution of third aspect present invention provides a kind of computer installation, including processor, and processor is for holding
The recommendation for any one that the technical solution of the first aspect of the present invention proposes is realized when the computer program stored in line storage
The step of method.
The technical solution of fourth aspect present invention provides a kind of computer readable storage medium, is stored thereon with computer
Program (instruction), computer program (instruction) realize what the technical solution of the first aspect of the present invention proposed when being executed by processor
The step of recommended method of any one.
Detailed description of the invention
Above-mentioned and/or additional aspect of the invention and advantage will become from the description of the embodiment in conjunction with the following figures
Obviously and it is readily appreciated that, in which:
Fig. 1 shows the flow diagram of recommended method involved in one embodiment of the invention；
Fig. 2 shows the flow diagrams of recommender system involved in another embodiment of the present invention；
Fig. 3 shows Dynamic Graph attention model figure involved in the embodiment of the present invention；
Fig. 4 shows Dynamic Graph attention network involved in the embodiment of the present invention.
Specific embodiment
To better understand the objects, features and advantages of the present invention, with reference to the accompanying drawing and specific real
Applying mode, the present invention is further described in detail.It should be noted that in the absence of conflict, the implementation of the application
Feature in example and embodiment can be combined with each other.
In the following description, numerous specific details are set forth in order to facilitate a full understanding of the present invention, still, the present invention may be used also
To be implemented using other than the one described here other modes, therefore, protection scope of the present invention is not limited to following public affairs
The limitation for the specific embodiment opened.
Referring to Fig. 1 to Fig. 4 description recommended method according to some embodiments of the invention and system, computer installation,
Computer readable storage medium.
In order to the user of on-line communities provide it is effective suggest, it is proposed that simultaneously to the dynamic interest of user and
Social influence dependent on situation is modeled.Final problem is defined as by we:
Defining (session level social recommendation) enables U indicate user's set, and I indicates article set, and G=(U, E) is social networks,
Wherein E is the social networks between user.For user u, a new session is givenThe target of session level social recommendation be simultaneously using the dynamic interest of user u (come
FromInformation) and social influence (come fromInformation, wherein N (u) is friend's set of user u) recommend
The a subset of I, the article in subset be user u the (n+1)th step may interested article.
For this purpose, the present invention proposes a kind of novel Dynamic Graph attention model Dynamic Graph
Recommendation (DGRec), as shown in figure 3, it can model preference and the friends of user of user itself simultaneously
Preference.
DGRec is made of four modules.First module is a Recognition with Recurrent Neural Network (RNN), it can be modeled in use
The article sequence being consumed in the current sessions of family.The interest of the friend of user is by the combination of their short-term preferences and long-term preference
Come what is modeled.Short-term preference, such as the article in its nearest session, are also encoded using RNN.The long-term preference of friend is to pass through
The personal insertion for learning to obtain indicates to encode.Then, model uses expression of the figure attention network by active user and its
The expression of friend combines.This is our model and the key component of contribution: it is proposed that mechanism association worked as according to user
Preceding interest measures the influence of each friend.In the final step, the model by will the current preference of user and he suffered by
To (dependent on situation) social influence combine and generate recommendation results.
As shown in Figure 1, the recommended method of the dialogue-based and social influence according to one embodiment of the invention, including it is following
Step:
S100 constructs the article set being consumed in target user's current sessionsThe social networks of corresponding target user
G is enabled:
G=(U, E)；
Wherein, u indicates that target user, U indicate the set of the friend of target user u in social networks, and E indicates to use with target
The social networks of family u, i expression are consumed commodity；
In order to capture the fast-changing interest of user, we model (target) user using RNN in current sessions
Movement (is clicked).RNN is the Series Modeling tool of standard and is used to modeling user (sequence) preference data recently.
RNN can infer user conversation with inputting an input with oneExpression.It can be with
The expression of the expression and newest input that input before recursively will be all combines, it may be assumed that S200, according to article set
The dynamic personal interest model of target user is established, is enabled
Wherein, hnIndicate the newest interest of target user, hn-1Indicate the previous interest of newest interest, f expression will be newest
It is consumed nonlinear function of the commodity in conjunction with previous interest；
It is believed that user may be subjected to the influence of the nearest interest of friends in social networks G.For this reason,
We model the short-term and Long-term Interest of friend in different ways.
Article sequence (for example, newest on-line session of friend) that we are consumed recently using friend models his short-term
Interest.Long-term Interest represents the total interest of a friend, and is modeled using personal insertion expression.
The session current to a target userThe short-term interest of his friend is respective in session T+1 with them
That session before indicates.The movement of each friend kIt is modeled with RNN.In fact,
Here we reuse the RNN of modeling target user's session to model the session of friend.In other words, two RNN are shared same
The parameter of sample.We indicate the short-term preference of friend k, i.e. S300, according to article set with RNN final outputBuilding
The short-term interest model of social networks GIt enables:
Wherein,Indicate the newest interest of friend k in social networks G,Indicate the previous interest of friend k；
The long-term preference of friend reflects their total interest.Since long-term preference is not time-sensitive, we make
Them are indicated with a vector.
S400 constructs the Long-term Interest model of social networks GIt enables:
Wherein, in social networks G friend k Long-term Interest modelIt is that user is embedded in representing matrix WuRow k；
S500, according to short-term interest modelWith Long-term Interest modelSpliced, obtains split-join model sk, enable
Wherein, ReLU (x)=max (0, x) is a nonlinear activation primitive, W1It is transformation matrix；
We obtain the mixing of the interest of target user and the interest of his friends using novel figure attention network
It indicates.Firstly, we encode friendship network in figure, interior joint corresponds to user (i.e. target user and its friend),
Side indicates friends.In addition, each node uses the expression of its corresponding user as (dynamic) feature.Secondly, using disappearing
It ceases pass-algorithm and propagates these features along side.It is to measure edge using attention mechanism in place of the main novel of our method
The weight for the feature that each side is propagated.Weight corresponds to friend to the influence degree of target user.By the message of fixed number of times
After transmitting iteration, the result feature at target user's node is exactly the expression after combination.
For each user, we construct one using the user and its friend as the figure of node.If user u has | N
(u) | a friend, then the figure just has | and N (u) |+1 node.The initial representation h of user unBy the initial characteristics as node u(each user u consumed one it is newIn article after, this feature will be updated).It is corresponding for friend k
Node diagnostic be arranged to SkAnd it is remained unchanged during timestamp is T+1.For formalization, the character representation of node
It is
We have proposed a novel Dynamic Graph attention networks, and context-sensitive social shadow is modeled with it
It rings, and guides the propagation of influence using attention mechanism.Whole process is expounded in Fig. 4.
S600, the node for calculating target user indicateIt is indicated with the node of friend k in social networks GIt enables:
Wherein,Expression for target user u at l layers,To calculate the similar of two nodes
The function of degree,For user k in social networks G about target user u weight (with existingFor background) or shadow
Loud rank；
S700 merges feature weight about the weight computing of target user u according to friend k in social networks G, enables:
Wherein,It is fusion of the interest of the social networks G of target user u on l layer；S800, to merging feature weight
Nonlinear transformation is carried out, is obtained:
Wherein, W(l)It is l layers of a weight matrix that is shared and can learning, wherein each layer represents picture scroll product
Convolution operation of network, the value obtain the final expression of each node after beginning to pass through l layers of convolution by first layer.Merge
The expression of (social influence) we useTo indicate.
As soon as because the interest of user is codetermined by both his nearest behavior and social influence, his final expression
Merging the two by full articulamentum to obtain, i.e. the final interest of user is calculated according to dynamic personal interest model by S900, it enables:
Wherein, W2For the matrix of a linear transformation,For the final interest of target user；
S1000 obtains the probability for recommending article to be y according to the final interest of user, it may be assumed that
Wherein, N (u) is the number of user in social networks G, zyIt is indicated for the insertion of article y, | I | it is the number of article；
S1100 is the log-likelihood function value of the probability calculation article of y according to recommendation article:The function is optimized with gradient descent method.
In this embodiment, this method is intended to model the dynamic interest and real-time social influence of user simultaneously.Specifically,
We are modeled using the nerual network technique session current to user, extract the interest preference of user, and calculate in real time
The influence that the friend of user generates it under current scene influences to carry out article recommendation in conjunction with itself hobby and friend.
As shown in Fig. 2, the recommender system 1000 of the dialogue-based and social influence according to another embodiment of the present invention, packet
It includes:
Module 10 is constructed, the article set being consumed in building target user's current sessions is arranged to be used forIt is corresponding
The social networks G of target user is enabled:
G=(U, E)；
Wherein, u indicates that target user, U indicate the set of the friend of target user u in social networks, and E indicates to use with target
The social networks of family u, i expression are consumed commodity；
Dynamic personal interest model building module 20, is arranged to be used for according to article setEstablish target user's
Dynamic personal interest model enables:
Wherein, hnIndicate the newest interest of target user, hn-1Indicate the previous interest of newest interest, f expression will be newest
It is consumed nonlinear function of the commodity in conjunction with previous interest；
Short-term interest model construction module 30 is arranged to be used for according to article setConstruct the short-term of social networks G
Interest modelIt enables:
Wherein,Indicate the newest interest of friend k in social networks G,Indicate the previous interest of friend k；
Long-term Interest model construction module 40 is arranged to be used for the Long-term Interest model of building social networks GIt enables:
Wherein, in social networks G friend k Long-term Interest modelIt is that user is embedded in representing matrix WuRow k；
Splicing module 50 is arranged to be used for according to short-term interest modelWith Long-term Interest modelSpliced, is obtained
To split-join model sk, enable
Wherein, ReLU (x)=max (0, x) is a nonlinear activation primitive, W1It is transformation matrix；
Computing module 60, the node for being arranged to be used for calculating target user indicateWith friend k in social networks G
Node indicatesIt enables:
Wherein,Expression for target user u at l layers,To calculate the similar of two nodes
The function of degree,For weight of the user k about target user u in social networks G；
Merge feature weight computing module 70, is arranged to be used for according to friend k in social networks G about target user u
Weight computing merge feature weight, enable:
Wherein,It is fusion of the interest of the social networks G of target user u on l layer；
Nonlinear transformation module 80 is arranged to be used for obtaining to feature weight progress nonlinear transformation is merged:
Wherein, W(l)It is l layers of a weight matrix that is shared and can learning；
Final interest computing module 90, is arranged to be used for that the final of user is calculated according to dynamic personal interest model
Interest enables:
Wherein, W2For the matrix of a linear transformation,For the final interest of target user；
Probability evaluation entity 100 is arranged to be used for the probability for obtaining that article is recommended to be y according to the final interest of user,
That is:
Wherein, N (u) is the number of user in social networks G, zyIt is indicated for the insertion of article y, | I | it is the number of article；
Log-likelihood function value computing module 110 is arranged to be used for according to article is recommended being the probability calculation article of y
Log-likelihood function value:
In this embodiment, in recommender system, while considering the social networks of user and the dynamic hobby of user
Factor, to promote the accuracy recommended；And the hobby according to user oneself, dynamic infer that influence power is higher in social networks
Friend, make friend similar in current interest on recommendation results influence it is bigger.In order to be lifted at the accurate of line platform recommender system
Property, the present invention in propose to model the dynamic hobby and dynamic social networks of user.It is comprehensive when generating recommendation results
It closes and considers user's own interests and its friend's interest, joined the dynamic inference module to friend's influence power, to preferably grab
The recent hobby for taking associated friends, the article for being allowed to serve target user are recommended.
In any of the above-described embodiment, it is preferable that by the newest non-linear letter for being consumed commodity in conjunction with previous interest
Number f are as follows:
Wherein, σ is sigmoid function, σ (x)=(1+exp (- x))-1。
The computer installation of further embodiment of the present invention, including processor, processor are stored for executing in memory
Computer program when realize the first aspect of the present invention technical solution propose any one recommended method the step of.
The computer readable storage medium of another embodiment of the invention, is stored thereon with computer program (instruction), counts
Calculation machine program (instruction) realizes the recommendation for any one that the technical solution of the first aspect of the present invention proposes when being executed by processor
The step of method.
In the present invention, the terms such as term " installation ", " connected ", " connection ", " fixation " shall be understood in a broad sense, for example,
" connection " may be fixed connection or may be dismantle connection, or integral connection；" connected " can be directly connected,
It can be indirectly connected through an intermediary.For the ordinary skill in the art, on can understanding as the case may be
State the concrete meaning of term in the present invention.
In description of the invention, it is to be understood that the instructions such as term " on ", "lower", "left", "right", "front", "rear"
Orientation or positional relationship is to be based on the orientation or positional relationship shown in the drawings, and is merely for convenience of the description present invention and simplification is retouched
It states, rather than the device or unit of indication or suggestion meaning must have specific direction, be constructed and operated in a specific orientation,
It is thus impossible to be interpreted as limitation of the present invention.
In the description of this specification, the description of term " one embodiment ", " some embodiments ", " specific embodiment " etc.
Mean that particular features, structures, materials, or characteristics described in conjunction with this embodiment or example are contained at least one reality of the invention
It applies in example or example.In the present specification, schematic expression of the above terms are not necessarily referring to identical embodiment or reality
Example.Moreover, description particular features, structures, materials, or characteristics can in any one or more of the embodiments or examples with
Suitable mode combines.
The foregoing is only a preferred embodiment of the present invention, is not intended to restrict the invention, for the skill of this field
For art personnel, the invention may be variously modified and varied.All within the spirits and principles of the present invention, made any to repair
Change, equivalent replacement, improvement etc., should all be included in the protection scope of the present invention.
Claims (6)
1. a kind of dialogue-based and social influence recommended method, which comprises the following steps:
The article set being consumed in building target user's current sessionsThe social networks G of the corresponding target user, enables:
G=(U, E)；
Wherein, u indicates that the target user, U indicate the set of the friend of target user u described in the social networks, and E is indicated
With the social networks of the target user u, i expression is consumed commodity；
According to the article setThe dynamic personal interest model of the target user is established, is enabled:
Wherein, hnIndicate the newest interest of the target user, hn-1Indicate the previous interest of the newest interest, f is indicated will
The newest nonlinear function for being consumed commodity in conjunction with the previous interest；
According to the article setConstruct the short-term interest model of the social networks GIt enables:
Wherein,Indicate the newest interest of friend k in social networks G,Indicate the previous interest of friend k；
Construct the Long-term Interest model of the social networks GIt enables:
Wherein, in the social networks G friend k Long-term Interest modelIt is that user is embedded in representing matrix WuRow k；
According to the short-term interest modelWith the Long-term Interest modelSpliced, obtains split-join model sk, enable
Wherein, ReLU (x)=max (0, x) is a nonlinear activation primitive, W1It is transformation matrix；
The node for calculating the target user indicatesIt is indicated with the node of friend k in the social networks GIt enables:
Wherein,Expression for the target user u at l layers,To calculate the similar of two nodes
The function of degree,For weight of the user k about the target user u in the social networks G；
Merge feature weight about the weight computing of the target user u according to friend k in the social networks G, enable:
Wherein,It is fusion of the interest of the social networks G of the target user u on l layer；
Nonlinear transformation is carried out to the merging feature weight, is obtained:
Wherein, W(l)It is l layers of a weight matrix that is shared and can learning；
The final interest of user is calculated according to the dynamic personal interest model, enables:
Wherein, W2For the matrix of a linear transformation,For the final interest of the target user；
The probability for recommending article to be y is obtained according to the final interest of the user, it may be assumed that
Wherein, N (u) is the number of user in the social networks G, zyIt is indicated for the insertion of article y, | I | it is the number of article；
It is the log-likelihood function value of the probability calculation article of y according to recommendation article:
2. according to claim 1 dialogue-based and social influence recommended method, it is characterised in that: by the newest quilt
Nonlinear function f of the consumables in conjunction with the previous interest are as follows:
Wherein, σ is sigmoid function, σ (x)=(1+exp (- x))-1。
3. a kind of dialogue-based and social influence recommender system characterized by comprising
Module is constructed, the article set being consumed in building target user's current sessions is arranged to be used forThe corresponding mesh
The social networks G of user is marked, is enabled:
G=(U, E)；
Wherein, u indicates that the target user, U indicate the set of the friend of target user u described in the social networks, and E is indicated
With the social networks of the target user u, i expression is consumed commodity；
Dynamic personal interest model building module is arranged to be used for according to the article setEstablish the target user
Dynamic personal interest model, enable:
Wherein, hnIndicate the newest interest of the target user, hn-1Indicate the previous interest of the newest interest, f is indicated will
The newest nonlinear function for being consumed commodity in conjunction with the previous interest；
Short-term interest model construction module is arranged to be used for according to the article setConstruct the short of the social networks G
Phase interest modelIt enables:
Wherein,Indicate the newest interest of friend k in social networks G,Indicate the previous interest of friend k；
Long-term Interest model construction module is arranged to be used for constructing the Long-term Interest model of the social networks GIt enables:
Wherein, in the social networks G friend k Long-term Interest modelIt is that user is embedded in representing matrix WuRow k；
Splicing module is arranged to be used for according to the short-term interest modelWith the Long-term Interest modelSpliced,
Obtain split-join model sk, enable
Wherein, ReLU (x)=max (0, x) is a nonlinear activation primitive, W1It is transformation matrix；
Computing module, the node for being arranged to be used for calculating the target user indicateWith friend k in the social networks G
Node indicateIt enables:
Wherein,Expression for the target user u at l layers,To calculate the similar of two nodes
The function of degree,For weight of the user k about the target user u in the social networks G；
Merge feature weight computing module, is arranged to be used for according to friend k in the social networks G about the target user
The weight computing of u merges feature weight, enables:
Wherein,It is fusion of the interest of the social networks G of the target user u on l layer；
Nonlinear transformation module is arranged to be used for carrying out nonlinear transformation to the merging feature weight, obtain:
Wherein, W(l)It is l layers of a weight matrix that is shared and can learning；
Final interest computing module, is arranged to be used for that the final emerging of user is calculated according to the dynamic personal interest model
Interest enables:
Wherein, W2For the matrix of a linear transformation,For the final interest of the target user；
Probability evaluation entity is arranged to be used for the probability for obtaining that article is recommended to be y according to the final interest of the user, it may be assumed that
Wherein, N (u) is the number of user in the social networks G, zyIt is indicated for the insertion of article y, | I | it is the number of article；
Log-likelihood function value computing module, be arranged to be used for according to recommend article be y probability calculation article logarithm seemingly
Right functional value:
4. according to claim 3 dialogue-based and social influence recommender system, it is characterised in that: by the newest quilt
Nonlinear function f of the consumables in conjunction with the previous interest are as follows:
Wherein, σ is sigmoid function, σ (x)=(1+exp (- x))-1。
5. a kind of computer installation, including processor, the processor is for when executing the computer program stored in memory
It realizes as described in claims 1 or 2 the step of recommended method.
6. a kind of computer readable storage medium, is stored thereon with computer program, it is characterised in that: the computer program quilt
The step of processor realizes recommended method as claimed in claim 1 or 2 when executing.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910032524.1A CN109767301B (en) | 2019-01-14 | 2019-01-14 | Recommendation method and system, computer device and computer readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910032524.1A CN109767301B (en) | 2019-01-14 | 2019-01-14 | Recommendation method and system, computer device and computer readable storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109767301A true CN109767301A (en) | 2019-05-17 |
| CN109767301B CN109767301B (en) | 2021-05-07 |
Family
ID=66453656
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910032524.1A Active CN109767301B (en) | 2019-01-14 | 2019-01-14 | Recommendation method and system, computer device and computer readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109767301B (en) |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110147911A (en) * | 2019-05-28 | 2019-08-20 | 吉林大学 | A kind of social influence power prediction model and prediction technique based on perception of content |
| CN110245293A (en) * | 2019-05-28 | 2019-09-17 | 腾讯科技（深圳）有限公司 | A kind of Web content recalls method and apparatus |
| CN110502701A (en) * | 2019-07-08 | 2019-11-26 | 华南师范大学 | Introduce friend recommendation method, system and the storage medium of attention mechanism |
| CN111125445A (en) * | 2019-12-17 | 2020-05-08 | 北京百度网讯科技有限公司 | Community theme generation method and device, electronic equipment and storage medium |
| CN111125537A (en) * | 2019-12-31 | 2020-05-08 | 中国计量大学 | Session recommendation method based on graph representation |
| CN111222054A (en) * | 2020-01-03 | 2020-06-02 | 中国计量大学 | Session social contact recommendation method based on context neighbor relation modeling |
| CN111258593A (en) * | 2020-01-08 | 2020-06-09 | 恒安嘉新（北京）科技股份公司 | Application program prediction model establishing method and device, storage medium and terminal |
| CN111260447A (en) * | 2020-02-11 | 2020-06-09 | 深圳前海达闼云端智能科技有限公司 | Commodity recommendation method and device for vending robot |
| CN111460331A (en) * | 2020-04-07 | 2020-07-28 | 中国计量大学 | Conversation recommendation method based on space-time diagram |
| CN111782765A (en) * | 2020-06-24 | 2020-10-16 | 安徽农业大学 | Recommendation method based on graph attention machine mechanism |
| CN112163147A (en) * | 2020-06-09 | 2021-01-01 | 中森云链(成都)科技有限责任公司 | Recommendation method for website session scene |
| CN112445981A (en) * | 2020-11-04 | 2021-03-05 | 西安电子科技大学 | Social and consumption joint recommendation system, method, storage medium and computer equipment |
| CN112559904A (en) * | 2020-12-16 | 2021-03-26 | 中国计量大学 | Conversational social recommendation method based on door mechanism and multi-modal graph network |
| CN112883268A (en) * | 2021-02-22 | 2021-06-01 | 中国计量大学 | Session recommendation method considering user multiple interests and social influence |
| CN112905887A (en) * | 2021-02-22 | 2021-06-04 | 中国计量大学 | Conversation recommendation method based on multi-interest short-term priority model |
| CN112950325A (en) * | 2021-03-16 | 2021-06-11 | 山西大学 | Social behavior fused self-attention sequence recommendation method |
| CN112948683A (en) * | 2021-03-16 | 2021-06-11 | 山西大学 | Socialized recommendation method with dynamic fusion of social information |
| CN113407862A (en) * | 2021-05-20 | 2021-09-17 | 桂林电子科技大学 | Sequential social contact recommendation method, system and storage medium based on door mechanism |
| CN114491150A (en) * | 2022-03-28 | 2022-05-13 | 苏州浪潮智能科技有限公司 | Video recommendation method, system, device and computer readable storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101944218A (en) * | 2010-01-27 | 2011-01-12 | 北京大学 | Personalized recommended method based on picture under social network and system thereof |
| US20170032257A1 (en) * | 2015-07-29 | 2017-02-02 | Google Inc. | Modeling personal entities |
| WO2018191555A1 (en) * | 2017-04-14 | 2018-10-18 | Drishti Technologies. Inc | Deep learning system for real time analysis of manufacturing operations |
-
2019
- 2019-01-14 CN CN201910032524.1A patent/CN109767301B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101944218A (en) * | 2010-01-27 | 2011-01-12 | 北京大学 | Personalized recommended method based on picture under social network and system thereof |
| US20170032257A1 (en) * | 2015-07-29 | 2017-02-02 | Google Inc. | Modeling personal entities |
| WO2018191555A1 (en) * | 2017-04-14 | 2018-10-18 | Drishti Technologies. Inc | Deep learning system for real time analysis of manufacturing operations |
Non-Patent Citations (1)
| Title |
|---|
| 孙广路: ""基于用户社交信息和搭配知识融合的服饰推荐算法研究"", 《中国博士学位论文全文数据库信息科技辑》 * |
Cited By (32)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110245293A (en) * | 2019-05-28 | 2019-09-17 | 腾讯科技（深圳）有限公司 | A kind of Web content recalls method and apparatus |
| CN110147911B (en) * | 2019-05-28 | 2022-09-09 | 吉林大学 | Social influence prediction model and prediction method based on content perception |
| CN110245293B (en) * | 2019-05-28 | 2023-07-11 | 深圳市雅阅科技有限公司 | Network content recall method and device |
| CN110147911A (en) * | 2019-05-28 | 2019-08-20 | 吉林大学 | A kind of social influence power prediction model and prediction technique based on perception of content |
| CN110502701B (en) * | 2019-07-08 | 2022-03-25 | 华南师范大学 | Friend recommendation method, system and storage medium introducing attention mechanism |
| CN110502701A (en) * | 2019-07-08 | 2019-11-26 | 华南师范大学 | Introduce friend recommendation method, system and the storage medium of attention mechanism |
| CN111125445A (en) * | 2019-12-17 | 2020-05-08 | 北京百度网讯科技有限公司 | Community theme generation method and device, electronic equipment and storage medium |
| CN111125445B (en) * | 2019-12-17 | 2023-08-15 | 北京百度网讯科技有限公司 | Community theme generation method and device, electronic equipment and storage medium |
| CN111125537A (en) * | 2019-12-31 | 2020-05-08 | 中国计量大学 | Session recommendation method based on graph representation |
| CN111125537B (en) * | 2019-12-31 | 2020-12-22 | 中国计量大学 | Session recommendation method based on graph representation |
| CN111222054B (en) * | 2020-01-03 | 2020-12-11 | 中国计量大学 | Session social contact recommendation method based on context neighbor relation modeling |
| CN111222054A (en) * | 2020-01-03 | 2020-06-02 | 中国计量大学 | Session social contact recommendation method based on context neighbor relation modeling |
| CN111258593B (en) * | 2020-01-08 | 2023-10-27 | 恒安嘉新（北京）科技股份公司 | Application program prediction model building method and device, storage medium and terminal |
| CN111258593A (en) * | 2020-01-08 | 2020-06-09 | 恒安嘉新（北京）科技股份公司 | Application program prediction model establishing method and device, storage medium and terminal |
| CN111260447A (en) * | 2020-02-11 | 2020-06-09 | 深圳前海达闼云端智能科技有限公司 | Commodity recommendation method and device for vending robot |
| CN111460331B (en) * | 2020-04-07 | 2020-12-25 | 中国计量大学 | Conversation recommendation method based on space-time diagram |
| CN111460331A (en) * | 2020-04-07 | 2020-07-28 | 中国计量大学 | Conversation recommendation method based on space-time diagram |
| CN112163147A (en) * | 2020-06-09 | 2021-01-01 | 中森云链(成都)科技有限责任公司 | Recommendation method for website session scene |
| CN111782765A (en) * | 2020-06-24 | 2020-10-16 | 安徽农业大学 | Recommendation method based on graph attention machine mechanism |
| CN112445981A (en) * | 2020-11-04 | 2021-03-05 | 西安电子科技大学 | Social and consumption joint recommendation system, method, storage medium and computer equipment |
| CN112559904A (en) * | 2020-12-16 | 2021-03-26 | 中国计量大学 | Conversational social recommendation method based on door mechanism and multi-modal graph network |
| CN112559904B (en) * | 2020-12-16 | 2021-10-22 | 中国计量大学 | Conversational social recommendation method based on door mechanism and multi-modal graph network |
| CN112883268A (en) * | 2021-02-22 | 2021-06-01 | 中国计量大学 | Session recommendation method considering user multiple interests and social influence |
| CN112883268B (en) * | 2021-02-22 | 2022-02-01 | 中国计量大学 | Session recommendation method considering user multiple interests and social influence |
| CN112905887A (en) * | 2021-02-22 | 2021-06-04 | 中国计量大学 | Conversation recommendation method based on multi-interest short-term priority model |
| CN112948683A (en) * | 2021-03-16 | 2021-06-11 | 山西大学 | Socialized recommendation method with dynamic fusion of social information |
| CN112950325A (en) * | 2021-03-16 | 2021-06-11 | 山西大学 | Social behavior fused self-attention sequence recommendation method |
| CN112950325B (en) * | 2021-03-16 | 2023-10-03 | 山西大学 | Self-attention sequence recommendation method for social behavior fusion |
| CN113407862B (en) * | 2021-05-20 | 2022-06-17 | 桂林电子科技大学 | Sequential social contact recommendation method, system and storage medium based on door mechanism |
| CN113407862A (en) * | 2021-05-20 | 2021-09-17 | 桂林电子科技大学 | Sequential social contact recommendation method, system and storage medium based on door mechanism |
| CN114491150A (en) * | 2022-03-28 | 2022-05-13 | 苏州浪潮智能科技有限公司 | Video recommendation method, system, device and computer readable storage medium |
| CN114491150B (en) * | 2022-03-28 | 2022-07-15 | 苏州浪潮智能科技有限公司 | Video recommendation method, system, equipment and computer readable storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109767301B (en) | 2021-05-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109767301A (en) | Recommended method and system, computer installation, computer readable storage medium | |
| CN111222054B (en) | Session social contact recommendation method based on context neighbor relation modeling | |
| Xiao et al. | Hidden semi-Markov model-based reputation management system for online to offline (O2O) e-commerce markets | |
| Khan et al. | Group decision support using fuzzy cognitive maps for causal reasoning | |
| CN109582956A (en) | text representation method and device applied to sentence embedding | |
| Howick et al. | Building confidence in models for multiple audiences: the modelling cascade | |
| AU2013249251A1 (en) | Method and apparatus for an intuitive customer experience | |
| CN114595396B (en) | Federal learning-based sequence recommendation method and system | |
| CN111210008A (en) | Method and device for processing interactive data by using LSTM neural network model | |
| CN112685657B (en) | Conversation social recommendation method based on multi-mode cross fusion graph network | |
| CN112115358B (en) | Personalized recommendation method utilizing multi-hop path characteristics in knowledge graph | |
| CN110689110B (en) | Method and device for processing interaction event | |
| CN111931054B (en) | Sequence recommendation method and system based on improved residual error structure | |
| CN110210233A (en) | Joint mapping method, apparatus, storage medium and the computer equipment of prediction model | |
| CN112528165A (en) | Session social recommendation method based on dynamic routing graph network | |
| CN112819575B (en) | Session recommendation method considering repeated purchasing behavior | |
| US8560490B2 (en) | Collaborative networking with optimized inter-domain information quality assessment | |
| CN115270001A (en) | Privacy protection recommendation method and system based on cloud collaborative learning | |
| CN112559904B (en) | Conversational social recommendation method based on door mechanism and multi-modal graph network | |
| Ahamed et al. | A recommender system based on deep neural network and matrix factorization for collaborative filtering | |
| García-Magariño et al. | A repository of method fragments for agent-oriented development of learning-based edge computing systems | |
| Sha et al. | Modeling participation behaviors in design crowdsourcing using a bipartite network-based approach | |
| CN114154080A (en) | Dynamic socialization recommendation method based on graph neural network | |
| CN111652673B (en) | Intelligent recommendation method, device, server and storage medium | |
| Azadi et al. | A new fuzzy goal directed benchmarking for supplier selection |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |