CN109241412A - A kind of recommended method, system and electronic equipment based on network representation study - Google Patents
A kind of recommended method, system and electronic equipment based on network representation study Download PDFInfo
- Publication number
- CN109241412A CN109241412A CN201810942135.8A CN201810942135A CN109241412A CN 109241412 A CN109241412 A CN 109241412A CN 201810942135 A CN201810942135 A CN 201810942135A CN 109241412 A CN109241412 A CN 109241412A
- Authority
- CN
- China
- Prior art keywords
- article
- user
- node
- network
- occurrence
- 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
This application involves a kind of recommended method, system and electronic equipments based on network representation study.This method comprises: step a: constructing user-article co-occurrence network based on bipartite graph network and injection shadow figure；Step b: it is directed to the user-article co-occurrence net definitions search strategy, obtains the neighbor node of each user node and article node；Step c: it according to each user node and article node and respective neighbor node, is indicated using the vector that network representation learns to obtain each user node and article node；Step d: indicating according to the vector of each user node and article node, the maximally related article node of each user node is calculated by vector, and recommend maximally related article to each user according to calculated result.The problem of the application slows down the sparsity problem of collaborative filtering, makes the explanatory stronger of recommender system, alleviates scalability in collaborative filtering significantly.
Description
Technical field
The application belongs to data mining and recommended technology field, in particular to a kind of recommendation side based on network representation study
Method, system and electronic equipment.
Background technique
With the arriving of big data era, recommender system more and more attention has been paid to, it help the quick garbled data of people,
Solve the problems, such as to do well on information overload, can to the potentially possible article liked of each user-customized recommended, such as
The similar article of Taobao is recommended, the music of Netease's cloud music is recommended etc..Nowadays recommender system obtains broad development, has penetrated into
The various aspects of people's daily life, such as: music recommendation, film recommendation, e-commerce, mobile phone application etc..
With the universal of recommender system, various recommended methods are suggested: including content-based recommendation, being cooperateed with
Filter, recommendation based on figure etc..Wherein, collaborative filtering is used to predict that the matrix disassembling method of scoring of the user to article is undoubtedly most
One of successful proposed algorithm, existing recommender system uses collaborative filtering method more.Collaborative filtering assumes phase in historical record
As user and article in the future also can be similar, wherein matrix disassembling method application at most, user to the score information of article
Stored using rating matrix, be further decomposed into low-dimensional article matrix and low-dimensional user's matrix multiple, it is can thus being lost or
The no score information of person, and then complete to recommend.
Application number " 201410007387.3 ", the patent of invention of entitled " collaborative filtering recommending method based on Web Community "
Disclose a kind of collaborative filtering recommending method based on Web Community, recommended as follows: acquisition user, which treats, to push away
The score information of project is recommended, and generates the relationship between user and user indirectly using the score data that user treats recommended project
Network；Calculate the similarity between user；It is detected by the community based on similarity and customer relationship network is divided into several
Communities of users；The maximum k user of similarity forms neighbour user's set in community where choosing user, is collected according to neighbour user
Conjunction carries out prediction scoring to the project that target user does not score；By project recommendation maximum in score in predicting value to user.
Application number " 201710799698.1 ", it is entitled " a kind of resource recommendation system and method based on Network Study Environment "
Using the thought of collaborative filtering, user group similar with target user is filtered out, in conjunction with similar users group and target user
Similarity and the confidence level of user's scoring carry out the scoring and recommendation of education resource, are directed to the scoring of education resource with user
Property and scoring objectivity, to recommend personalized and high quality education resource for user.
However, collaborative filtering suffers from the influence of sparsity and scalability problem.Firstly, in actual life, user
Score information to article be it is seldom, many sluggish users are to seldom article scoring or many unwelcome objects
Product have obtained seldom scoring, and score information concentrates in several popular articles, and therefore, rating matrix is very in practical situations
Sparse and distribution is irregular.Secondly, recommender system will often recommend the different article of different user, and then realize individual character
Change demand, however different recommendations is carried out for different users, recommendation process needs global calculation, and with user and article
Quantity is continuously increased, and the consumption of global calculation becomes constantly to increase, so that scalability becomes main problem.
Summary of the invention
This application provides a kind of recommended method, system and electronic equipments based on network representation study, it is intended at least exist
One of above-mentioned technical problem in the prior art is solved to a certain extent.
To solve the above-mentioned problems, this application provides following technical solutions:
A kind of recommended method based on network representation study, comprising the following steps:
Step a: user-article co-occurrence network is constructed based on bipartite graph network and injection shadow figure；
Step b: it is directed to the user-article co-occurrence net definitions search strategy, obtains each user node and article section
The neighbor node of point；
Step c: according to each user node and article node and respective neighbor node, learnt using network representation
It is indicated to the vector of each user node and article node；
Step d: it is indicated according to the vector of each user node and article node, each use is calculated by vector
The maximally related article node of family node, and maximally related article is recommended to each user according to calculated result.
The technical solution that the embodiment of the present application is taken further include: described to be based on bipartite graph network and list in the step a
Projection figure building user-article co-occurrence network specifically includes:
Step a1: the scoring using bipartite graph network storage user to article constructs user-article bipartite graph；
Step a2: using the cooccurrence relation between injection shadow figure storage article, article co-occurrence network is constructed；
Step a3: it is based on user-article bipartite graph and article co-occurrence network struction user-article co-occurrence network, and is arranged
OT parameter and PR parameter filter the useless cooccurrence relation in the user-article co-occurrence network.
The technical solution that the embodiment of the present application is taken further include: in the step a3, the OT parameter is for reflecting two
Cooccurrence relation between a article is strong and weak, the OT calculation formula between two articles are as follows:
In above formula,For judging in favorites list that whether article appears in user i,Calculation formula
Are as follows:
The PR parameter is used to filter out influence of the user of habit favorable comment for article cooccurrence relation, PR calculation formula
Are as follows:
In above formula,Indicate the article scoring set of i-th of user,Represent the happiness of i-th of user
Like article set.
The technical solution that the embodiment of the present application is taken further include: in the step b, defined search strategy includes wide
Preferential sampling policy and depth-first sampling policy are spent, the neighborhood node of the breadth First sampling policy is restricted to directly connect
It is connected to the node of source node；The neighborhood node of the depth-first sampling policy is by the node group apart from source node continuous sampling
At.
The technical solution that the embodiment of the present application is taken further include: in the step c, it is described according to each user node and
Article node and respective neighbor node learn to obtain the vector table of each user node and article node using network representation
Show specifically: using skip-gram training, and combine stochastic gradient descent and negative sampling, obtain each user node and article
The vector of node indicates.
A kind of another technical solution that the embodiment of the present application is taken are as follows: recommender system based on network representation study, comprising:
User-article co-occurrence network struction module: for same based on bipartite graph network and injection shadow figure building user-article
Existing network；
Search strategy definition module: for being directed to the user-article co-occurrence net definitions search strategy, each use is obtained
The neighbor node of family node and article node；
Network representation study module: for making according to each user node and article node and respective neighbor node
It is indicated with the vector that network representation learns to obtain each user node and article node；
Vector calculation module: for being indicated according to the vector of each user node and article node, by meter
Calculation obtains the maximally related article node of each user node, and recommends maximally related article to each user according to calculated result.
The technical solution that the embodiment of the present application is taken further include: the user-article co-occurrence network struction module includes:
Bipartite graph construction unit: for using scoring of the bipartite graph network storage user to article, user-article two is constructed
Component；
Article co-occurrence network struction unit: for using the cooccurrence relation between injection shadow figure storage article, article is constructed
Co-occurrence network；
User-article co-occurrence network struction unit: user-article bipartite graph and article co-occurrence network struction user-are based on
Article co-occurrence network, and OT parameter is set and PR parameter filters useless cooccurrence relation in the user-article co-occurrence network.
The technical solution that the embodiment of the present application is taken further include: the co-occurrence that the OT parameter is used to reflect between two articles is closed
System is strong and weak, the OT calculation formula between two articles are as follows:
In above formula,For judging in favorites list that whether article appears in user i,Calculation formula
Are as follows:
The PR parameter is used to filter out influence of the user of habit favorable comment for article cooccurrence relation, PR calculation formula
Are as follows:
In above formula,Indicate the article scoring set of i-th of user,Represent the happiness of i-th of user
Like article set.
The technical solution that the embodiment of the present application is taken further include: the search strategy packet of described search policy definition module definition
Breadth First sampling policy and depth-first sampling policy are included, the neighborhood node of the breadth First sampling policy is restricted to directly
It is connected to the node of source node in succession；The neighborhood node of the depth-first sampling policy is by the node apart from source node continuous sampling
Composition.
The technical solution that the embodiment of the present application is taken further include: the network representation study module is according to each user node
With article node and respective neighbor node, learn to obtain the vector of each user node and article node using network representation
It indicates specifically: using skip-gram training, and combine stochastic gradient descent and negative sampling, obtain each user node and object
The vector of moral integrity point indicates.
The another technical solution that the embodiment of the present application is taken are as follows: a kind of electronic equipment, comprising:
At least one processor；And
The memory being connect at least one described processor communication；Wherein,
The memory is stored with the instruction that can be executed by one processor, and described instruction is by described at least one
It manages device to execute, so that at least one described processor is able to carry out the following of the above-mentioned recommended method based on network representation study
Operation:
Step a: user-article co-occurrence network is constructed based on bipartite graph network and injection shadow figure；
Step b: it is directed to the user-article co-occurrence net definitions search strategy, obtains each user node and article section
The neighbor node of point；
Step c: according to each user node and article node and respective neighbor node, learnt using network representation
It is indicated to the vector of each user node and article node；
Step d: it is indicated according to the vector of each user node and article node, each use is calculated by vector
The maximally related article node of family node, and maximally related article is recommended to each user according to calculated result.
Compared with the existing technology, the embodiment of the present application generate beneficial effect be: the embodiment of the present application based on network
It is same to indicate that recommended method, system and the electronic equipment of study combine bipartite graph and single mode perspective view to construct user-article first
Then existing network has formulated breadth-first search and the sampling of depth-first search strategy traverses network, finally by network representation
Learning method obtains recommendation results, compared with the existing technology, the application has the following advantages:
1, co-occurrence network is combined into using bipartite graph and injection shadow figure to show the relationship of user and article in recommender system,
Slow down the sparsity problem of collaborative filtering；
2, using depth-first and breadth first traversal network, user can be explored simultaneously to the scoring relationship of article, object
Product cooccurrence relation, user's similarity relationships make the explanatory stronger of recommender system；
3, using the application of the network representation method combination stochastic gradient descent learnt and negative sampling, global calculation is not needed
The problem of can also recommending, alleviating scalability in collaborative filtering significantly.
Detailed description of the invention
Fig. 1 is the flow chart of the recommended method based on network representation study of the embodiment of the present application；
Fig. 2 (a) is the user-article bipartite graph network diagram of the embodiment of the present application；
Fig. 2 (b) is article co-occurrence network diagram；
Fig. 3 (a) is user-article co-occurrence network diagram；
Fig. 3 (b) and Fig. 3 (c) is respectively user-article co-occurrence network compactness and similitude schematic diagram；
Fig. 4 is search strategy schematic diagram；
Fig. 5 is the structural schematic diagram of the recommender system based on network representation study of the embodiment of the present application；
Fig. 6 is the hardware device structural representation of the recommended method provided by the embodiments of the present application based on network representation study
Figure.
Specific embodiment
It is with reference to the accompanying drawings and embodiments, right in order to which the objects, technical solutions and advantages of the application are more clearly understood
The application is further elaborated.It should be appreciated that specific embodiment described herein is only to explain the application, not
For limiting the application.
Referring to Fig. 1, being the flow chart of the recommended method based on network representation study of the embodiment of the present application.The application is real
Apply example based on network representation study recommended method the following steps are included:
Step 100: the scoring using bipartite graph network storage user to article constructs user-article bipartite graph；
It is the user-article bipartite graph network diagram of the embodiment of the present application as shown in Fig. 2 (a) in step 100.G=
(U, O, E), wherein U indicates user (user) node set { u1,u2,u3...um, O indicates item (article) node set { o1,
o2,o3...on, it is evident that the two set are disjoint.E indicates the set on the side in bipartite graph network.Such as Fig. 2 (a) institute
The user of the m=4 shown, n=5-article bipartite graph network, have side to make and if only if user to the article between user and article
Scoring (1-5).Weight on side represents scoring of the user to article, when scoring is less than given threshold (the application given threshold
It is 3, can be specifically set according to practical application) when, show that user does not like the article.On the contrary, when scoring is greater than setting threshold
When value, show that user likes the article, so the weight on side also reflects user to the fancy grade of article.The application usesThe article scoring set for indicating i-th of user, usesRepresent i-th of user likes article set.
For example, in Fig. 2 (a)And
Step 200: using the cooccurrence relation between injection shadow figure storage article, constructing article co-occurrence network；
It is article co-occurrence network diagram as shown in Fig. 2 (b) in step 200.The application defines object using injection shadow figure
Product co-occurrence network: when two nodes in item set provide high score by the same user in user set, this
It is connected between two node points with a line, the weight on side represents the two nodes and appears in the same user between node
Provide the Co-occurrence Times (co-occurrence number, abbreviation OT) of high score.
Step 300: being based on user-article bipartite graph and article co-occurrence network struction user-article co-occurrence network, and be arranged
Nothing in OT parameter and PR (Personal Rating Habit, individual's grading habit) parameter filtering user-article co-occurrence network
Use cooccurrence relation；
In step 300, the interactive information of existing user and article in user-article co-occurrence network, while again including article
Between cooccurrence relation, make recommend sparsity problem be eased.It needs to filter the lower information of value ratio when constructing network, collects
In valuable information, to avoid the too big situation of network density.Therefore, the application passes through setting the two parameter mistakes of OT and PR
Filter garbage.
Specifically, OT has reflected the power of the cooccurrence relation between two articles, the OT calculation formula between two articles are as follows:
In formula (1),For judging in favorites list that whether article appears in user i,Calculating
Formula are as follows:
The size of the OT value of two articles reflects a possibility that they are liked by the same user simultaneously size, objective anti-
The characteristic similarity of two articles is reflected.OT value size and its existing number are presented long-tail distribution, small (such as the appearance one of OT value
It is secondary) cooccurrence relation it is more, in huge cyberrelationship, as OT be 1 cooccurrence relation can ignore completely because not having
There is identification and reference value is also very low, if they are also added in user-article co-occurrence network, not but not improves efficiency
The calculated load of network representation study is increased instead.Therefore, the application filters out OT value by setting OT parameter and sets less than OT
The cooccurrence relation for determining threshold value (the application OT given threshold is 2, specifically can be according to being using being set), promotes recommendation results
Accuracy rate.
Similarly, one PR parameter of the application setting filters out the user of habit favorable comment (positive rating 100%) for article
The influence of cooccurrence relation:
OT=2, user-article co-occurrence network of PR=1, filtering while filtering out the cooccurrence relation less than 2 times are set
Falling positive rating is absolutely influence of the people for article cooccurrence relation, obtains user shown in Fig. 3 (a)-article co-occurrence net
Network schematic diagram.
Step 400: being directed to user-article co-occurrence net definitions search strategy, obtain each user node and article node
Neighbor node sequence；
In step 400, it for user-article co-occurrence network, needs to obtain each user using different search strategies
The neighbor node sequence of node and article node gives skip-gram training.Before defining search strategy, first define in network
Compactness and similitude, compactness refer to what the node for belonging to a cluster should be leaned in the low-dimensional vector that network representation learns
Closely；Similitude refers to that the identical node of figure should lean on close in each cluster；Compactness and similitude are specifically such as Fig. 3
(b) and shown in Fig. 3 (c).
In the embodiment of the present application, the search strategy of definition includes breadth First sampling policy (BFS) and depth-first sampling
Tactful (DFS), wherein BFS can have found potential user preference structure well and then find similar user, and DFS can be with
The interactive relation between article and user and article in close relations is found, article is commented so as to explore user simultaneously
Divide relationship, article cooccurrence relation, user's similarity relationships, alleviate the scalability problem of recommender system significantly, makes to recommend system
That unites is explanatory stronger.
Breadth First sampling: neighborhood node is restricted to be directly connected to the node of source node.As shown in figure 4, being saved from source
It is t, x respectively that point v, which takes 3 neighbor nodes,1,x2。
Depth-first sampling: neighborhood node is made of the node apart from source node continuous sampling.In Fig. 4, from source node v
Start to sample 3 neighbor nodes are as follows: x3、x4、x5。
Step 500: according to known each user node and article node and respective neighbor node, using net list
Stochastic gradient descent and negative sampling are practised and combined to dendrography, and the vector for obtaining each user node and article node indicates；
In step 500, the embodiment of the present application using skip-gram training obtain each user node and article node to
Amount indicates that Skip-gram is the method that natural language processing is used, and can predict the interior of the context of each word fixed size
Hold, and obtains the vector of each word.The application Skip-gram will be applied to network representation study the inside, and combine with
The problem of machine gradient declines and the application of negative sampling, alleviates scalability in collaborative filtering significantly.
Step 600: being indicated according to the vector of user node and article node, each user node is calculated by vector
Maximally related article node, and maximally related article is recommended to the user according to calculated result.
Referring to Fig. 5, being the structural schematic diagram of the recommender system based on network representation study of the embodiment of the present application.This Shen
Please the recommender system based on network representation study of embodiment include that user-article co-occurrence network struction module, search strategy are fixed
Adopted module, network representation study module and vector calculation module.
User-article co-occurrence network struction module: for same based on bipartite graph network and injection shadow figure building user-article
Existing network；The interactive information of existing user and article in the user of the embodiment of the present application-article co-occurrence network, while inclusion again
Cooccurrence relation between product makes that sparsity problem is recommended to be eased.
Specifically, user-article co-occurrence network struction module includes:
Bipartite graph construction unit: for using scoring of the bipartite graph network storage user to article, user-article two is constructed
Component；It is the user-article bipartite graph network diagram of the embodiment of the present application as shown in Fig. 2 (a).G=(U, O, E), wherein U
Indicate user (user) node set { u1,u2,u3...um, O indicates item (article) node set { o1,o2,o3...on, very
Obviously, the two set are disjoint.E indicates the set on the side in bipartite graph network.The m=4 as shown in Fig. 2 (a), n=5
User-article bipartite graph network, there is side to be made that scoring (1-5) to the article and if only if user between user and article.
Weight on side represents scoring of the user to article, and when scoring is less than given threshold, (the application given threshold is 3, specifically can root
Set according to practical application) when, show that user does not like the article.On the contrary, showing user when scoring is greater than given threshold
The article is liked, so the weight on side also reflects user to the fancy grade of article.The application usesIndicate the
The article of i user, which scores, to be gathered, and is usedRepresent i-th of user likes article set.For example, in Fig. 2 (a)And
Article co-occurrence network struction unit: for using the cooccurrence relation between injection shadow figure storage article, article is constructed
Co-occurrence network；It is article co-occurrence network diagram as shown in Fig. 2 (b).The application defines article co-occurrence net using injection shadow figure
Network: when two nodes in item set provide high score by the same user in user set, the two nodes
It is connected between point with a line, the weight on side, which represents the two nodes and appears in the same user, between node provides high score
Co-occurrence Times (co-occurrence number, abbreviation OT).
User-article co-occurrence network struction unit: for being used in conjunction with user-article bipartite graph and article co-occurrence network struction
Family-article co-occurrence network, and the useless cooccurrence relation being arranged in OT parameter and PR parameter filtering user-article co-occurrence network；Structure
It needs to filter value ratio lower information when establishing network, concentrates valuable information, to avoid the too big situation of network density.Cause
This, the application passes through setting the two parameter filtering useless information of OT and PR.
Specifically, OT has reflected that the cooccurrence relation of two articles is strong and weak, the OT calculation formula between two articles are as follows:
In formula (1),For judging in favorites list that whether article appears in user i,Calculating
Formula are as follows:
The size of the OT value of two articles reflects a possibility that they are liked by the same user simultaneously size, objective anti-
The characteristic similarity of two articles is reflected.OT value size and its existing number are presented long-tail distribution, small (such as the appearance one of OT value
It is secondary) cooccurrence relation it is more, in huge cyberrelationship, as OT be 1 cooccurrence relation can ignore completely because not having
There is identification and reference value is also very low, if they are also added in user-article co-occurrence network, not but not improves efficiency
The calculated load of network representation study is increased instead.Therefore, the application filters out OT value by setting OT parameter and sets less than OT
The cooccurrence relation for determining threshold value (the application OT given threshold is 2, specifically can be according to being using being set), promotes recommendation results
Accuracy rate.
Similarly, one PR parameter of the application setting filters out the user of habit favorable comment (positive rating 100%) for article
The influence of cooccurrence relation:
OT=2, user-article co-occurrence network of PR=1, filtering while filtering out the cooccurrence relation less than 2 times are set
Falling positive rating is absolutely influence of the people for article cooccurrence relation, obtains user shown in Fig. 3 (a)-article co-occurrence net
Network schematic diagram.
Search strategy definition module: for being directed to user-article co-occurrence net definitions search strategy, each user's section is obtained
The neighbor node sequence of point and article node；Before defining search strategy, the compactness and similitude in network are first defined, closely
It is close that property refers to that the node for belonging to a cluster should be leaned in the low-dimensional vector that network representation learns；Similitude refers to each
The identical node of figure should lean on close in cluster.
In the embodiment of the present application, the search strategy of definition includes breadth First sampling policy (BFS) and depth-first sampling
Tactful (DFS), wherein BFS can have found potential user preference structure well and then find similar user, and DFS can be with
The interactive relation between article and user and article in close relations is found, article is commented so as to explore user simultaneously
Divide relationship, article cooccurrence relation, user's similarity relationships, alleviate the scalability problem of recommender system significantly, makes to recommend system
That unites is explanatory stronger.
Breadth First sampling: neighborhood node is restricted to be directly connected to the node of source node.As shown in figure 4, being saved from source
It is t, x respectively that point v, which takes 3 neighbor nodes,1,x2。
Depth-first sampling: neighborhood node is made of the node apart from source node continuous sampling.In Fig. 4, from source node v
Start to sample 3 neighbor nodes are as follows: x3、x4、x5。
Network representation study module: for being saved according to known each user node and article node and respective neighbours
Point learns using network representation and combines stochastic gradient descent and negative sampling, obtain each user node and article node to
Amount indicates；The embodiment of the present application obtains the vector expression of each user node and article node using skip-gram training,
Skip-gram is the method that natural language processing is used, and can predict the content of the context of each word fixed size, and
Obtain the vector of each word.The application will apply to Skip-gram network representation study the inside, and combine under stochastic gradient
The problem of dropping and bear the application sampled, alleviating scalability in collaborative filtering significantly.
Vector calculation module: it for being indicated according to the vector of user node and article node, is calculated often by vector
A maximally related article node of user node, and maximally related article is recommended to the user according to calculated result.
Fig. 6 is the hardware device structural representation of the recommended method provided by the embodiments of the present application based on network representation study
Figure.As shown in fig. 6, the equipment includes one or more processors and memory.It takes a processor as an example, which may be used also
To include: input system and output system.
Processor, memory, input system and output system can be connected by bus or other modes, in Fig. 6 with
For being connected by bus.
Memory as a kind of non-transient computer readable storage medium, can be used for storing non-transient software program, it is non-temporarily
State computer executable program and module.Processor passes through operation non-transient software program stored in memory, instruction
And module realizes the place of above method embodiment thereby executing the various function application and data processing of electronic equipment
Reason method.
Memory may include storing program area and storage data area, wherein storing program area can storage program area, extremely
Application program required for a few function；It storage data area can storing data etc..In addition, memory may include that high speed is random
Memory is accessed, can also include non-transient memory, a for example, at least disk memory, flush memory device or other are non-
Transient state solid-state memory.In some embodiments, it includes the memory remotely located relative to processor that memory is optional, this
A little remote memories can pass through network connection to processing system.The example of above-mentioned network includes but is not limited to internet, enterprise
Intranet, local area network, mobile radio communication and combinations thereof.
Input system can receive the number or character information of input, and generate signal input.Output system may include showing
Display screen etc. shows equipment.
One or more of module storages in the memory, are executed when by one or more of processors
When, execute the following operation of any of the above-described embodiment of the method:
Step a: user-article co-occurrence network is constructed based on bipartite graph network and injection shadow figure；
Step b: it is directed to the user-article co-occurrence net definitions search strategy, obtains each user node and article section
The neighbor node of point；
Step c: according to each user node and article node and respective neighbor node, learnt using network representation
It is indicated to the vector of each user node and article node；
Step d: it is indicated according to the vector of each user node and article node, each use is calculated by vector
The maximally related article node of family node, and maximally related article is recommended to each user according to calculated result.
Method provided by the embodiment of the present application can be performed in the said goods, has the corresponding functional module of execution method and has
Beneficial effect.The not technical detail of detailed description in the present embodiment, reference can be made to method provided by the embodiments of the present application.
The embodiment of the present application provides a kind of non-transient (non-volatile) computer storage medium, and the computer storage is situated between
Matter is stored with computer executable instructions, the executable following operation of the computer executable instructions:
Step a: user-article co-occurrence network is constructed based on bipartite graph network and injection shadow figure；
Step b: it is directed to the user-article co-occurrence net definitions search strategy, obtains each user node and article section
The neighbor node of point；
Step c: according to each user node and article node and respective neighbor node, learnt using network representation
It is indicated to the vector of each user node and article node；
Step d: it is indicated according to the vector of each user node and article node, each use is calculated by vector
The maximally related article node of family node, and maximally related article is recommended to each user according to calculated result.
The embodiment of the present application provides a kind of computer program product, and the computer program product is non-temporary including being stored in
Computer program on state computer readable storage medium, the computer program include program instruction, when described program instructs
When being computer-executed, the computer is made to execute following operation:
Step a: user-article co-occurrence network is constructed based on bipartite graph network and injection shadow figure；
Step b: it is directed to the user-article co-occurrence net definitions search strategy, obtains each user node and article section
The neighbor node of point；
Step c: according to each user node and article node and respective neighbor node, learnt using network representation
It is indicated to the vector of each user node and article node；
Step d: it is indicated according to the vector of each user node and article node, each use is calculated by vector
The maximally related article node of family node, and maximally related article is recommended to each user according to calculated result.
Recommended method, system and the electronic equipment based on network representation study of the embodiment of the present application combines bipartite graph first
User-article co-occurrence network is constructed with single mode perspective view, has then formulated breadth-first search and depth-first search strategy
Traverses network sampling, obtain recommendation results finally by network representation learning method, compared with the existing technology, the application have with
Lower advantage:
1, co-occurrence network is combined into using bipartite graph and injection shadow figure to show the relationship of user and article in recommender system,
Slow down the sparsity problem of collaborative filtering；
2, using depth-first and breadth first traversal network, user can be explored simultaneously to the scoring relationship of article, object
Product cooccurrence relation, user's similarity relationships make the explanatory stronger of recommender system；
3, using the application of the network representation method combination stochastic gradient descent learnt and negative sampling, global calculation is not needed
The problem of can also recommending, alleviating scalability in collaborative filtering significantly.
The foregoing description of the disclosed embodiments makes professional and technical personnel in the field can be realized or use the application.
Various modifications to these embodiments will be readily apparent to those skilled in the art, defined herein
General Principle can realize in other embodiments without departing from the spirit or scope of the application.Therefore, this Shen
These embodiments shown in the application please be not intended to be limited to, and are to fit to special with principle disclosed in the present application and novelty
The consistent widest scope of point.
Claims (11)
1. a kind of recommended method based on network representation study, which comprises the following steps:
Step a: user-article co-occurrence network is constructed based on bipartite graph network and injection shadow figure；
Step b: being directed to the user-article co-occurrence net definitions search strategy, obtains each user node and article node
Neighbor node；
Step c: according to each user node and article node and respective neighbor node, learn to obtain using network representation every
The vector of a user node and article node indicates；
Step d: indicating according to the vector of each user node and article node, and each user is calculated by vector and saves
The maximally related article node of point, and maximally related article is recommended to each user according to calculated result.
2. the recommended method according to claim 1 based on network representation study, which is characterized in that in the step a,
It is described to be specifically included based on bipartite graph network and injection shadow figure building user-article co-occurrence network:
Step a1: the scoring using bipartite graph network storage user to article constructs user-article bipartite graph；
Step a2: using the cooccurrence relation between injection shadow figure storage article, article co-occurrence network is constructed；
Step a3: being based on user-article bipartite graph and article co-occurrence network struction user-article co-occurrence network, and OT ginseng is arranged
Several and PR parameter filters the useless cooccurrence relation in the user-article co-occurrence network.
3. the recommended method according to claim 2 based on network representation study, which is characterized in that in the step a3
In, the cooccurrence relation that the OT parameter is used to reflect between two articles is strong and weak, the OT calculation formula between two articles are as follows:
In above formula,For judging in favorites list that whether article appears in user i,Calculation formula are as follows:
The PR parameter is used to filter out influence of the user of habit favorable comment for article cooccurrence relation, PR calculation formula are as follows:
In above formula,Indicate the article scoring set of i-th of user,Represent the preference of i-th of user
Product set.
4. the recommended method according to claim 3 based on network representation study, which is characterized in that in the step b,
Defined search strategy includes breadth First sampling policy and depth-first sampling policy, the breadth First sampling policy
Neighborhood node is restricted to be directly connected to the node of source node；The neighborhood node of the depth-first sampling policy is by distance sources
The node of node continuous sampling forms.
5. the recommended method according to claim 4 based on network representation study, which is characterized in that in the step c,
It is described according to each user node and article node and respective neighbor node, learn to obtain each user using network representation
The vector of node and article node indicates specifically: using skip-gram training, and stochastic gradient descent and negative sampling are combined,
The vector for obtaining each user node and article node indicates.
6. a kind of recommender system based on network representation study characterized by comprising
User-article co-occurrence network struction module: for constructing user-article co-occurrence net based on bipartite graph network and injection shadow figure
Network；
Search strategy definition module: for being directed to the user-article co-occurrence net definitions search strategy, each user's section is obtained
The neighbor node of point and article node；
Network representation study module: for using net according to each user node and article node and respective neighbor node
Network indicates that study obtains each user node and the vector of article node and indicates；
Vector calculation module: it for being indicated according to the vector of each user node and article node, is calculated by vector
Recommend maximally related article to each user to the maximally related article node of each user node, and according to calculated result.
7. the recommender system according to claim 6 based on network representation study, which is characterized in that the user-article
Co-occurrence network struction module includes:
Bipartite graph construction unit: for using scoring of the bipartite graph network storage user to article, building user-article two divides
Figure；
Article co-occurrence network struction unit: for using the cooccurrence relation between injection shadow figure storage article, article co-occurrence is constructed
Network；
User-article co-occurrence network struction unit: user-article bipartite graph and article co-occurrence network struction user-article are based on
Co-occurrence network, and OT parameter is set and PR parameter filters useless cooccurrence relation in the user-article co-occurrence network.
8. the recommender system according to claim 7 based on network representation study, which is characterized in that the OT parameter is used for
Reflect that the cooccurrence relation between two articles is strong and weak, the OT calculation formula between two articles are as follows:
In above formula,For judging in favorites list that whether article appears in user i,Calculation formula are as follows:
The PR parameter is used to filter out influence of the user of habit favorable comment for article cooccurrence relation, PR calculation formula are as follows:
In above formula,Indicate the article scoring set of i-th of user,Represent the preference of i-th of user
Product set.
9. the recommender system according to claim 8 based on network representation study, which is characterized in that described search strategy is fixed
The search strategy of adopted module definition includes breadth First sampling policy and depth-first sampling policy, and the breadth First samples plan
Neighborhood node slightly is restricted to be directly connected to the node of source node；The neighborhood node of the depth-first sampling policy by away from
Node composition from source node continuous sampling.
10. the recommender system according to claim 9 based on network representation study, which is characterized in that the network representation
Study module learns to obtain each according to each user node and article node and respective neighbor node using network representation
The vector of user node and article node indicates specifically: using skip-gram training, and stochastic gradient descent is combined to adopt with negative
Sample, the vector for obtaining each user node and article node indicate.
11. a kind of electronic equipment, comprising:
At least one processor；And
The memory being connect at least one described processor communication；Wherein,
The memory is stored with the instruction that can be executed by one processor, and described instruction is by least one described processor
It executes, so that at least one described processor is able to carry out above-mentioned 1 to 5 described in any item recommendations based on network representation study
The following operation of method:
Step a: user-article co-occurrence network is constructed based on bipartite graph network and injection shadow figure；
Step b: being directed to the user-article co-occurrence net definitions search strategy, obtains each user node and article node
Neighbor node；
Step c: according to each user node and article node and respective neighbor node, learn to obtain using network representation every
The vector of a user node and article node indicates；
Step d: indicating according to the vector of each user node and article node, and each user is calculated by vector and saves
The maximally related article node of point, and maximally related article is recommended to each user according to calculated result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810942135.8A CN109241412B (en) | 2018-08-17 | 2018-08-17 | Recommendation method and system based on network representation learning and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810942135.8A CN109241412B (en) | 2018-08-17 | 2018-08-17 | Recommendation method and system based on network representation learning and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109241412A true CN109241412A (en) | 2019-01-18 |
| CN109241412B CN109241412B (en) | 2020-12-11 |
Family
ID=65071536
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810942135.8A Active CN109241412B (en) | 2018-08-17 | 2018-08-17 | Recommendation method and system based on network representation learning and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109241412B (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109918543A (en) * | 2019-02-26 | 2019-06-21 | 华中科技大学 | The link prediction method of node is directed in a kind of figure stream |
| CN110310185A (en) * | 2019-07-10 | 2019-10-08 | 云南大学 | Popular and novelty Method of Commodity Recommendation based on weighting bigraph (bipartite graph) |
| CN110335112A (en) * | 2019-06-18 | 2019-10-15 | 中国平安财产保险股份有限公司 | A kind of Products Show method, apparatus and computer storage medium |
| CN110879864A (en) * | 2019-10-29 | 2020-03-13 | 南京大学 | Context recommendation method based on graph neural network and attention mechanism |
| CN111368205A (en) * | 2020-03-09 | 2020-07-03 | 腾讯科技（深圳）有限公司 | Data recommendation method and device, computer equipment and storage medium |
| CN111797663A (en) * | 2019-08-02 | 2020-10-20 | 北京京东尚科信息技术有限公司 | Collocation scene recognition method and device |
| CN112131480A (en) * | 2020-09-30 | 2020-12-25 | 中国海洋大学 | Personalized commodity recommendation method and system based on multilayer heterogeneous attribute network representation learning |
| CN112464647A (en) * | 2020-11-23 | 2021-03-09 | 北京智源人工智能研究院 | Recommendation system-oriented negative sampling method and device and electronic equipment |
| CN112784171A (en) * | 2021-01-21 | 2021-05-11 | 重庆邮电大学 | Movie recommendation method based on context typicality |
| CN113255712A (en) * | 2020-02-12 | 2021-08-13 | 华为技术有限公司 | Recommendation method and device |
| CN113360764A (en) * | 2021-06-23 | 2021-09-07 | 甄付(上海)网络科技有限公司 | Social network construction method and system based on life service consumption scene |
| CN113763014A (en) * | 2021-01-05 | 2021-12-07 | 北京沃东天骏信息技术有限公司 | Article co-occurrence relation determining method and device and judgment model obtaining method and device |
| CN113888138A (en) * | 2021-10-27 | 2022-01-04 | 重庆邮电大学 | Project management method based on block chain and network representation learning recommendation |
| EP4088197A4 (en) * | 2020-02-12 | 2023-01-11 | Huawei Technologies Co., Ltd. | Recommender system using bayesian graph convolution networks |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103116639A (en) * | 2013-02-20 | 2013-05-22 | 新浪网技术(中国)有限公司 | Item recommendation method and system based on user-item bipartite model |
| US8660970B1 (en) * | 2009-04-23 | 2014-02-25 | The Boeing Company | Passive learning and autonomously interactive system for leveraging user knowledge in networked environments |
| CN106897254A (en) * | 2015-12-18 | 2017-06-27 | 清华大学 | A kind of network representation learning method |
| CN107391623A (en) * | 2017-07-07 | 2017-11-24 | 中国人民大学 | A kind of knowledge mapping embedding grammar for merging more background knowledges |
| CN108153912A (en) * | 2018-01-24 | 2018-06-12 | 北京理工大学 | A kind of knowledge based represents the Harmonious Matrix decomposition method of study |
-
2018
- 2018-08-17 CN CN201810942135.8A patent/CN109241412B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8660970B1 (en) * | 2009-04-23 | 2014-02-25 | The Boeing Company | Passive learning and autonomously interactive system for leveraging user knowledge in networked environments |
| CN103116639A (en) * | 2013-02-20 | 2013-05-22 | 新浪网技术(中国)有限公司 | Item recommendation method and system based on user-item bipartite model |
| CN106897254A (en) * | 2015-12-18 | 2017-06-27 | 清华大学 | A kind of network representation learning method |
| CN107391623A (en) * | 2017-07-07 | 2017-11-24 | 中国人民大学 | A kind of knowledge mapping embedding grammar for merging more background knowledges |
| CN108153912A (en) * | 2018-01-24 | 2018-06-12 | 北京理工大学 | A kind of knowledge based represents the Harmonious Matrix decomposition method of study |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109918543A (en) * | 2019-02-26 | 2019-06-21 | 华中科技大学 | The link prediction method of node is directed in a kind of figure stream |
| CN110335112A (en) * | 2019-06-18 | 2019-10-15 | 中国平安财产保险股份有限公司 | A kind of Products Show method, apparatus and computer storage medium |
| CN110310185A (en) * | 2019-07-10 | 2019-10-08 | 云南大学 | Popular and novelty Method of Commodity Recommendation based on weighting bigraph (bipartite graph) |
| CN110310185B (en) * | 2019-07-10 | 2022-02-18 | 云南大学 | Weighted bipartite graph-based popular and novel commodity recommendation method |
| CN111797663A (en) * | 2019-08-02 | 2020-10-20 | 北京京东尚科信息技术有限公司 | Collocation scene recognition method and device |
| CN110879864A (en) * | 2019-10-29 | 2020-03-13 | 南京大学 | Context recommendation method based on graph neural network and attention mechanism |
| CN110879864B (en) * | 2019-10-29 | 2022-06-07 | 南京大学 | Context recommendation method based on graph neural network and attention mechanism |
| CN113255712A (en) * | 2020-02-12 | 2021-08-13 | 华为技术有限公司 | Recommendation method and device |
| EP4088197A4 (en) * | 2020-02-12 | 2023-01-11 | Huawei Technologies Co., Ltd. | Recommender system using bayesian graph convolution networks |
| CN111368205A (en) * | 2020-03-09 | 2020-07-03 | 腾讯科技（深圳）有限公司 | Data recommendation method and device, computer equipment and storage medium |
| CN112131480A (en) * | 2020-09-30 | 2020-12-25 | 中国海洋大学 | Personalized commodity recommendation method and system based on multilayer heterogeneous attribute network representation learning |
| CN112131480B (en) * | 2020-09-30 | 2022-06-10 | 中国海洋大学 | Personalized commodity recommendation method and system based on multilayer heterogeneous attribute network representation learning |
| CN112464647A (en) * | 2020-11-23 | 2021-03-09 | 北京智源人工智能研究院 | Recommendation system-oriented negative sampling method and device and electronic equipment |
| CN113763014A (en) * | 2021-01-05 | 2021-12-07 | 北京沃东天骏信息技术有限公司 | Article co-occurrence relation determining method and device and judgment model obtaining method and device |
| CN112784171A (en) * | 2021-01-21 | 2021-05-11 | 重庆邮电大学 | Movie recommendation method based on context typicality |
| CN113360764A (en) * | 2021-06-23 | 2021-09-07 | 甄付(上海)网络科技有限公司 | Social network construction method and system based on life service consumption scene |
| CN113888138A (en) * | 2021-10-27 | 2022-01-04 | 重庆邮电大学 | Project management method based on block chain and network representation learning recommendation |
| CN113888138B (en) * | 2021-10-27 | 2024-05-14 | 重庆邮电大学 | Project management method based on blockchain and network representation learning recommendation |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109241412B (en) | 2020-12-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109241412A (en) | A kind of recommended method, system and electronic equipment based on network representation study | |
| US10958748B2 (en) | Resource push method and apparatus | |
| CN111177569B (en) | Recommendation processing method, device and equipment based on artificial intelligence | |
| US20210397980A1 (en) | Information recommendation method and apparatus, electronic device, and readable storage medium | |
| EP3917383A1 (en) | Systems and methods for organizing and finding data | |
| US10535106B2 (en) | Selecting user posts related to trending topics on online social networks | |
| WO2019047790A1 (en) | Method and system for generating combined features of machine learning samples | |
| Xu et al. | Improving user recommendation by extracting social topics and interest topics of users in uni-directional social networks | |
| WO2021135562A1 (en) | Feature validity evaluation method and apparatus, and electronic device and storage medium | |
| CN110532479A (en) | A kind of information recommendation method, device and equipment | |
| WO2014056408A1 (en) | Information recommending method, device and server | |
| US20150193535A1 (en) | Identifying influencers for topics in social media | |
| WO2020238502A1 (en) | Article recommendation method and apparatus, electronic device and storage medium | |
| WO2023108980A1 (en) | Information push method and device based on text adversarial sample | |
| CN112100221B (en) | Information recommendation method and device, recommendation server and storage medium | |
| CN113449204B (en) | Social event classification method and device based on local aggregation graph attention network | |
| Cai et al. | Neighborhood-enhanced transfer learning for one-class collaborative filtering | |
| CN105869058B (en) | A kind of method that multilayer latent variable model user portrait extracts | |
| CN113312480A (en) | Scientific and technological thesis level multi-label classification method and device based on graph convolution network | |
| Zhang et al. | An approach of service discovery based on service goal clustering | |
| Dron et al. | On the design of collective applications | |
| Layton | Learning data mining with Python | |
| Ali et al. | Big social data as a service (BSDaaS): a service composition framework for social media analysis | |
| CN112507185B (en) | User portrait determination method and device | |
| Yan et al. | Tackling the achilles heel of social networks: Influence propagation based language model smoothing |
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 |