CN109816101A - A kind of session sequence of recommendation method and system based on figure convolutional neural networks - Google Patents
A kind of session sequence of recommendation method and system based on figure convolutional neural networks Download PDFInfo
- Publication number
- CN109816101A CN109816101A CN201910098574.XA CN201910098574A CN109816101A CN 109816101 A CN109816101 A CN 109816101A CN 201910098574 A CN201910098574 A CN 201910098574A CN 109816101 A CN109816101 A CN 109816101A
- Authority
- CN
- China
- Prior art keywords
- vector
- session
- user
- sequence
- session sequence
- 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.)
- Pending
Links
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The session sequence of recommendation method and system based on figure convolutional neural networks that the present disclosure proposes a kind of are a digraph for each session sequence construct, and for each digraph, being input to figure convolutional neural networks to obtain the implicit of all nodes indicates vector；Implicit expression vector based on obtained node, global preferences vector sum, which is generated, using soft attention mechanism network locally clicks preference vector, wherein global preferences vector sum is locally clicked preference vector and is made of the implicit expression vector of node, later, each session sequence is expressed as the combination that user partial in global preferences vector and the session clicks preference vector；For each session sequence, each project of combined prediction that user partial clicks preference vector in global preferences vector calculated and the session becomes the probability clicked next time.It is indicated by introducing global and local implicit vector, eliminates some noises in native vector space, obtain more accurate prediction effect.
Description
Technical field
This disclosure relates to computer processing technical field, more particularly to a kind of session sequence based on figure convolutional neural networks
Column recommended method and system.
Background technique
As internet can obtain the rapid growth of information, information overload is denounced by user always.Recommender system can be with
User is helped quickly to be accurately obtained wanted information.Existing most of recommender system hypothesis can constantly record user
Activity.However, the personal identification of user may be unknown in many cases, for website end, and only have currently just
It is known for website in the user behavior history of the ession for telecommunication of progress.Therefore, in a session to limited behavior into
It is significant that row, which models and correspondingly carries out commending contents,.On the contrary, dependent on sufficient user-project interaction biography
System recommended method is easy to produce the recommendation results of inaccuracy in this case.
Recently, next row that a few thing based on Markov chain passes through the previous action prediction user of user
For.Although achieving good results, because the independence assumption of these methods is excessively strong, forecasting accuracy is limited
Further increase.In recent years, Recognition with Recurrent Neural Network (RNN) is applied to the recommender system of dialogue-based sequence by most of researchs
And obtain effect outstanding.
Although the above method achieve it is satisfactory as a result, they still have some limitations.
Firstly, since often without enough user behaviors in a session, and in dialogue-based recommender system
Words are usually anonymous and large number of, therefore session is clicked user behavior involved in sequence and is but limited.Therefore difficult
Accurately to estimate that the implicit of each user indicates vector from each session.
Secondly, the conversion, connection relationship between the project of user's click sequence are critically important.It can be in dialogue-based recommendation
In regard as and influence the local factor that user clicks, but these methods only only account for the unidirectional conversion between adopting consecutive click chemical reaction project,
Have ignored the transformational relation between context (other of user click item).
Summary of the invention
In order to solve the deficiencies in the prior art, embodiment of the present disclosure provides a kind of meeting based on figure convolutional neural networks
Sequence of recommendation method is talked about, carries out being modeled as figure by clicking project to user, user in sequence has been fully taken into account and has clicked project
Between relationship and can accurately predict that user's is next so as to accurately estimate that user clicks the expression of project
Click action.
To achieve the goals above, the application uses following technical scheme:
A kind of session sequence of recommendation method based on figure convolutional neural networks, comprising:
The history for obtaining user clicks item data, and is a digraph for each session sequence construct, for each
Digraph, being input to figure convolutional neural networks to obtain the implicit of all nodes involved in digraph indicates vector；
Implicit expression vector based on obtained node generates the drawn game of global preferences vector using soft attention mechanism network
Portion clicks preference vector, and wherein global preferences vector sum is locally clicked preference vector and is made of the implicit expression vector of node,
Later, each session sequence is expressed as the combination that user partial in global preferences vector and the session clicks preference vector；
For each session sequence, in global preferences vector calculated and the session user partial click preference to
The each project of the combined prediction of amount becomes the probability clicked next time.
As the further technical solution of the disclosure, user in each node on behalf session sequence in the digraph
One click item, in figure while show user's adopting consecutive click chemical reaction this while two nodes connecting.
As the further technical solution of the disclosure, after constructing digraph, the weight on side in digraph is subjected to normalizing
Change operation, side right value be calculated as while frequency of occurrence divided by this while start node out-degree.
As the further technical solution of the disclosure, in global preferences vector and the session user partial click preference to
The combination insertion of amount is used as the expression of session sequence.
As the further technical solution of the disclosure, after generating the insertion expression of each session, by being implied table
Show and click item multiplied by each candidate, obtains a recommendation scores, then recommendation scores are normalized by function, select
Divide highest item as recommendation foundation.
As the further technical solution of the disclosure, processing of the figure convolutional neural networks to digraph, specifically: will own
The user being related in session sequence clicks Xiang Jun and is embedded in a unified implicit representation space, is formed as based on user's point
Hit the implicit expression vector of project, finally, each session sequence be expressed as the implicit expression from clicking item in the digraph to
Measure the vector of composition.
As the further technical solution of the disclosure, vector is indicated based on implicit, and the part click for firstly generating user is inclined
Good vector, in order to predict the next item down click, then influence it is maximum should be user current click；
Later, consider to construct the project being related in entire user conversation sequence using attention mechanism user's overall situation
Preference vector；
Next, converting to above-mentioned two vector, group is combined into the mixing expression vector of user conversation sequence.
As the further technical solution of the disclosure, the implicit expression vector sum user of item is clicked by obtaining each user
After the mixing expression vector of session sequence, dot product operation is carried out to all candidate items, recommendation scores are obtained, then to all scores
Operation is normalized, the click of user's the next item down is predicted.
As the further technical solution of the disclosure, when prediction, defining loss function be that prediction is clicked and the friendship of true value
Entropy is pitched, parameter is learnt using algorithm.
Embodiment of the disclosure also discloses a kind of session sequence of recommendation system based on figure convolutional neural networks, packet
It includes:
Knot vector indicates unit, is configured as: the history for obtaining user clicks item data, and is directed to each session sequence
It is configured to a digraph, for each digraph, is input to figure convolutional neural networks to obtain involved in digraph
The implicit expression vector of all nodes；
Session sequence vector indicates unit, and be configured as: the implicit expression vector based on obtained node uses soft attention
Power mechanism network generate global preferences vector sum locally click preference vector, wherein global preferences vector sum locally click preference to
Amount is made of the implicit expression vector of node, later, each session sequence is expressed as global preferences vector and the session
Middle user partial clicks the combination of preference vector；
Project clicks predicting unit, for each session sequence, in global preferences vector calculated and the session
The each project of combined prediction that user partial clicks preference vector becomes the probability clicked next time.
Embodiment of the disclosure also discloses a kind of session sequence of recommendation system based on figure convolutional neural networks, packet
It includes:
Acquisition unit is configured as: the history for acquiring user clicks item data, and forms session sequence；
Server, the server are configured as: the history for obtaining user clicks item data, and is directed to each session sequence
It is configured to a digraph, for each digraph, is input to figure convolutional neural networks to obtain involved in digraph
The implicit expression vector of all nodes；
Implicit expression vector based on obtained node generates the drawn game of global preferences vector using soft attention mechanism network
Portion clicks preference vector, and wherein global preferences vector sum is locally clicked preference vector and is made of the implicit expression vector of node,
Later, each session sequence is expressed as the combination that user partial in global preferences vector and the session clicks preference vector；
For each session sequence, in global preferences vector calculated and the session user partial click preference to
The each project of the combined prediction of amount becomes the probability clicked next time；
Display unit, the display unit are configured as: display server, which is formed by each project, becomes point next time
The probability hit.
Embodiment of the disclosure also discloses a kind of computer equipment, comprising: processor, network interface and memory,
The memory is stored with executable program code, and the network interface is used for messaging, institute by the control of the processor
Processor is stated for calling the executable program code, executes a kind of session based on figure convolutional neural networks described above
Sequence of recommendation method.
A kind of storage medium is stored with instruction in the storage medium, when run on a computer, so that computer
Execute a kind of session sequence of recommendation method based on figure convolutional neural networks described above.
Compared with prior art, the beneficial effect of the disclosure is:
The technical solution of the disclosure solves other models cannot model connection relationship between sequence very well, cannot learn very well
The problems such as expression of habit project and Sparse and cold start-up, it can be suitably used for the huge and complicated quotient of current sessions sequence data amount
Industry business scenario, such as shopping cart is recommended, product mix and ad click rate predict multiple fields.
The technical solution of the disclosure is directed to the expression for obtaining the project being related in accurate session sequence, and can apply
In the situation of high-volume conversation sequence.It is indicated, is eliminated in native vector space by introducing global and local implicit vector
Some noises, obtain more accurate prediction effect.
The technical solution of the disclosure proposes session Series Modeling to be digraph, can capture complicated conversion between node and closes
System, in this, as the foundation of recommendation, can be improved the accuracy rate of prediction.
Detailed description of the invention
The Figure of description for constituting a part of this disclosure is used to provide further understanding of the disclosure, and the disclosure is shown
Meaning property embodiment and its explanation do not constitute the improper restriction to the disclosure for explaining the disclosure.
Fig. 1 is the work flow diagram of embodiment of the present disclosure；
Fig. 2 (a)-Fig. 2 (b) is test result schematic diagram of the embodiment of the present disclosure on data set.
Specific embodiment
It is noted that following detailed description is all illustrative, it is intended to provide further instruction to the disclosure.Unless another
It indicates, all technical and scientific terms used herein has usual with disclosure person of an ordinary skill in the technical field
The identical meanings of understanding.
It should be noted that term used herein above is merely to describe specific embodiment, and be not intended to restricted root
According to the illustrative embodiments of the disclosure.As used herein, unless the context clearly indicates otherwise, otherwise singular
Also it is intended to include plural form, additionally, it should be understood that, when in the present specification using term "comprising" and/or " packet
Include " when, indicate existing characteristics, step, operation, device, component and/or their combination.
In a kind of typical embodiment of the disclosure, referring to figure 1, provide a kind of based on picture scroll product nerve net
The session sequence of recommendation method of network, including specific steps are as follows:
Step S1. determines the session sequence of user according to cookies information of the user on website, and according in sequence
Each timestamp is ranked up.It is a digraph according to each individual user conversation sequence construct.Have for each
Xiang Tu is input to figure convolutional neural networks successively to obtain the implicit vector of all nodes involved in each figure and indicate.
Step S2., which implies vector based on the obtained node of step S2, to be indicated, is generated using soft attention mechanism network global
With the implicit expression of local session sequence, wherein global preferences and partial vector are made of the implicit vector of node.Later, will
Each session sequence is expressed as the combination of user's current interest in global preferences and the session.
Step S3. is become next time each session sequence by each project of vector forecasting calculated in step S2
The probability of click.
Embodiment of the present disclosure propose the session sequence of recommendation method based on figure convolutional neural networks, by by user with
The interaction item of system is (depending on different recommendation scenes.In concrete scene, it may be possible to browsing, click plus the shopping of user
The behaviors such as vehicle, collection, purchase) and session sequence be embedded into implicit vector space, to each item and session Sequence Learning one d dimension
Feature vector vi, si.By being digraph by user conversation Series Modeling, the implicit vector that the disclosure learns is more accurate.
Specifically, being a digraph by each session Series Modeling in step sl.In this digraph, often
The click item of user in one node on behalf session sequence.Each edge represents what the user in session sequence s successively clicked
Two projects, such as: the side (v in figure1, v2) show user's adopting consecutive click chemical reaction v1And v2。
In view of user may click duplicate project, cause to be likely to occur multiple duplicate click items in sequence.At this moment
Embodiment of the present disclosure considers the weight on side operation is normalized.Side right value be calculated as while frequency of occurrence divided by this while
The out-degree of start node.Out-degree is calculated as using the node as the number on the side of start node.
Later, the user being related in all session sequences is clicked Xiang Jun by figure convolutional network by embodiment of the present disclosure
It is embedded into a unified d to tie up in implicit representation space.The implicit expression vector of project, each session sequence are clicked based on user
Column can be expressed as the vector being made of the implicit expression vector for clicking item in the sequence chart.
In one embodiment, each session sequence can be expressed as by the implicit expression of the project in the oriented sequence chart
The vector of vector composition, specific steps are as follows:
The sequence chart g constituted for user conversation sequence ss, calculate the adjacency matrix with itself circuitWith
Vertex Degree matrixIt is input in picture scroll product neural net layer, obtains each implicit expression matrix G for clicking item:
Wherein, each implicit expression vector v for clicking item is a line of G, θ: to need the weight matrix learnt.It needs to infuse
Meaning is eigenmatrix HsSetting it is more flexible, the figure that constitutes without additional information of item, H are clicked for usersIt can be set to
I。
Specifically, in step s 2, generating the implicit of session sequence based on the implicit expression for obtaining clicking item in step S1
Vector is expressed, each session sequence is expressed as to the implicit expression vector of the node as involved in the session.In order to be better anticipated
The click next time of user, in conjunction with long-term (overall situation) preference of the user embodied in session sequence and current (part) interest, and
This combination insertion is used as to the expression of session sequence.
In one embodiment, the implicit expression vector based on the implicit expression generation session sequence by obtaining clicking item,
Specifically: click preference vector s in the part for firstly generating userl, in order to predict that the next item down is clicked, then influence maximum should be
The current click of user.
Later, consider to construct the project being related in entire user conversation sequence using attention mechanism user's overall situation
Preference vector sg。
Next, converting to above-mentioned two vector, group is combined into the mixing expression vector s of user conversation sequenceh.Specifically
Mapping mode linear transformation and nonlinear transformation (such as using multi-layer perception (MLP)) can be considered.
Specifically, being indicated by being implied multiplied by each time after step S2 generates the insertion expression of each session
Item is hit in reconnaissance, obtains a recommendation scores.Recommendation scores are normalized by softmax function again, select highest scoring
Item as recommend foundation.
In one embodiment, the click of user's the next item down is predicted.Each user's point is respectively obtained through the above steps
After hitting the implicit expression of item and the implicit expression vector of user conversation sequence, dot product operation is carried out to all candidate items, is pushed away
Recommend score.Next softmax normalization operation is carried out to all scores, is convenient for subsequent prediction and recommendation.
In order to train the neural network model, the loss function of Definition Model is the cross entropy of prediction click and true value
(cross entropy), then learns model parameter using Back-Propagation algorithm.
This specific embodiment can carry out being modeled as figure by clicking project to user, fully taken into account in sequence and used
Relationship between family click project so as to accurately estimate that user clicks the expression of project, and can accurately be predicted to use
Next click action at family.
In the specific embodiment of the disclosure, since independent of one another, separation session sequence is modeled as graph structure
Data, and the implicit expression vector of accurate node is obtained using the neural network based on picture scroll product.In reality, the purchase of user
Object behavior often has certain connection, for example, periodically, repeatability and fixed buying behavior, this embodiment example proposes will
Session Series Modeling is figure, and can capture complicated transformational relation between node can be improved prediction in this, as the foundation of recommendation
Accuracy rate.
In the specific embodiment of the disclosure, due to the implicit expression vector independent of user, but by each meeting
Words sequence is embedded in humble implicit space.The implicit expression vector of node involved in each individual session sequence can be based only upon
To be recommended.Therefore, the quantity for needing the parameter learnt can be greatly reduced, can be improved the speed and essence of model learning
Degree.
In the specific embodiment of the disclosure, the recommender system problem of dialogue-based sequence is accurate particular for obtaining
Session sequence in the expression of project that is related to, and can apply in the situation of high-volume conversation sequence.It is global by introducing
It is indicated with the implicit vector of part, eliminates some noises in native vector space, obtain more accurate prediction effect.
In order to enable those skilled in the art can clearly understand the technical solution of the disclosure, below with reference to tool
The technical solution of the disclosure is described in detail in the embodiment and comparative example of body.
Effect to the recommender system for constructing dialogue-based sequence for a better understanding of the present invention, and the verifying present invention
Implementation result, be next illustrated by taking the recommendation of online shopping platform as an example, this example using Yoochoose and
Diginetica database.The two databases contain click steam of the user on certain electronic shopping platform, have one
Fixed representativeness.Each data set includes that a certain number of users click sequence.In order to other methods carry out it is fair compared with,
The sequence that project of the frequency of occurrence less than 5 times and length are 1 is ignored.In view of the quantity mistake of sequence in Yoochoose data
In huge, nearest 1/64 and 1/4 data are only used only.
The method that the effect of embodiment of the present disclosure is verified on the two data sets is as follows:
Step S1. is a digraph according to each individually session sequence construct.For each digraph, successively by it
Figure convolutional neural networks are input to obtain the implicit vector of all nodes involved in each figure and indicate.
Step S2., which implies vector based on the obtained node of step S2, to be indicated, is generated using soft attention mechanism network global
With the implicit expression of local session sequence, wherein global preferences and partial vector are made of the implicit vector of node.Later, will
Each session sequence is expressed as the combination of user's current interest in global preferences and the session.
Step S3. is become next time each session sequence by each project of vector forecasting calculated in step S2
The probability of click.
The disclosure is compared with following method NARM and STAMP, and current best method, using 20 He of Precision@
MRR@20 is used as evaluation index.When assessment, using the most suitable parameter of verifying Resource selection of 10% scale, and limit
Scale d=100 of implicit expression of space.
In addition, for list entries s=[vS, 1, vS, 2..., vS, n], generate following sequence and corresponding prediction label:
([vS, 1], vS, 2), ([vS, 1, vS, 2], vS, 3) ..., ([vS, 1, vS, 2..., vS, n-1], vS, n),
Wherein, [vS, 1, vS, 2..., vS, n-1] it is the sequence generated, vS, n: for corresponding prediction label.
Comparison result, referring to shown in attached drawing 2 (a)-attached drawing 2 (b).It can be seen from the figure that disclosed method with it is newest
NARM compared with STAMP, other all leading two methods of 20 two evaluation indexes of Precision@20 and MRR@.Demonstrate this
The validity of method.
One examples of implementation of the disclosure also disclose a kind of session sequence of recommendation system based on figure convolutional neural networks, packet
It includes:
Knot vector indicates unit, is configured as: the history for obtaining user clicks item data, and is directed to each session sequence
It is configured to a digraph, for each digraph, is input to figure convolutional neural networks to obtain involved in digraph
The implicit expression vector of all nodes；
Session sequence vector indicates unit, and be configured as: the implicit expression vector based on obtained node uses soft attention
Power mechanism network generate global preferences vector sum locally click preference vector, wherein global preferences vector sum locally click preference to
Amount is made of the implicit expression vector of node, later, each session sequence is expressed as global preferences vector and the session
Middle user partial clicks the combination of preference vector；
Project clicks predicting unit, for each session sequence, in global preferences vector calculated and the session
The each project of combined prediction that user partial clicks preference vector becomes the probability clicked next time.
One examples of implementation of the disclosure also disclose a kind of session sequence of recommendation system based on figure convolutional neural networks, packet
It includes:
Acquisition unit is configured as: the history of liniment user clicks item data, and forms session sequence；
Server, the server are configured as: the history for obtaining user clicks item data, and is directed to each session sequence
It is configured to a digraph, for each digraph, is input to figure convolutional neural networks to obtain involved in digraph
The implicit expression vector of all nodes；
Implicit expression vector based on obtained node generates the drawn game of global preferences vector using soft attention mechanism network
Portion clicks preference vector, and wherein global preferences vector sum is locally clicked preference vector and is made of the implicit expression vector of node,
Later, each session sequence is expressed as the combination that user partial in global preferences vector and the session clicks preference vector；
For each session sequence, in global preferences vector calculated and the session user partial click preference to
The each project of the combined prediction of amount becomes the probability clicked next time；
Display unit, the display unit are configured as: display server, which is formed by each project, becomes point next time
The probability hit.
Another embodiment of the present disclosure also discloses a kind of computer equipment, comprising: processor, network interface and storage
Device, the memory are stored with executable program code, and the network interface is used for messaging by the control of the processor,
The processor executes a kind of meeting based on figure convolutional neural networks described above for calling the executable program code
Talk about sequence of recommendation method.
A kind of storage medium of another embodiment of the present disclosure is stored with instruction in the storage medium, when it is in computer
When upper operation, so that computer executes a kind of session sequence of recommendation method based on figure convolutional neural networks described above.
It is understood that in the description of this specification, reference term " embodiment ", " another embodiment ", " other
The description of embodiment " or " first embodiment~N embodiment " etc. means specific spy described in conjunction with this embodiment or example
Sign, structure, material or feature are included at least one embodiment or example of the invention.In the present specification, to above-mentioned
The schematic representation of term may not refer to the same embodiment or example.Moreover, the specific features of description, structure, material
Person's feature can be combined in any suitable manner in any one or more of the embodiments or examples.
The foregoing is merely preferred embodiment of the present disclosure, are not limited to the disclosure, for the skill of this field
For art personnel, the disclosure can have various modifications and variations.It is all within the spirit and principle of the disclosure, it is made any to repair
Change, equivalent replacement, improvement etc., should be included within the protection scope of the disclosure.
Claims (10)
1. a kind of session sequence of recommendation method based on figure convolutional neural networks, characterized in that include:
The history for obtaining user clicks item data, and is a digraph for each session sequence construct, for each oriented
Figure, being input to figure convolutional neural networks to obtain the implicit of all nodes involved in digraph indicates vector；
Implicit expression vector based on obtained node generates global preferences vector sum partial points using soft attention mechanism network
Preference vector is hit, wherein global preferences vector sum is locally clicked preference vector and is made of the implicit expression vector of node, later,
Each session sequence is expressed as the combination that user partial in global preferences vector and the session clicks preference vector；
For each session sequence, user partial clicks preference vector in global preferences vector calculated and the session
The each project of combined prediction becomes the probability clicked next time.
2. a kind of session sequence of recommendation method based on figure convolutional neural networks as described in claim 1, characterized in that described
In digraph in each node on behalf session sequence user a click item, the side in figure shows that user's adopting consecutive click chemical reaction should
Two nodes of side connection.
3. a kind of session sequence of recommendation method based on figure convolutional neural networks as claimed in claim 2, characterized in that building
After digraph, the weight on side in digraph is normalized operation, side right value be calculated as while frequency of occurrence divided by this while
Start node out-degree.
Further, the combination insertion that user partial clicks preference vector in global preferences vector and the session is used as session sequence
The expression of column is indicated to click item multiplied by each candidate, be obtained after generating the insertion expression of each session by being implied
One recommendation scores, then recommendation scores are normalized by function, the item of highest scoring is selected as recommendation foundation.
4. a kind of session sequence of recommendation method based on figure convolutional neural networks as described in claim 1, characterized in that picture scroll
Processing of the product neural network to digraph, specifically: the user being related in all session sequences click Xiang Jun is embedded in one
In the implicit representation space of a unification, be formed as the implicit expression vector that project is clicked based on user, finally, each session sequence
List is shown as the vector being made of the implicit expression vector for clicking item in the digraph.
5. a kind of session sequence of recommendation method based on figure convolutional neural networks as described in claim 1, characterized in that be based on
Implicit to indicate vector, preference vector is clicked in the part for firstly generating user, in order to predict that the next item down is clicked, then influences maximum answer
When the current click for being user；
Later, consider to construct user's global preferences to the project being related in entire user conversation sequence using attention mechanism
Vector；
Next, converting to above-mentioned two vector, group is combined into the mixing expression vector of user conversation sequence.
6. a kind of session sequence of recommendation method based on figure convolutional neural networks as claimed in claim 5, characterized in that pass through
After obtaining the mixing expression vector for the implicit expression vector sum user conversation sequence that each user clicks item, to all candidate items into
The operation of row dot product, obtains recommendation scores, operation then is normalized to all scores, clicks and is carried out in advance to user's the next item down
It surveys.
7. a kind of session sequence of recommendation method based on figure convolutional neural networks as described in claim 1, characterized in that prediction
When, the cross entropy that loss function is prediction click and true value is defined, parameter is learnt using algorithm.
8. a kind of session sequence of recommendation system based on figure convolutional neural networks, characterized in that include:
Knot vector indicates unit, is configured as: the history for obtaining user clicks item data, and is directed to each session sequence construct
Is input to by figure convolutional neural networks and is owned involved in digraph to obtain for each digraph for a digraph
The implicit expression vector of node；
Session sequence vector indicates unit, and be configured as: the implicit expression vector based on obtained node uses soft attention machine
Network processed generates global preferences vector sum and locally clicks preference vector, and wherein it is equal locally to click preference vector for global preferences vector sum
It is made of the implicit expression vector of node, later, each session sequence is expressed as using in global preferences vector and the session
Locally click the combination of preference vector in family；
Project clicks predicting unit, for each session sequence, user in global preferences vector calculated and the session
The each project of combined prediction that preference vector is clicked in part becomes the probability clicked next time.
9. a kind of storage medium, it is stored with instruction in the storage medium, when run on a computer, so that computer is held
A kind of any session sequence of recommendation method based on figure convolutional neural networks of row the claims 1-7.
10. a kind of computer equipment, comprising: processor, network interface and memory, the memory are stored with executable program
Code, the network interface are used for messaging by the control of the processor, characterized in that the processor is for calling institute
Executable program code is stated, a kind of any session sequence based on figure convolutional neural networks of the claims 1-7 is executed
Column recommended method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910098574.XA CN109816101A (en) | 2019-01-31 | 2019-01-31 | A kind of session sequence of recommendation method and system based on figure convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910098574.XA CN109816101A (en) | 2019-01-31 | 2019-01-31 | A kind of session sequence of recommendation method and system based on figure convolutional neural networks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109816101A true CN109816101A (en) | 2019-05-28 |
Family
ID=66606133
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910098574.XA Pending CN109816101A (en) | 2019-01-31 | 2019-01-31 | A kind of session sequence of recommendation method and system based on figure convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109816101A (en) |
Cited By (27)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110309360A (en) * | 2019-06-13 | 2019-10-08 | 山东大学 | A kind of the topic label personalized recommendation method and system of short-sighted frequency |
| CN110490717A (en) * | 2019-09-05 | 2019-11-22 | 齐鲁工业大学 | Method of Commodity Recommendation and system based on user conversation and figure convolutional neural networks |
| CN110543935A (en) * | 2019-08-15 | 2019-12-06 | 阿里巴巴集团控股有限公司 | Method and device for processing interactive sequence data |
| CN110569437A (en) * | 2019-09-05 | 2019-12-13 | 腾讯科技（深圳）有限公司 | click probability prediction and page content recommendation methods and devices |
| CN110796313A (en) * | 2019-11-01 | 2020-02-14 | 北京理工大学 | Session recommendation method based on weighted graph volume and item attraction model |
| CN111080400A (en) * | 2019-11-25 | 2020-04-28 | 中山大学 | Commodity recommendation method and system based on gate control graph convolution network and storage medium |
| CN111368203A (en) * | 2020-03-09 | 2020-07-03 | 电子科技大学 | News recommendation method and system based on graph neural network |
| CN111582443A (en) * | 2020-04-22 | 2020-08-25 | 成都信息工程大学 | Recommendation method based on Mask mechanism and level attention mechanism |
| CN111667067A (en) * | 2020-05-28 | 2020-09-15 | 平安医疗健康管理股份有限公司 | Recommendation method and device based on graph neural network and computer equipment |
| CN111931076A (en) * | 2020-09-22 | 2020-11-13 | 平安国际智慧城市科技股份有限公司 | Method and device for carrying out relationship recommendation based on authorized directed graph and computer equipment |
| CN111949865A (en) * | 2020-08-10 | 2020-11-17 | 杭州电子科技大学 | Interest point recommendation method based on graph neural network and user long-term and short-term preference |
| CN112150210A (en) * | 2020-06-19 | 2020-12-29 | 南京理工大学 | Improved neural network recommendation method and system based on GGNN (global warming network) |
| CN112364976A (en) * | 2020-10-14 | 2021-02-12 | 南开大学 | User preference prediction method based on session recommendation system |
| WO2021027256A1 (en) * | 2019-08-15 | 2021-02-18 | 创新先进技术有限公司 | Method and apparatus for processing interactive sequence data |
| CN112381581A (en) * | 2020-11-17 | 2021-02-19 | 东华理工大学 | Advertisement click rate estimation method based on improved Transformer |
| US10936950B1 (en) | 2019-08-15 | 2021-03-02 | Advanced New Technologies Co., Ltd. | Processing sequential interaction data |
| CN112528161A (en) * | 2021-02-07 | 2021-03-19 | 电子科技大学 | Conversation recommendation method based on item click sequence optimization |
| CN112733018A (en) * | 2020-12-31 | 2021-04-30 | 哈尔滨工程大学 | Session recommendation method based on graph neural network GNN and multi-task learning |
| WO2021114590A1 (en) * | 2019-12-09 | 2021-06-17 | 北京百度网讯科技有限公司 | Session recommendation method and apparatus, and electronic device |
| CN113239147A (en) * | 2021-05-12 | 2021-08-10 | 平安科技（深圳）有限公司 | Intelligent conversation method, system and medium based on graph neural network |
| CN113326425A (en) * | 2021-04-20 | 2021-08-31 | 中国电子科技集团公司第五十四研究所 | Session recommendation method and system based on structure and semantic attention stacking |
| CN113449201A (en) * | 2021-06-22 | 2021-09-28 | 上海明略人工智能(集团)有限公司 | Cross-session recommendation method, system, storage medium and electronic device |
| CN113487856A (en) * | 2021-06-04 | 2021-10-08 | 兰州理工大学 | Traffic flow combination prediction model based on graph convolution network and attention mechanism |
| CN113641811A (en) * | 2021-08-19 | 2021-11-12 | 中山大学 | Session recommendation method, system, device and storage medium for promoting purchasing behavior |
| EP3901788A3 (en) * | 2020-09-21 | 2022-03-09 | Beijing Baidu Netcom Science And Technology Co. Ltd. | Conversation-based recommending method, conversation-based recommending apparatus, and device |
| WO2023197910A1 (en) * | 2022-04-12 | 2023-10-19 | 华为技术有限公司 | User behavior prediction method and related device thereof |
| CN111667067B (en) * | 2020-05-28 | 2024-10-29 | 深圳平安医疗健康科技服务有限公司 | Recommendation method and device based on graph neural network and computer equipment |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102467728A (en) * | 2010-11-09 | 2012-05-23 | 上海悦易网络信息技术有限公司 | Multi-party transaction system and transaction method |
| CN106603387A (en) * | 2016-12-20 | 2017-04-26 | 西南石油大学 | Microblog forwarding path prediction method and system based on microblog forwarding relationship |
| CN107341611A (en) * | 2017-07-06 | 2017-11-10 | 浙江大学 | A kind of operation flow based on convolutional neural networks recommends method |
| CN109002488A (en) * | 2018-06-26 | 2018-12-14 | 北京邮电大学 | A kind of recommended models training method and device based on first path context |
-
2019
- 2019-01-31 CN CN201910098574.XA patent/CN109816101A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102467728A (en) * | 2010-11-09 | 2012-05-23 | 上海悦易网络信息技术有限公司 | Multi-party transaction system and transaction method |
| CN106603387A (en) * | 2016-12-20 | 2017-04-26 | 西南石油大学 | Microblog forwarding path prediction method and system based on microblog forwarding relationship |
| CN107341611A (en) * | 2017-07-06 | 2017-11-10 | 浙江大学 | A kind of operation flow based on convolutional neural networks recommends method |
| CN109002488A (en) * | 2018-06-26 | 2018-12-14 | 北京邮电大学 | A kind of recommended models training method and device based on first path context |
Non-Patent Citations (1)
| Title |
|---|
| SHU WU ET AL.: "Session-based Recommendation with Graph Neural Networks", 《ARXIV》 * |
Cited By (39)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110309360B (en) * | 2019-06-13 | 2021-09-28 | 山东大学 | Short video label labeling method and system |
| CN110309360A (en) * | 2019-06-13 | 2019-10-08 | 山东大学 | A kind of the topic label personalized recommendation method and system of short-sighted frequency |
| CN110543935B (en) * | 2019-08-15 | 2023-06-20 | 创新先进技术有限公司 | Method and device for processing interactive sequence data |
| CN110543935A (en) * | 2019-08-15 | 2019-12-06 | 阿里巴巴集团控股有限公司 | Method and device for processing interactive sequence data |
| US10936950B1 (en) | 2019-08-15 | 2021-03-02 | Advanced New Technologies Co., Ltd. | Processing sequential interaction data |
| US11636341B2 (en) | 2019-08-15 | 2023-04-25 | Advanced New Technologies Co., Ltd. | Processing sequential interaction data |
| WO2021027256A1 (en) * | 2019-08-15 | 2021-02-18 | 创新先进技术有限公司 | Method and apparatus for processing interactive sequence data |
| CN110490717A (en) * | 2019-09-05 | 2019-11-22 | 齐鲁工业大学 | Method of Commodity Recommendation and system based on user conversation and figure convolutional neural networks |
| CN110569437A (en) * | 2019-09-05 | 2019-12-13 | 腾讯科技（深圳）有限公司 | click probability prediction and page content recommendation methods and devices |
| CN110569437B (en) * | 2019-09-05 | 2022-03-04 | 腾讯科技（深圳）有限公司 | Click probability prediction and page content recommendation methods and devices |
| CN110796313A (en) * | 2019-11-01 | 2020-02-14 | 北京理工大学 | Session recommendation method based on weighted graph volume and item attraction model |
| CN110796313B (en) * | 2019-11-01 | 2022-05-31 | 北京理工大学 | Session recommendation method based on weighted graph volume and item attraction model |
| CN111080400A (en) * | 2019-11-25 | 2020-04-28 | 中山大学 | Commodity recommendation method and system based on gate control graph convolution network and storage medium |
| CN111080400B (en) * | 2019-11-25 | 2023-04-18 | 中山大学 | Commodity recommendation method and system based on gate control graph convolution network and storage medium |
| WO2021114590A1 (en) * | 2019-12-09 | 2021-06-17 | 北京百度网讯科技有限公司 | Session recommendation method and apparatus, and electronic device |
| JP7074964B2 (en) | 2019-12-09 | 2022-05-25 | ベイジン バイドゥ ネットコム サイエンス テクノロジー カンパニー リミテッド | Recommended session methods, equipment and devices |
| JP2022516593A (en) * | 2019-12-09 | 2022-03-01 | 北京百度網訊科技有限公司 | Recommended session methods, equipment and devices |
| CN111368203A (en) * | 2020-03-09 | 2020-07-03 | 电子科技大学 | News recommendation method and system based on graph neural network |
| CN111582443A (en) * | 2020-04-22 | 2020-08-25 | 成都信息工程大学 | Recommendation method based on Mask mechanism and level attention mechanism |
| CN111667067B (en) * | 2020-05-28 | 2024-10-29 | 深圳平安医疗健康科技服务有限公司 | Recommendation method and device based on graph neural network and computer equipment |
| CN111667067A (en) * | 2020-05-28 | 2020-09-15 | 平安医疗健康管理股份有限公司 | Recommendation method and device based on graph neural network and computer equipment |
| CN112150210A (en) * | 2020-06-19 | 2020-12-29 | 南京理工大学 | Improved neural network recommendation method and system based on GGNN (global warming network) |
| CN111949865A (en) * | 2020-08-10 | 2020-11-17 | 杭州电子科技大学 | Interest point recommendation method based on graph neural network and user long-term and short-term preference |
| EP3901788A3 (en) * | 2020-09-21 | 2022-03-09 | Beijing Baidu Netcom Science And Technology Co. Ltd. | Conversation-based recommending method, conversation-based recommending apparatus, and device |
| CN111931076A (en) * | 2020-09-22 | 2020-11-13 | 平安国际智慧城市科技股份有限公司 | Method and device for carrying out relationship recommendation based on authorized directed graph and computer equipment |
| CN111931076B (en) * | 2020-09-22 | 2021-02-09 | 平安国际智慧城市科技股份有限公司 | Method and device for carrying out relationship recommendation based on authorized directed graph and computer equipment |
| CN112364976B (en) * | 2020-10-14 | 2023-04-07 | 南开大学 | User preference prediction method based on session recommendation system |
| CN112364976A (en) * | 2020-10-14 | 2021-02-12 | 南开大学 | User preference prediction method based on session recommendation system |
| CN112381581A (en) * | 2020-11-17 | 2021-02-19 | 东华理工大学 | Advertisement click rate estimation method based on improved Transformer |
| CN112381581B (en) * | 2020-11-17 | 2022-07-08 | 东华理工大学 | Advertisement click rate estimation method based on improved Transformer |
| CN112733018A (en) * | 2020-12-31 | 2021-04-30 | 哈尔滨工程大学 | Session recommendation method based on graph neural network GNN and multi-task learning |
| CN112528161A (en) * | 2021-02-07 | 2021-03-19 | 电子科技大学 | Conversation recommendation method based on item click sequence optimization |
| CN113326425A (en) * | 2021-04-20 | 2021-08-31 | 中国电子科技集团公司第五十四研究所 | Session recommendation method and system based on structure and semantic attention stacking |
| CN113239147A (en) * | 2021-05-12 | 2021-08-10 | 平安科技（深圳）有限公司 | Intelligent conversation method, system and medium based on graph neural network |
| CN113487856A (en) * | 2021-06-04 | 2021-10-08 | 兰州理工大学 | Traffic flow combination prediction model based on graph convolution network and attention mechanism |
| CN113449201A (en) * | 2021-06-22 | 2021-09-28 | 上海明略人工智能(集团)有限公司 | Cross-session recommendation method, system, storage medium and electronic device |
| CN113641811A (en) * | 2021-08-19 | 2021-11-12 | 中山大学 | Session recommendation method, system, device and storage medium for promoting purchasing behavior |
| CN113641811B (en) * | 2021-08-19 | 2023-09-01 | 中山大学 | Session recommendation method, system, equipment and storage medium for promoting purchasing behavior |
| WO2023197910A1 (en) * | 2022-04-12 | 2023-10-19 | 华为技术有限公司 | User behavior prediction method and related device thereof |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109816101A (en) | A kind of session sequence of recommendation method and system based on figure convolutional neural networks | |
| Kokkodis et al. | Reputation transferability in online labor markets | |
| US11070643B2 (en) | Discovering signature of electronic social networks | |
| Ipeirotis et al. | Quizz: targeted crowdsourcing with a billion (potential) users | |
| CN111813921B (en) | Topic recommendation method, electronic device and computer-readable storage medium | |
| US8838688B2 (en) | Inferring user interests using social network correlation and attribute correlation | |
| CN112733018B (en) | Session recommendation method based on graph neural network GNN and multi-task learning | |
| CN110879864B (en) | Context recommendation method based on graph neural network and attention mechanism | |
| CN110796313B (en) | Session recommendation method based on weighted graph volume and item attraction model | |
| CN111259222A (en) | Article recommendation method, system, electronic device and storage medium | |
| Yeo et al. | Conversion prediction from clickstream: Modeling market prediction and customer predictability | |
| CN112967112A (en) | Electronic commerce recommendation method for self-attention mechanism and graph neural network | |
| CN108053050A (en) | Clicking rate predictor method, device, computing device and storage medium | |
| CN111274501A (en) | Method, system and non-transitory storage medium for pushing information | |
| US20170300959A9 (en) | Method, computer readable medium and system for determining true scores for a plurality of touchpoint encounters | |
| US20210350284A1 (en) | Tackling delayed user response by modifying training data for machine-learned models | |
| CN110738314B (en) | Click rate prediction method and device based on deep migration network | |
| CN114625969A (en) | Recommendation method based on interactive neighbor session | |
| CN111368195B (en) | Model training method, device, equipment and storage medium | |
| Jankowski | Increasing website conversions using content repetitions with different levels of persuasion | |
| CN112559905B (en) | Conversation recommendation method based on dual-mode attention mechanism and social similarity | |
| CN113569139A (en) | Personalized session recommendation method and system | |
| Zhang et al. | Autor3: Au tomated R eal-time R anking with R einforcement Learning in E-commerce Sponsored Search Advertising | |
| Chen et al. | A novel recommender algorithm based on graph embedding and diffusion sampling | |
| WO2022259487A1 (en) | Prediction device, prediction method, and program |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190528 |
|
| RJ01 | Rejection of invention patent application after publication |