CN110162709A - A kind of personalized arrangement method of the robust of combination antithesis confrontation generation network - Google Patents
A kind of personalized arrangement method of the robust of combination antithesis confrontation generation network Download PDFInfo
- Publication number
- CN110162709A CN110162709A CN201910441213.0A CN201910441213A CN110162709A CN 110162709 A CN110162709 A CN 110162709A CN 201910441213 A CN201910441213 A CN 201910441213A CN 110162709 A CN110162709 A CN 110162709A
- Authority
- CN
- China
- Prior art keywords
- user
- article
- network
- robust
- factor expression
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
- G06F18/24155—Bayesian classification
Abstract
The present invention is used for computer recommending field, proposes a kind of personalized arrangement method of robust, it is intended to provide the personalized recommendation list of robust for user, user is helped more effectively to find that oneself may interested article.For recommender system application platform, the effective article recommendation list of robust also contributes to improving user to the degree of belief of the recommender system, improves user's viscosity.Robust problem is owed for present in existing personalized arrangement method, the present invention is dedicated to improving the robustness of the latent factor expression of user and article while learning the latent factor expression of user and article, to obtain a more robust personalized ranking model.Compared to the personalized arrangement method of existing robust, it is a kind of personalized arrangement method of the robust of conclusion formula that personalization arrangement method proposed by the present invention, which can inductively model the noise in the data set of Expansion,.
Description
Technical field
The present invention carries out personalized recommendation, the personalized arrangement method of specifically a kind of robust using computer technology.
Background technique
The high speed development of Internet technology, so that information overload becomes people and fast and effeciently obtains information one main barrier
Hinder.Recommender system can filter out partial information in bulk information and recommend user, to meet the information requirement of user, thus
Problem of information overload is effectively relieved.Using computer technology, recommender system recommends the information that user may like or material object
User, has been applied to the every aspect of people's life, such as commercial product recommending, and news push sings single recommendation, video recommendations etc.
Deng.
For the individual demand for meeting different user, personalized recommendation system becomes indispensable important in network application
Component part.In personalized recommendation system, personalized ranking has played important function.Personalized ranking is intended to inclined according to user
It is good, an orderly commercial product recommending list is generated for user.However, existing personalization arrangement method often lacks to ranking Shandong
The research of stick.The personalized ranking of robust refers to, when minor change occurs for the interaction data of user, personalized arrangement method
Larger change does not occur for the recommendation list provided.Without the personalized ranking of robust, for example, in film recommendation, when user's
When minor alteration occurs for viewing record, violent variation will occur for the movie listings of recommendation, or even not like most likely with family
Article come the article liked before, it is serious to reduce the accuracy recommended.In the application of actual personalized recommendation, Ke Yifa
Existing, the recommendation list of robust tends to the interest preference for more effectively reflecting user.And personalized recommendation application platform is come
It says, the recommendation list of robust also contributes to improving user's viscosity, brings continual and steady economic interests.
Existing recommender system is much based on collaborative filtering, and the main thought of this method is according to user's history
Interbehavior modeling user to the preference of article.Latent factor model is all popular in academia and industry.The latent factor can
To be interpreted as the stealthy reason that a user likes an article, for example user likes certain film, it may be possible to because of the film
In there is him to like some element or some performer etc. that he likes；If another film has similar element
And performer, then he probably can also like this film.Based on this, factor model of diving is that user and article learn the latent factor
Expression, on many data sets, latent factor model all shows good prediction accuracy, and effect can be higher than general association
Same filter algorithm.In fact, many personalization arrangement methods inherently include a latent factor model, it is that user and article learn
One low-dimensional expression.But the low-dimensional expression obtained based on noise interaction data does not have robustness, it is difficult to which effectively reflection is used
The feature preferences at family and the characteristic mass of article, the personalized ranking model after causing lack robustness.Although in recent years very
Multiphase closes the research that personalized ranking has been carried out in scholar, but existing research work is to the robust of personalized arrangement method
Journal of Sex Research is also insufficient.The personalized arrangement method of existing robust mainly includes two kinds, and first method is based on noise reduction self-encoding encoder,
Its assume interaction data be it is noiseless, according to certain noise level give interaction data add noise, then by minimize pair
The reconstruction error really to score trains self-encoding encoder, so that trained encoder is identified the noise in data, study is arrived
The low-dimensional of robust is expressed；Second method, to the dual training process of disturbance rejection, improves the Shandong of model parameter using an addition
Stick, to obtain the personalized ranking recommended models of robust.But there is common a, instruction in both methods
The model perfected only on current data set have robustness, they all cannot inductively treatment scale extension data set.
The reason is that trained model parameter depends on the noise level of training data, which is closely related with data scale.
And in fact, user-article interaction data noise level is unknown, and depend on different data sets, noise reduction self-encoding encoder root
The input data of model is obtained according to the noise level that a hyper parameter determines, during model training, the setting root of the hyper parameter
It is empirically determined according to current training set, therefore, for the interaction data of Expansion, needs to reset the hyper parameter；And
The dual training to disturbance rejection is added, disturbance parameter obtained in model training also relies on current training set, therefore fights
The model parameter for the robust that training obtains is also only effective on current data set.The present invention proposes that an antithesis confrontation generates net
Network differentiates user that network and article differentiate that network learns Bayes's personalization arrangement method and article using user
Latent factor expression forces prior-constrained, the robustness of the latent factor expression of raising respectively, so that the noise in interaction data has been modeled,
Improve the robustness of personalized ranking.
Summary of the invention
It is an object of the invention to solve the problems, such as existing personalized arrangement method and deficiency, a kind of robust is provided
Personalized arrangement method.
To achieve the purpose of the present invention, the present invention proposes that a kind of combination antithesis confrontation generates the personalized row of the robust of network
Name method.Firstly, user and article are embedded into a common lower dimensional space；Then, using Bayes's personalization ranking and
The latent factor expression of dual training two process alternative optimization users and article.The user of trained robust and article it is latent because
Submodel can inductively model the noise in interaction data, so as to the data set for the treatment of scale extension；Finally, utilizing Shandong
The user of stick and the latent factor expression prediction user of article generate the Top-K recommendation list of robust to the score of article.The invention
It mainly comprises the steps that
Step 1: acquiring the interaction data of user and article from internet, and pre-processed；
Step 2: data set is divided into training set, verifying collection and test set；
Step 3: according to the interaction data of user and article, user and article being learnt by Bayes's personalization arrangement method
Latent factor expression；
Step 4: the latent factor expression of the user and article that obtain for step 3, the confrontation by generating confrontation network are instructed
Practice process and carry out canonical, improves the robustness of the latent factor expression of user and article；
Step 5: alternately above-mentioned steps 3 and step 4 obtain the user of robust and the latent factor expression of article, thus
It carries out Score on Prediction and generates the personalized recommendation list of robust.
The interaction data of user and article is pre-processed in the step 1, the specific method is as follows: user is gathered
U, size n indicate n user；Article set I, size m indicate m article；Define user-article of n row m column
Interactive matrix, matrix element value are as follows:
User-article interaction data is handled as implicit feedback, value 1, indicates that user u occurred to interact with article i,
But do not mean that user u really likes article i；Value is that 0 expression user did not meet the article；According to the interaction of implicit feedback
Data pass through Bayes's personalization ranking, i.e. BPR, to learn ranking model.
The step 2 is that data set is divided into training set and test set, method particularly includes: for each data set,
According to leaving-one method, retains newest primary interaction for each user and constitute test set, be trained with remaining interaction data；?
During model training, randomly retain primary interaction from training data for each user and constitute verifying collection, to adjust super ginseng
Number.
The step 3 according to the interaction data of user and article, by Bayes's personalization arrangement method learn user and
The latent factor expression of article, detailed process is as follows:
A1. according to user-article Interactive matrix R, training dataset D is constructed:Its
InIndicate the article set that user u was interacted, I indicates storewide set；
A2. it for each user in training dataset D, is encoded using one-hot as input, is compiled by a user
Code device GUObtain the insertion vector p of user；For example, obtaining it for user u and being embedded in vector pu=GU(u；ΛU), wherein ΛUFor
Subscriber-coded device GUParameter；
A3. for each article in training dataset D, an object is equally passed through as input using one-hot coding
Product encoder GVObtain the insertion vector q of article；For example, obtaining it for article v and being embedded in vector qv=GV(v；ΛV), wherein
ΛVFor article code device GVParameter；
A4. BPR training user and article code device are used, the specific method is as follows:,
B1. for the example (u, i, j) in each training set, pass through encoder GUAnd GVRespectively obtain the latent factor of user
Express puWith the latent factor expression q of articlei、qj；
B2. for user-article to (u, i), predict user u to the score of article i according to (1) formula:
B3. same mode obtains user u to the score of article jOptimize subscriber-coded device and object using BPR later
The optimization problem of the parameter of product encoder, BPR is as follows:
The step be by generate user that the dual training process of confrontation network obtains step 3 and article it is latent because
Sublist improves the robustness of latent factor expression up to canonical is carried out, method particularly includes: a pair of of generation is constructed respectively for user and article
Fight network, wherein the subscriber-coded device G in step 3UWith article code device GVHerein as respective generation network, in addition
Realize that user differentiates network D respectively using two deep neural networksUNetwork D is differentiated with articleV, respectively user and article
Latent factor expression introducing is prior-constrained, to model the noise in interaction data；The specific steps of dual training are as follows:
C1. it fixes user and generates network GU, network D is differentiated using the latent factor expression of user as sample input userU；Its
In, the latent factor expression from prior distribution x~P (x) is true sample, generates network GUThe latent factor expression generated is false sample
This, training differentiates network, and can correctly distinguish input sample is still to carry out self-generating net from prior distribution x~P (x)
Network GU；Using cross entropy loss function, optimization user differentiates network DU, obtain following (3) formula of objective function, wherein ΘUFor user
Differentiate network DUThe network parameter to be learnt, n are number of users:
C2. it fixes user and differentiates network DU, training generation network GU, the latent factor expression of its generation is made to obey prior distribution x
~P (x), to make to differentiate network DUMistakenly judgement carrys out self-generating network GUUser dive factor expression be true sample；It is used
Family generates network GUFollowing (4) formula of objective function, wherein ΛUNetwork G is generated for userUThe network parameter to be learnt, n are to use
Amount:
C3. so that the latent factor expression of article is obeyed a prior distribution t~P (t) in the same way, it is latent to obtain article
The differentiation network D of factor expressionVWith generation network GVObjective function, following (5) formula and (6) formula respectively, wherein ΘVAnd ΛVFor
The network parameter to be learnt, m are article number:
The step 5 is the BPR learning process of alternately step 3 and the dual training process of step 4, obtains robust
The latent factor expression of user and article, and Score on Prediction is carried out based on the latent factor expression, according to prediction scoreIt is arranged
Name generates the personalized recommendation list of robust for user u.
Detailed description of the invention
Fig. 1 is the personalized arrangement method flow chart of robust of the invention.
Specific embodiment
To achieve the purpose of the present invention, the present invention proposes that a kind of combination antithesis confrontation generates the personalized row of the robust of network
Name method.Firstly, user and article are embedded into a common lower dimensional space；Then, using Bayes's personalization ranking and
The latent factor expression of dual training two process alternative optimization users and article.The user of trained robust and article it is latent because
Submodel can inductively model the noise in interaction data, so as to the data set for the treatment of scale extension；Finally, utilizing Shandong
The user of stick and the latent factor expression prediction user of article generate the Top-K recommendation list of robust to the score of article.The invention
It mainly comprises the steps that
Step 1: acquiring the interaction data of user and article from internet, and pre-processed；
Step 2: data set is divided into training set, verifying collection and test set；
Step 3: according to the interaction data of user and article, user and article being learnt by Bayes's personalization arrangement method
Latent factor expression；
Step 4: the latent factor expression of the user and article that obtain for step 3, the confrontation by generating confrontation network are instructed
Practice process and carry out canonical, improves the robustness of the latent factor expression of user and article；
Step 5: alternately above-mentioned steps 3 and step 4 obtain the user of robust and the latent factor expression of article, thus
It carries out Score on Prediction and generates the personalized recommendation list of robust.
The interaction data of user and article is pre-processed in the step 1, the specific method is as follows: user is gathered
U, size n indicate n user；Article set I, size m indicate m article；Define user-article of n row m column
Interactive matrix, matrix element value are as follows:
User-article interaction data is handled as implicit feedback, value 1, indicates that user u occurred to interact with article i,
But do not mean that user u really likes article i；Value is that 0 expression user did not meet the article；According to the interaction of implicit feedback
Data pass through Bayes's personalization ranking, i.e. BPR, to learn ranking model.
The step 2 is that data set is divided into training set and test set, method particularly includes: for each data set,
According to leaving-one method, retains newest primary interaction for each user and constitute test set, be trained with remaining interaction data；?
During model training, randomly retain primary interaction from training data for each user and constitute verifying collection, to adjust super ginseng
Number.
The step 3 according to the interaction data of user and article, by Bayes's personalization arrangement method learn user and
The latent factor expression of article, detailed process is as follows:
A1. according to user-article Interactive matrix R, training dataset D is constructed:Its
InIndicate the article set that user u was interacted, I indicates storewide set；
A2. it for each user in training dataset D, is encoded using one-hot as input, is compiled by a user
Code device GUObtain the insertion vector p of user；For example, obtaining it for user u and being embedded in vector pu=GU(u；ΛU), wherein ΛUFor
Subscriber-coded device GUParameter；
A3. for each article in training dataset D, an object is equally passed through as input using one-hot coding
Product encoder GVObtain the insertion vector q of article；For example, obtaining it for article v and being embedded in vector qv=GV(v；ΛV), wherein
ΛVFor article code device GVParameter；
A4. BPR training user and article code device are used, the specific method is as follows:,
B1. for the example (u, i, j) in each training set, pass through encoder GUAnd GVRespectively obtain the latent factor of user
Express puWith the latent factor expression q of articlei、qj；
B2. for user-article to (u, i), predict user u to the score of article i according to (1) formula:
B3. same mode obtains user u to the score of article jOptimize subscriber-coded device and object using BPR later
The optimization problem of the parameter of product encoder, BPR is as follows:
The step be by generate user that the dual training process of confrontation network obtains step 3 and article it is latent because
Sublist improves the robustness of latent factor expression up to canonical is carried out, method particularly includes: a pair of of generation is constructed respectively for user and article
Fight network, wherein the subscriber-coded device G in step 3UWith article code device GVHerein as respective generation network, in addition
Realize that user differentiates network D respectively using two deep neural networksUNetwork D is differentiated with articleV, respectively user and article
Latent factor expression introducing is prior-constrained, to model the noise in interaction data；By taking user as an example, make user using dual training
Latent factor expression obey a prior distribution x~P (x), the specific steps of dual training are as follows:
C1. it fixes user and generates network GU, training user's differentiation network DU: the latent factor expression of user is defeated as sample
Access customer differentiates network DU；Wherein, the latent factor expression from prior distribution x~P (x) is true sample, generates network GUIt generates
Latent factor expression is dummy copy, and training differentiates network, and can correctly distinguish input sample is from prior distribution x~P
(x), still carry out self-generating network GU；Using cross entropy loss function, optimization user differentiates network DU, objective function is as follows (3)
Formula, wherein ΘUNetwork D is differentiated for userUThe network parameter to be learnt, n are number of users:
C2. it fixes user and differentiates network DU, training generation network GU, the latent factor expression of its generation is made to obey prior distribution x
~P (x), to make to differentiate network DUMistakenly judgement carrys out self-generating network GUUser dive factor expression be true sample；It is used
Family generates network GUFollowing (4) formula of objective function, wherein ΛUNetwork G is generated for userUThe network parameter to be learnt, n are to use
Amount:
C3. so that the latent factor expression of article is obeyed a prior distribution t~P (t) in the same way, it is latent to obtain article
The differentiation network D of factor expressionVWith generation network GVObjective function, following (5) formula and (6) formula respectively, wherein ΘVAnd ΛVFor
The network parameter to be learnt, m are article number:
The step 5 is the BPR learning process of alternately step 3 and the dual training process of step 4, obtains robust
The latent factor expression of user and article, and Score on Prediction is carried out based on the latent factor expression, according to prediction scoreIt is arranged
Name generates the personalized recommendation list of robust for user u.
The present invention is a kind of personalized arrangement method of the robust of combination antithesis confrontation generation network.Using institute in the present invention
It is prior-constrained that the dual training mode stated is that the latent factor expression of user and article is forced, so as to model well user with
Noise in article interaction data, improves the robustness of the latent factor expression of user and article, and then obtains an of robust
Property ranking model, for user carry out robust article recommend.The latent factor determined by trained differentiation network in the present invention
The prior distribution of expression can inductively be suitable for the data set of the Expansion of same source.
Claims (6)
1. the personalized arrangement method that a kind of combination antithesis confrontation generates the robust of network, it is characterised in that the following steps are included:
Step 1: acquiring the interaction data of user and article from internet, and pre-processed；
Step 2: data set is divided into training set, verifying collection and test set；
Step 3: according to the interaction data of user and article, the latent of user and article being learnt by Bayes's personalization arrangement method
Factor expression；
Step 4: the latent factor expression of the user and article that are obtained for step 3, by the dual training mistake for generating confrontation network
Cheng Jinhang canonical improves the robustness of the latent factor expression of user and article；
Step 5: alternately above-mentioned steps 3 and step 4 obtain the user of robust and the latent factor expression of article, to carry out
Score on Prediction and the personalized recommendation list for generating robust.
2. a kind of combination antithesis confrontation according to claim 1 generates the personalized arrangement method of the robust of network, special
Sign is: pre-processing in the step 1 to the interaction data of user and article, the specific method is as follows: user is gathered
U, size n indicate n user；Article set I, size m indicate m article；Define user-article of n row m column
Interactive matrix, matrix element value are as follows:
User-article interaction data is handled as implicit feedback, value 1, indicates that user u occurred to interact with article i, but not
Mean that user u really likes article i；Value is that 0 expression user did not meet the article；According to the interaction number of implicit feedback
According to passing through Bayes's personalization ranking, i.e. BPR, to learn ranking model.
3. a kind of combination antithesis confrontation according to claim 1 generates the personalized arrangement method of the robust of network, special
Sign is: the step 2 is that data set is divided into training set and test set, method particularly includes: for each data set, root
According to leaving-one method, retains newest primary interaction for each user and constitute test set, be trained with remaining interaction data；In mould
In type training process, randomly retains primary interaction from training data for each user and constitute verifying collection, to adjust hyper parameter.
4. a kind of combination antithesis confrontation according to claim 1 generates the personalized arrangement method of the robust of network, special
Sign is: the step 3 learns user and object by Bayes's personalization arrangement method according to the interaction data of user and article
The latent factor expression of product, detailed process is as follows:
A1. according to user-article Interactive matrix R, training dataset D is constructed:WhereinIndicate the article set that user u was interacted, I indicates storewide set；
A2. it for each user in training dataset D, is encoded using one-hot as input, passes through a subscriber-coded device
GUObtain the insertion vector p of user；For example, obtaining it for user u and being embedded in vector pu=GU(u；ΛU), wherein ΛUFor user
Encoder GUParameter；
A3. it for each article in training dataset D, is equally compiled as input by an article using one-hot coding
Code device GVObtain the insertion vector q of article；For example, obtaining it for article v and being embedded in vector qv=GV(v；ΛV), wherein ΛVFor
Article code device GVParameter；
A4. BPR training user and article code device are used, the specific method is as follows:
B1. for the example (u, i, j) in each training set, pass through encoder GUAnd GVRespectively obtain the latent factor expression of user
puWith the latent factor expression q of articlei、qj；
B2. for user-article to (u, i), predict user u to the score of article i according to (1) formula:
B3. same mode obtains user u to the score of article jOptimize subscriber-coded device and article code using BPR later
The optimization problem of the parameter of device, BPR is as follows:
。
5. a kind of combination antithesis confrontation according to claim 1 generates the personalized arrangement method of the robust of network, special
Sign is: the step 4 be by generate user that the dual training process of confrontation network obtains step 3 and article it is latent because
Sublist improves the robustness of latent factor expression up to canonical is carried out, method particularly includes: a pair of of generation is constructed respectively for user and article
Fight network, wherein the subscriber-coded device G in step 3UWith article code device GVHerein as respective generation network, in addition
Realize that user differentiates network D respectively using two deep neural networksUNetwork D is differentiated with articleV, respectively user and article
Latent factor expression introducing is prior-constrained, to model the noise in interaction data；The specific steps of dual training are as follows:
C1. it fixes user and generates network GU, network D is differentiated using the latent factor expression of user as sample input userU；Wherein, come
It is true sample from the latent factor expression of prior distribution x~P (x), generates network GUThe latent factor expression generated is dummy copy, training
Differentiate network, can correctly distinguish input sample is still to carry out self-generating network G from prior distribution x~P (x)U；Make
With cross entropy loss function, optimization user differentiates network DU, obtain following (3) formula of objective function, wherein ΘUNet is differentiated for user
Network DUThe network parameter to be learnt, n are number of users:
C2. it fixes user and differentiates network DU, training generation network GU, the latent factor for generating it, which reaches, obeys prior distribution x~P
(x), to make to differentiate network DUMistakenly judgement carrys out self-generating network GUUser dive factor expression be true sample；Obtain user
Generate network GUFollowing (4) formula of objective function, wherein ΛUNetwork G is generated for userUThe network parameter to be learnt, n are user
Number:
C3. so that the latent factor expression of article is obeyed a prior distribution t~P (t) in the same way, obtain the latent factor of article
The differentiation network D of expressionVWith generation network GVObjective function, following (5) formula and (6) formula respectively, wherein ΘVAnd ΛVTo learn
The network parameter of habit, m are article number:
。
6. a kind of combination antithesis confrontation according to claim 1 generates the personalized arrangement method of the robust of network, special
Sign is: the step 5 is the BPR learning process of alternately step 3 and the dual training process of step 4, obtains robust
The latent factor expression of user and article, and Score on Prediction is carried out based on the latent factor expression, according to prediction scoreIt is arranged
Name generates the personalized recommendation list of robust for user u.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910441213.0A CN110162709B (en) | 2019-05-24 | 2019-05-24 | Robust personalized ranking method for generating network by combining dual confrontation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910441213.0A CN110162709B (en) | 2019-05-24 | 2019-05-24 | Robust personalized ranking method for generating network by combining dual confrontation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110162709A true CN110162709A (en) | 2019-08-23 |
| CN110162709B CN110162709B (en) | 2020-03-13 |
Family
ID=67632768
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910441213.0A Active CN110162709B (en) | 2019-05-24 | 2019-05-24 | Robust personalized ranking method for generating network by combining dual confrontation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110162709B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110727844A (en) * | 2019-10-21 | 2020-01-24 | 东北林业大学 | Online commented commodity feature viewpoint extraction method based on generation countermeasure network |
| CN110781401A (en) * | 2019-11-07 | 2020-02-11 | 电子科技大学 | Top-n project recommendation method based on collaborative autoregressive flow |
| CN111104601A (en) * | 2019-12-26 | 2020-05-05 | 河南理工大学 | Antagonistic multi-feedback-level paired personalized ranking method |
| CN111241417A (en) * | 2020-01-06 | 2020-06-05 | 浙江工业大学 | Method for enhancing robustness of collaborative recommendation system |
| CN111259233A (en) * | 2020-01-06 | 2020-06-09 | 浙江工业大学 | Method for improving stability of collaborative filtering model |
| CN111460130A (en) * | 2020-03-27 | 2020-07-28 | 咪咕数字传媒有限公司 | Information recommendation method, device, equipment and readable storage medium |
| CN113221003A (en) * | 2021-05-20 | 2021-08-06 | 北京建筑大学 | Mixed filtering recommendation method and system based on dual theory |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101714198A (en) * | 2009-10-29 | 2010-05-26 | 北京航空航天大学 | System based on Bayesian estimation for evaluating credibility of countermeasure information of computer network |
| CN104199904A (en) * | 2014-08-27 | 2014-12-10 | 腾讯科技（深圳）有限公司 | Social information push method, server, user terminal and system |
| US20180053210A1 (en) * | 2016-08-17 | 2018-02-22 | Adobe Systems Incorporated | Personalization of Digital Content Recommendations |
| US10218800B2 (en) * | 2014-07-28 | 2019-02-26 | Oath Inc. | Systems and methods for providing recommendations and explanations |
| CN109657149A (en) * | 2018-12-25 | 2019-04-19 | 合肥学院 | A kind of recommended method and system based on generation confrontation network and double focusing class |
| CN109657156A (en) * | 2019-01-22 | 2019-04-19 | 杭州师范大学 | A kind of personalized recommendation method generating confrontation network based on circulation |
-
2019
- 2019-05-24 CN CN201910441213.0A patent/CN110162709B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101714198A (en) * | 2009-10-29 | 2010-05-26 | 北京航空航天大学 | System based on Bayesian estimation for evaluating credibility of countermeasure information of computer network |
| US10218800B2 (en) * | 2014-07-28 | 2019-02-26 | Oath Inc. | Systems and methods for providing recommendations and explanations |
| CN104199904A (en) * | 2014-08-27 | 2014-12-10 | 腾讯科技（深圳）有限公司 | Social information push method, server, user terminal and system |
| US20180053210A1 (en) * | 2016-08-17 | 2018-02-22 | Adobe Systems Incorporated | Personalization of Digital Content Recommendations |
| CN109657149A (en) * | 2018-12-25 | 2019-04-19 | 合肥学院 | A kind of recommended method and system based on generation confrontation network and double focusing class |
| CN109657156A (en) * | 2019-01-22 | 2019-04-19 | 杭州师范大学 | A kind of personalized recommendation method generating confrontation network based on circulation |
Non-Patent Citations (1)
| Title |
|---|
| 陈晋音等: "基于图过滤的快速密度聚类双层网络推荐算法", 《控制理论与应用》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110727844A (en) * | 2019-10-21 | 2020-01-24 | 东北林业大学 | Online commented commodity feature viewpoint extraction method based on generation countermeasure network |
| CN110727844B (en) * | 2019-10-21 | 2022-07-01 | 东北林业大学 | Online commented commodity feature viewpoint extraction method based on generation countermeasure network |
| CN110781401A (en) * | 2019-11-07 | 2020-02-11 | 电子科技大学 | Top-n project recommendation method based on collaborative autoregressive flow |
| CN111104601A (en) * | 2019-12-26 | 2020-05-05 | 河南理工大学 | Antagonistic multi-feedback-level paired personalized ranking method |
| CN111104601B (en) * | 2019-12-26 | 2022-09-13 | 河南理工大学 | Antagonistic multi-feedback-level paired personalized ranking method |
| CN111241417A (en) * | 2020-01-06 | 2020-06-05 | 浙江工业大学 | Method for enhancing robustness of collaborative recommendation system |
| CN111259233A (en) * | 2020-01-06 | 2020-06-09 | 浙江工业大学 | Method for improving stability of collaborative filtering model |
| CN111241417B (en) * | 2020-01-06 | 2022-07-26 | 浙江工业大学 | Method for enhancing robustness of collaborative recommendation system |
| CN111259233B (en) * | 2020-01-06 | 2022-07-26 | 浙江工业大学 | Method for improving stability of collaborative filtering model |
| CN111460130A (en) * | 2020-03-27 | 2020-07-28 | 咪咕数字传媒有限公司 | Information recommendation method, device, equipment and readable storage medium |
| CN113221003A (en) * | 2021-05-20 | 2021-08-06 | 北京建筑大学 | Mixed filtering recommendation method and system based on dual theory |
| CN113221003B (en) * | 2021-05-20 | 2023-05-02 | 北京建筑大学 | Mixed filtering recommendation method and system based on dual theory |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110162709B (en) | 2020-03-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110162709A (en) | A kind of personalized arrangement method of the robust of combination antithesis confrontation generation network | |
| Okura et al. | Embedding-based news recommendation for millions of users | |
| CN111797321B (en) | Personalized knowledge recommendation method and system for different scenes | |
| WO2021223567A1 (en) | Content processing method and apparatus, computer device, and storage medium | |
| CN108763362B (en) | Local model weighted fusion Top-N movie recommendation method based on random anchor point pair selection | |
| Nguyen et al. | Exploring the filter bubble: the effect of using recommender systems on content diversity | |
| CN112313697A (en) | System and method for generating interpretable description-based recommendations describing angle augmentation | |
| CN112163165A (en) | Information recommendation method, device, equipment and computer readable storage medium | |
| Yu et al. | Adaptive social similarities for recommender systems | |
| CN111274440A (en) | Video recommendation method based on visual and audio content relevancy mining | |
| KR102129583B1 (en) | Method and apparatus for recommending items based on deep learning | |
| CN110019943A (en) | Video recommendation method, device, electronic equipment and storage medium | |
| CN110032684B (en) | Information cross-domain parallel sequence recommendation method, medium and equipment based on shared account | |
| CN111339439A (en) | Collaborative filtering recommendation method and device fusing comment text and time sequence effect | |
| Mehta et al. | Movie recommendation systems using sentiment analysis and cosine similarity | |
| CN111125428B (en) | Time-dependent movie recommendation method based on score prediction function fitting structure | |
| Stöcker et al. | Riding the wave of misclassification: How we end up with extreme youtube content | |
| Sridhar et al. | Content-Based Movie Recommendation System Using MBO with DBN. | |
| CN116680363A (en) | Emotion analysis method based on multi-mode comment data | |
| CN115098786A (en) | News recommendation method and system based on gating multi-head self-attention | |
| CN113590965B (en) | Video recommendation method integrating knowledge graph and emotion analysis | |
| Selk et al. | Nonparametric regression and classification with functional, categorical, and mixed covariates | |
| CN109697255A (en) | A kind of Personalize News jettison system and method based on automatic measure on line | |
| CN110008406A (en) | A kind of public sentiment personalized recommendation algorithm based on similar community | |
| CN110347916A (en) | Cross-scenario item recommendation method, device, electronic equipment and storage medium |
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 | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right |
Denomination of invention: A robust personalized ranking method combined with dual confrontation generation network Effective date of registration: 20220509 Granted publication date: 20200313 Pledgee: Bank of Chengdu science and technology branch of Limited by Share Ltd. Pledgor: ZHONGSENYUNLIAN (CHENGDU) TECHNOLOGY Co.,Ltd. Registration number: Y2022980005318 |