CN109241440A - It is a kind of based on deep learning towards implicit feedback recommended method - Google Patents
It is a kind of based on deep learning towards implicit feedback recommended method Download PDFInfo
- Publication number
- CN109241440A CN109241440A CN201811145651.4A CN201811145651A CN109241440A CN 109241440 A CN109241440 A CN 109241440A CN 201811145651 A CN201811145651 A CN 201811145651A CN 109241440 A CN109241440 A CN 109241440A
- Authority
- CN
- China
- Prior art keywords
- vector
- user
- layer
- data
- project
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The proposed algorithm towards implicit feedback based on deep learning that the present invention relates to a kind of, selects hidden data as training data；The feature vector of user and project is obtained by way of variation automatic coding machine according to user-project Interactive matrix；The feature vector of user and project progress vector multiplication is obtained into new vector A；The feature vector of user and project are inlayed into the new vector D of composition, the input layer of D input multilayer deep neural network structural model is then obtained to the output vector E of input layer；The result input hidden layer of inlaying of vector A and vector E is continued to train, obtains new model parameter, while by the output input and output layer of hidden layer, obtaining final prediction result；The data predicted will be needed to be put into trained neural network structure model, obtain prediction result.The present invention solves the problems, such as that recommendation results are not in artificial caused deviation, and the data needed preferably obtain relatively, and simple and easy, hardware requirement is low, and time loss is few.
Description
Technical field
This method is related to proposed algorithm, especially a kind of proposed algorithm towards implicit feedback based on deep learning.
Background technique
With the development of society and the progress in epoch, information technology and Internet technology enter the mode of high speed development, society
The information overload epoch can be entered from the absence of information epoch.In information overload, the either publisher of information or letter
The recipient of breath all suffers from very big challenge.Publisher's problem encountered of information is: the diversification of publication how to be allowed to be believed
Breath is targetedly received and is paid close attention to by the recipient of information；Recipient's problem encountered of information is: how from magnanimity
The information oneself liked is found in information.In order to help people's fast and effectively filter information, proposed algorithm is come into being.Recommend
Algorithm is by a large amount of use in service industry, e-commerce, social networks, cache contents selection.Proposed algorithm by with
The analysis and modeling of family hobby and behavior, by user it is possible that interested project recommendation is to user.The effect of proposed algorithm and
The quality of the user information and used proposed algorithm that are collected into has direct relationship.Comment, scoring and the user of user
Relevant information etc. be that user is unsolicited, these information intuitively show the personal preference of user, this category information very much
Referred to as explicit feedback；For the information that user passively provides, such as the click of user, downloading, viewing, purchasing history are a series of
The information of individual subscriber hobby cannot be directly expressed, then referred to as implicit feedback.Currently, most of proposed algorithm only needle
To explicit feedback data, and implicit feedback data are had ignored, but implicit feedback total amount of data has far more than display feedback
Even only implicit feedback data (such as space flight, communication etc.) under scene, it can be seen that, the proposed algorithm for implicit feedback is ten
Divide necessary.
The problems such as negative factor evidence and less information dimension are lacked for implicit feedback data, now for implicit feedback
Method can be divided into following 3 kinds: (1) proposed algorithm based on the other collaborative filtering of unitary class, this algorithm will only exist positive sample
Collaborative filtering problem be summarized as single type collaborative filtering problem (One-Class Collaborative Filtering,
OCCF), all missing datas are considered as negative sample data (ALL Missing As Negative, AMAN) by algorithm, or will
All missing datas are considered as unknown data (All Missing As Uknown, AMAU), and each sample data is based on not
The model that the data of missing are included in weight together is simultaneously trained by same weight, or is divided the negative sample of missing data
Cloth is it is assumed that be the common solution of such algorithm.(2) the implicit feedback proposed algorithm of external information auxiliary is introduced.Only according to
The basic reason undesirable by implicit feedback recommendation effect is: lacking the direct judgement to user preferences.Therefore occur drawing
Enter the implicit feedback proposed algorithm of external information auxiliary.Such method is divided into following basic 3 class: introducing the implicit feedback of context
Proposed algorithm, the implicit feedback proposed algorithm for introducing cross-domain knowledge, the implicit feedback proposed algorithm for introducing social information.In introducing
Implicit feedback proposed algorithm hereafter is sufficiently to excavate contextual information, such as time locating for user, place, state, mood etc.,
These contextual informations have very big relationship to the selection of project to user, can mention contextual information as auxiliary information introducing
The effect that high implicit feedback is recommended；The implicit feedback proposed algorithm for introducing cross-domain knowledge is by a kind of feasible side of transfer learning
The knowledge learnt in other field is assisted the learning tasks in recommender system by method, and representing algorithm is that other in LFM model are led
The latent factor that domain is acquired can help the study of the characteristics algorithm of target domain；The implicit feedback for introducing social information is recommended
Algorithm is accurate, the comprehensive user information for aiding in user, and such as friend information, area information, community information etc. excavates user
Interested project.(3) based on the proposed algorithm of sequence.Now the proposed algorithm based on sequence gradually become recommend
Hot spot in algorithm.Sort recommendations algorithm is to be excavated and handled by existing data and form one according to corresponding rule
The model of a sequence, user object later can be ranked up according to this model.Be broadly divided into following 2 class: point-by-point sequence and by
Team's sequence.
The method of above-mentioned three kinds of implicit feedbacks has the following problems:
(1) hypothesis that weight is added is unstable, and deviation occurs in the result that will lead to recommendation；The time for calculating all samples is multiple
Higher, the time it takes higher cost of miscellaneous degree；Since the partial information of missing can not be utilized, it is random to sometimes result in result
The problems such as appearance.
(2) context information data of user and project interaction is difficult to obtain and excavate, and lacks diversified auxiliary letter
Breath is so that the result of proposed algorithm is poor.The migration circle span of knowledge is larger, can generate biggish deviation to the result of recommendation.
(3) time complexity for calculating sample is higher, has ignored the pass between sample and sample, between user and user
System, model, which calculates, sometimes needs a large amount of time and hardware to be calculated.
Summary of the invention
In order to which the hypothesis for solving above-mentioned addition weight is unstable, there is deviation, user and project in the result that will lead to recommendation
Interactive context information data is difficult to obtain and excavate, and calculates the higher problem of time complexity of sample, and the present invention proposes
Following technical scheme:
(1) select hidden data as training data, if data set used is the data set of explicit feedback, need into
Explicit data is converted to hidden data by the pretreatment of row data；And establish the corresponding user of hidden data-project interaction square
Battle array；
(2) user and project are mapped to together by way of variation automatic coding machine according to user-project Interactive matrix
In one latent space, the feature vector B of user and the feature vector C of project are obtained；
(3) the feature vector B of the user and feature vector C of the project multiplication for carrying out vector is operated, and it is complete to store operation
Finish obtained new vector A；
(4) the feature vector C of the feature vector B of user and project are carried out inlaying the new vector D of composition, D is inputted into multilayer
The input layer of deep neural network structural model is trained multilayer deep neural network structural model parameter, while obtaining defeated
Enter the output vector E of layer, i.e. object vector；Wherein new vector D is placed on by C is augmented to obtain behind B；The depth nerve net
Network is adopted using variation from coding structure, hidden layer and output layer by input layer, hidden layer, output layer up of three-layer, input layer
Use multi-layer perception (MLP)；
(5) new vector A and characterize data obtained in (4) that characterize data linear character is used for obtained in (3) is non-
The object vector E of linear character is inlayed to obtain new vector F, the model parameter that F input hidden layer is obtained according to step 4 after
Continuous training, obtains new model parameter, while obtaining the output vector of hidden layer；This general deep neural network carries out 20 times repeatedly
In generation, can also adjust the number of iterations according to required precision, but at least should not be below 10 times, and highest should not be greater than 200 times.
(6) by the output input and output layer of (5) obtained hidden layer, the final prediction result of output layer is obtained；Pass through
The gap between final prediction score and true score is minimized constantly to train this multilayer deep neural network structural model,
The optimized parameter of the network structure model is obtained, the parameter training to the network structure model is completed；
(7) data predicted will be needed to be put into trained neural network structure model, obtains prediction result.
User and project are abstracted into binaryzation unitary vector by input layer by training data；It is again that input layer is obtained
Unitary vector is sent into hidden layer, that is, multi-layer perception (MLP) and is trained, to excavate the potential connection in user and project, this hair
It is bright to be not only extracted the linear character of data, but also be extracted the nonlinear characteristic of data and blend them, the data extracted
Feature is more comprehensive.What final output layer obtained is prediction scoreTraining passes through minimumWith its target value yuiBetween
Point-by-point loss carries out.Implicit feedback data are sent into deep neural network by this method, extract data spy with nonlinear mode
Sign, and learn data characteristics from data using a linear kernel, then the data characteristics that two methods are learnt is mutual
Fusion is strengthened mutually, can preferably be analyzed user-project Interactive matrix, available preferable in a short time
Result.
Beneficial effect
The present invention establishes deep learning model according to the data set of collected user and project, and prediction user is possible to feel
The bulleted list of interest, and provide front and back sequence.This method is without artificial weight setting early period, so that recommendation results are not in people
For caused deviation；Due to being the method for implicit recommendation, the data of other dimensions are not needed, the data needed are relatively preferable
It obtains；Since the present invention uses deep neural network, compared with traditional proposed algorithm, required data volume is few, preferable to solve
It has determined cold start-up problem, this method is simple and easy, and hardware requirement is low, and time loss is few.
Detailed description of the invention
Fig. 1 method flow diagram
Fig. 2 accuracy rate change curve
Fig. 3 normalizing accoumulation of discount profit change curve
Specific embodiment
(1) use MovienLens (1M) as the training set of model and the source of test set, MovienLens (1M) has
6040 users and 3706 projects, this data set include 1,000,000 scorings, each user at least 20 scorings, this film
Score data collection is widely used, for the effect of proposed algorithm.Its data is handled, user has evaluation to project
Interaction is designated generally as 1, and the interaction that user does not evaluate project is designated generally as 0, and gives up in addition to timestamp remaining
Attribute data makes it be completely converted into implicit feedback data set, and according to the principle of leave-one-out, most by each user
Test set of the close primary interaction as model, training set of the remainder data as model.And establish user-project Interactive matrix
(user-item matrix)。
(2) project corresponding to user in user-project Interactive matrix (user-item matrix) is extracted, is obtained
To the corresponding bulleted list of user (item-list), the one-dimensional vector that this list is 3706 yuan, and it is automatic by multinomial variation
The mode of code machine (VARIATIONAL AUTO ENCODER), obtains initial user feature vector corresponding to user, this is special
The one-dimensional vector that vector is 1024 yuan is levied, then this feature vector is carried out one by the neural network (MLP) of one layer of full articulamentum
Secondary dimension compression, makes it become user characteristics vector (user latent vector), is denoted as vector B.By user-project interaction
User corresponding to project extracts in matrix (uesr-item matrix), obtains the corresponding user list (user- of project
List), the one-dimensional vector that this list is 6040 yuan, and pass through multinomial variation automatic coding machine (VARIATIONAL AUTO
ENCODER mode), obtains initial user feature vector corresponding to project, the one-dimensional vector that this feature vector is 1024 yuan,
This feature vector is subjected to a dimension compression by the neural network of one layer of full articulamentum (MLP) again, it is made to become project spy
It levies vector (item latent vector), is denoted as vector C.
(3) by user characteristics vector obtained in (2) (user latent vector) both vector B and item characteristic
Vector (item latent vector) both vector C, carries out the dot product of vector, obtains vector after dot product, be denoted as vector A.
(4) by user characteristics vector B (item latent vector) and item feature vector C (item latent
Vector it) inlays to obtain one 128 yuan of one-dimensional vector, is denoted as vector D, it is hidden as deep neural network used by this method
The input of layer-multi-layer perception (MLP) (MLP) is hidden, the training of deep neural network parameter is carried out.And object vector E is obtained in the method
Deep neural network-multi-layer perception (MLP) (MLP) used in hidden layer is a kind of tower network structure, and bottom is most wide, Mei Gehou
There is less neuronal quantity after layer.Network shares 4 layers, and every layer of neuronal quantity is respectively 128,64,32,8.This mind
Through network is defined as: W is the weight matrix in perceptron among these, and b is the mind of neural network
Through threshold values, a is the activation primitive of neural network, and as used herein is ReLU activation primitive；Z thus tie by training for neural network
Fruit, puFor the final feature vector of user, qiFor the final feature vector of project, yuiThe score of neural network prediction thus；It considers
yuiPossibility value have two-value (0 or 1), choose this deep neural network majorized function be two class cross entropy loss functions
(binary cross-entroy loss) is optimized using the training that random descent method carries out neural network (SGD).
(5) obtained vector E in vector A obtained in (3) and (4) is inlayed to obtain the unitary of one 72 dimension to
Amount, obtains one 72 yuan of one-dimensional vector, is denoted as vector F.Vector F is put into one one layer of the neural network connected entirely into
Row training, obtains the output result of hidden layer.
(6) model prediction score, and the score of true result will in the output result input and output layer of hidden layer, be obtained
Comparison is made, to optimize the parameter of deep neural network, after the number of iterations reaches 20 times, deep neural network tends to be steady
Fixed, parameters achieve the effect that more excellent.
(7) with the test set test depth neural network selected in (1), accuracy rate Precision and normalizing discount are selected
Accumulated profit (Normalized Discounted Cumulative Gain) tests the effect of the method for the invention.NtpIt is the quantity that the article of algorithm recommendation is liked for user, NfpFor algorithm recommend article be user not
The quantity liked；
When the sample space of test is 20, the effect of algorithm is as shown in Figure 2 and Figure 3.By Fig. 2 Fig. 3 as it can be seen that accuracy rate and
Normalizing accoumulation of discount profit, which only passes through an iteration, can obtain relatively stable effect, and accuracy rate is stablized 0.7 or so, normalizing
Accoumulation of discount profit, which is stablized, represents the number of iterations in 0.41 or so, Fig. 2,3 horizontal axis, the digital generation respectively below corresponding the number of iterations
Table corresponds to the pregroup rate and normalization cumulative gain of the number of iterations, and is all ideal result.
Claims (1)
1. a kind of proposed algorithm towards implicit feedback based on deep learning, it is characterised in that the following steps are included:
(1) hidden data is selected to be counted as training data if data set used is the data set of explicit feedback
According to pretreatment explicit data is converted into hidden data；And establish the corresponding user of hidden data-project Interactive matrix；
(2) user and project be mapped to by way of variation automatic coding machine according to user-project Interactive matrix same
In latent space, the feature vector B of user and the feature vector C of project are obtained；
(3) the feature vector B of the user and feature vector C of the project multiplication for carrying out vector is operated, and stores operation and finishes
The new vector A arrived；
(4) the feature vector C of the feature vector B of user and project are carried out inlaying the new vector D of composition, D is inputted into multilayer depth
The input layer of neural network structure model is trained multilayer deep neural network structural model parameter, while obtaining input layer
Output vector E, i.e. object vector；Wherein new vector D is placed on by C is augmented to obtain behind B；The deep neural network by
Input layer, hidden layer, output layer up of three-layer, using variation from coding structure, hidden layer and output layer are all made of more input layer
Layer perceptron；
(5) new vector A and characterize data obtained in (4) that characterize data linear character is used for obtained in (3) is non-linear
The object vector E of feature is inlayed to obtain new vector F, and F input hidden layer is continued to instruct according to the model parameter that step 4 obtains
Practice, obtains new model parameter, while obtaining the output vector of hidden layer；
(6) by the output input and output layer of (5) obtained hidden layer, the final prediction result of output layer is obtained；Pass through minimum
Change the gap between final prediction score and true score constantly to train this multilayer deep neural network structural model, obtains
The optimized parameter of the network structure model completes the parameter training to the network structure model；
(7) data predicted will be needed to be put into trained neural network structure model, obtains prediction result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811145651.4A CN109241440A (en) | 2018-09-29 | 2018-09-29 | It is a kind of based on deep learning towards implicit feedback recommended method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811145651.4A CN109241440A (en) | 2018-09-29 | 2018-09-29 | It is a kind of based on deep learning towards implicit feedback recommended method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109241440A true CN109241440A (en) | 2019-01-18 |
Family
ID=65055192
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811145651.4A Pending CN109241440A (en) | 2018-09-29 | 2018-09-29 | It is a kind of based on deep learning towards implicit feedback recommended method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109241440A (en) |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109871504A (en) * | 2019-01-24 | 2019-06-11 | 中国科学院软件研究所 | A kind of Course Recommendation System based on Heterogeneous Information network and deep learning |
| CN110059251A (en) * | 2019-04-22 | 2019-07-26 | 郑州大学 | Collaborative filtering recommending method based on more relationship implicit feedback confidence levels |
| CN110704728A (en) * | 2019-09-06 | 2020-01-17 | 北京达佳互联信息技术有限公司 | Information recommendation method and device, electronic equipment and storage medium |
| CN110727872A (en) * | 2019-10-21 | 2020-01-24 | 深圳微品致远信息科技有限公司 | Method and device for mining ambiguous selection behavior based on implicit feedback |
| CN110807150A (en) * | 2019-10-14 | 2020-02-18 | 腾讯科技（深圳）有限公司 | Information processing method and device, electronic equipment and computer readable storage medium |
| CN110838020A (en) * | 2019-09-16 | 2020-02-25 | 平安科技（深圳）有限公司 | Recommendation method and device based on vector migration, computer equipment and storage medium |
| CN110837596A (en) * | 2019-09-16 | 2020-02-25 | 中国平安人寿保险股份有限公司 | Intelligent recommendation method and device, computer equipment and storage medium |
| CN110910218A (en) * | 2019-11-21 | 2020-03-24 | 南京邮电大学 | Multi-behavior migration recommendation method based on deep learning |
| CN111177580A (en) * | 2019-12-21 | 2020-05-19 | 杭州电子科技大学 | Method for realizing personalized recommendation by utilizing multiple implicit feedback |
| CN111291274A (en) * | 2020-03-02 | 2020-06-16 | 苏州大学 | Article recommendation method, device, equipment and computer-readable storage medium |
| CN111432003A (en) * | 2020-03-27 | 2020-07-17 | 尹兵 | Data pushing method and device applied to cloud computing, electronic equipment and system |
| CN111476223A (en) * | 2020-06-24 | 2020-07-31 | 支付宝(杭州)信息技术有限公司 | Method and device for evaluating interaction event |
| CN111523940A (en) * | 2020-04-23 | 2020-08-11 | 华中科技大学 | Deep reinforcement learning-based recommendation method and system with negative feedback |
| CN112149734A (en) * | 2020-09-23 | 2020-12-29 | 哈尔滨工程大学 | Cross-domain recommendation method based on stacked self-encoder |
| CN112231584A (en) * | 2020-12-08 | 2021-01-15 | 平安科技（深圳）有限公司 | Data pushing method and device based on small sample transfer learning and computer equipment |
| CN112231582A (en) * | 2020-11-10 | 2021-01-15 | 南京大学 | Website recommendation method and equipment based on variational self-coding data fusion |
| CN112800344A (en) * | 2021-01-29 | 2021-05-14 | 重庆邮电大学 | Deep neural network-based movie recommendation method |
| CN113010802A (en) * | 2021-03-25 | 2021-06-22 | 华南理工大学 | Recommendation method facing implicit feedback based on multi-attribute interaction of user and article |
| CN113254795A (en) * | 2020-02-11 | 2021-08-13 | 北京京东振世信息技术有限公司 | Training method and device for recommendation model |
| CN113342963A (en) * | 2021-04-29 | 2021-09-03 | 山东大学 | Service recommendation method and system based on transfer learning |
| CN113590964A (en) * | 2021-08-04 | 2021-11-02 | 燕山大学 | Deep neural network Top-N recommendation algorithm based on heterogeneous modeling |
| CN114581161A (en) * | 2022-05-06 | 2022-06-03 | 深圳市明珞锋科技有限责任公司 | Information pushing method and system based on deep learning |
| CN115841366A (en) * | 2022-12-30 | 2023-03-24 | 中国科学技术大学 | Article recommendation model training method and device, electronic equipment and storage medium |
-
2018
- 2018-09-29 CN CN201811145651.4A patent/CN109241440A/en active Pending
Cited By (36)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109871504A (en) * | 2019-01-24 | 2019-06-11 | 中国科学院软件研究所 | A kind of Course Recommendation System based on Heterogeneous Information network and deep learning |
| CN109871504B (en) * | 2019-01-24 | 2020-09-29 | 中国科学院软件研究所 | Course recommendation system based on heterogeneous information network and deep learning |
| CN110059251B (en) * | 2019-04-22 | 2022-10-28 | 郑州大学 | Collaborative filtering recommendation method based on multi-relation implicit feedback confidence |
| CN110059251A (en) * | 2019-04-22 | 2019-07-26 | 郑州大学 | Collaborative filtering recommending method based on more relationship implicit feedback confidence levels |
| CN110704728A (en) * | 2019-09-06 | 2020-01-17 | 北京达佳互联信息技术有限公司 | Information recommendation method and device, electronic equipment and storage medium |
| CN110838020A (en) * | 2019-09-16 | 2020-02-25 | 平安科技（深圳）有限公司 | Recommendation method and device based on vector migration, computer equipment and storage medium |
| CN110837596A (en) * | 2019-09-16 | 2020-02-25 | 中国平安人寿保险股份有限公司 | Intelligent recommendation method and device, computer equipment and storage medium |
| CN110838020B (en) * | 2019-09-16 | 2023-06-23 | 平安科技（深圳）有限公司 | Recommendation method and device based on vector migration, computer equipment and storage medium |
| CN110807150A (en) * | 2019-10-14 | 2020-02-18 | 腾讯科技（深圳）有限公司 | Information processing method and device, electronic equipment and computer readable storage medium |
| CN110727872A (en) * | 2019-10-21 | 2020-01-24 | 深圳微品致远信息科技有限公司 | Method and device for mining ambiguous selection behavior based on implicit feedback |
| CN110910218B (en) * | 2019-11-21 | 2022-08-26 | 南京邮电大学 | Multi-behavior migration recommendation method based on deep learning |
| CN110910218A (en) * | 2019-11-21 | 2020-03-24 | 南京邮电大学 | Multi-behavior migration recommendation method based on deep learning |
| CN111177580A (en) * | 2019-12-21 | 2020-05-19 | 杭州电子科技大学 | Method for realizing personalized recommendation by utilizing multiple implicit feedback |
| CN113254795A (en) * | 2020-02-11 | 2021-08-13 | 北京京东振世信息技术有限公司 | Training method and device for recommendation model |
| CN113254795B (en) * | 2020-02-11 | 2023-11-07 | 北京京东振世信息技术有限公司 | Training method and device for recommendation model |
| CN111291274A (en) * | 2020-03-02 | 2020-06-16 | 苏州大学 | Article recommendation method, device, equipment and computer-readable storage medium |
| CN111432003A (en) * | 2020-03-27 | 2020-07-17 | 尹兵 | Data pushing method and device applied to cloud computing, electronic equipment and system |
| CN111432003B (en) * | 2020-03-27 | 2021-01-08 | 上海星地通讯工程研究所 | Data pushing method and device applied to cloud computing, electronic equipment and system |
| CN111523940A (en) * | 2020-04-23 | 2020-08-11 | 华中科技大学 | Deep reinforcement learning-based recommendation method and system with negative feedback |
| CN111476223B (en) * | 2020-06-24 | 2020-09-22 | 支付宝(杭州)信息技术有限公司 | Method and device for evaluating interaction event |
| CN111476223A (en) * | 2020-06-24 | 2020-07-31 | 支付宝(杭州)信息技术有限公司 | Method and device for evaluating interaction event |
| CN112149734A (en) * | 2020-09-23 | 2020-12-29 | 哈尔滨工程大学 | Cross-domain recommendation method based on stacked self-encoder |
| CN112231582A (en) * | 2020-11-10 | 2021-01-15 | 南京大学 | Website recommendation method and equipment based on variational self-coding data fusion |
| CN112231582B (en) * | 2020-11-10 | 2023-11-21 | 南京大学 | Website recommendation method and equipment based on variation self-coding data fusion |
| CN112231584A (en) * | 2020-12-08 | 2021-01-15 | 平安科技（深圳）有限公司 | Data pushing method and device based on small sample transfer learning and computer equipment |
| CN112800344A (en) * | 2021-01-29 | 2021-05-14 | 重庆邮电大学 | Deep neural network-based movie recommendation method |
| CN112800344B (en) * | 2021-01-29 | 2022-03-22 | 重庆邮电大学 | Deep neural network-based movie recommendation method |
| CN113010802A (en) * | 2021-03-25 | 2021-06-22 | 华南理工大学 | Recommendation method facing implicit feedback based on multi-attribute interaction of user and article |
| CN113342963A (en) * | 2021-04-29 | 2021-09-03 | 山东大学 | Service recommendation method and system based on transfer learning |
| CN113342963B (en) * | 2021-04-29 | 2022-03-04 | 山东大学 | Service recommendation method and system based on transfer learning |
| CN113590964A (en) * | 2021-08-04 | 2021-11-02 | 燕山大学 | Deep neural network Top-N recommendation algorithm based on heterogeneous modeling |
| CN113590964B (en) * | 2021-08-04 | 2023-05-23 | 燕山大学 | Deep neural network Top-N recommendation method based on heterogeneous modeling |
| CN114581161B (en) * | 2022-05-06 | 2022-08-16 | 深圳市明珞锋科技有限责任公司 | Information pushing method and system based on deep learning |
| CN114581161A (en) * | 2022-05-06 | 2022-06-03 | 深圳市明珞锋科技有限责任公司 | Information pushing method and system based on deep learning |
| CN115841366A (en) * | 2022-12-30 | 2023-03-24 | 中国科学技术大学 | Article recommendation model training method and device, electronic equipment and storage medium |
| CN115841366B (en) * | 2022-12-30 | 2023-08-29 | 中国科学技术大学 | Method and device for training object recommendation model, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109241440A (en) | It is a kind of based on deep learning towards implicit feedback recommended method | |
| Al-Gasawneh et al. | The relationship between perceived destination image, social media interaction and travel intentions relating to Neom city | |
| KR102107848B1 (en) | Book Recommendation Service Method | |
| CN107580704A (en) | Context personage recommends | |
| CN108363804A (en) | Partial model Weighted Fusion Top-N films based on user clustering recommend method | |
| CN107077486A (en) | Affective Evaluation system and method | |
| CN103917968A (en) | System and method for managing opinion networks with interactive opinion flows | |
| CN110162717A (en) | A kind of method and apparatus of commending friends | |
| CN110321291A (en) | Test cases intelligent extraction system and method | |
| CN104166668A (en) | News recommendation system and method based on FOLFM model | |
| CN103198086A (en) | Information processing device, information processing method, and program | |
| CN108694647A (en) | A kind of method for digging and device of trade company's rationale for the recommendation, electronic equipment | |
| CN111709810A (en) | Object recommendation method and device based on recommendation model | |
| CN103678518A (en) | Method and device for adjusting recommendation lists | |
| CN111428127B (en) | Personalized event recommendation method and system integrating theme matching and bidirectional preference | |
| CN113706251B (en) | Model-based commodity recommendation method, device, computer equipment and storage medium | |
| CN108053050A (en) | Clicking rate predictor method, device, computing device and storage medium | |
| Evripidou et al. | Quaestio-it. com: a social intelligent debating platform | |
| CN109902229A (en) | A kind of interpretable recommended method based on comment | |
| Maity et al. | Analysis and prediction of question topic popularity in community Q&A sites: a case study of Quora | |
| CN107885846A (en) | Recommend method in a kind of knowledge point excavated based on implicit attribute and implicit relationship | |
| CN109885776A (en) | Recommended models can be explained in open source community PR reviewer | |
| CN112699310A (en) | Cold start cross-domain hybrid recommendation method and system based on deep neural network | |
| CN104572915B (en) | One kind is based on the enhanced customer incident relatedness computation method of content environment | |
| Cai et al. | An extension of social network group decision-making based on trustrank and personas |
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: 20190118 |
|
| RJ01 | Rejection of invention patent application after publication |