CN110717100A - Context perception recommendation method based on Gaussian embedded representation technology - Google Patents

Context perception recommendation method based on Gaussian embedded representation technology Download PDF

Info

Publication number
CN110717100A
CN110717100A CN201910940873.3A CN201910940873A CN110717100A CN 110717100 A CN110717100 A CN 110717100A CN 201910940873 A CN201910940873 A CN 201910940873A CN 110717100 A CN110717100 A CN 110717100A
Authority
CN
China
Prior art keywords
context
user
gaussian
mean
article
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201910940873.3A
Other languages
Chinese (zh)
Other versions
CN110717100B (en
Inventor
王东京
张新
俞东进
邵逸凡
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hangzhou Dianzi University
Hangzhou Electronic Science and Technology University
Original Assignee
Hangzhou Electronic Science and Technology University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Hangzhou Electronic Science and Technology University filed Critical Hangzhou Electronic Science and Technology University
Priority to CN201910940873.3A priority Critical patent/CN110717100B/en
Publication of CN110717100A publication Critical patent/CN110717100A/en
Application granted granted Critical
Publication of CN110717100B publication Critical patent/CN110717100B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

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

Abstract

The invention discloses a context-aware recommendation method based on a Gaussian embedded representation technology. The invention comprises the following steps: the method comprises the steps of collecting context perception behavior records of a user and attributes of articles, learning interest vector representation of the user, feature vector representation of the articles and the like from collected data by utilizing a Gaussian embedding representation technology, and finally implementing recommendation based on the feature vectors obtained through learning. The method mainly utilizes a Gaussian embedding representation technology to accurately model the dynamics of the user interest and the multi-dynamics of the article characteristics, fully utilizes the context information of the user and the attributes of the articles, improves the recommendation accuracy rate and further improves the user satisfaction.

Description

Context perception recommendation method based on Gaussian embedded representation technology
Technical Field
The invention belongs to the technical field of data mining, information retrieval and recommendation, and particularly relates to a context-aware recommendation method based on a Gaussian embedded representation technology.
Background
The recommendation system can help a user to quickly find out the content in which the user is interested from mass data, so that the problem of information overload is relieved. Existing recommendation methods can be broadly classified into collaborative filtering-based and content-based recommendation methods. Most collaborative filtering systems (e.g., matrix factorization) use separate feature vectors to represent users and items, but these vector representations correspond to a single fixed point in a low dimensional space, and cannot model dynamics and uncertainty in the recommendation system. In fact, the user's interests are dynamically changing, for example, a user who likes to listen to rock music and light music tends to rock music while in motion and tends to listen to light music while at rest. Also, items tend to have composite features including multiple attributes, such as a science fiction movie having both science fiction and comedy attributes. Furthermore, the interaction behavior of the user and the item is often influenced by the context. Therefore, how to effectively model and calculate the similarity according to the dynamic interest of the user and the multi-modal characteristics of the item, and how to accurately fuse the context information is the key to realize more accurate recommendation.
Disclosure of Invention
Aiming at the technical problems in the prior art, the invention provides a context-aware recommendation method based on a Gaussian-embedding expression technology, which can improve the recommendation effect and performance.
The invention comprises the following steps:
10. a context-aware behavioral record of the user and attributes of the item are collected.
20. The user's interest vector representation, feature vector representation of the item, etc. are learned from the collected data using gaussian embedded representation techniques.
30. And implementing recommendation based on the interest and the feature vector obtained by learning.
Wherein the step 10 comprises:
101. the behavior records of all users are collected and represented as a "user-item-context" triple set: r { (U, i, c) | user U has an interaction record with item i under context c }, U ═ U { (U, c) |, U { (U, i, c) |1，u2，...，u|U|Is the set of all users in the dataset, I ═ U |, I ═ I1，i2，...，i|I|Is the set of all | I | items.
102. Collecting attribute data A of all the articles in the article set I, including information such as article categories, article labels and the like, wherein the attribute of the article I is represented as
Figure BDA0002222856890000029
Wherein step 20 comprises:
201. item imAnd attribute a thereofkThe attribution relationship between them is modeled as the integral of the product of the gaussian probability density functions:
Figure BDA0002222856890000021
wherein.
Figure BDA0002222856890000022
A vector x representing a gaussian distribution conforming to mean and covariance as mu and sigma respectively,and
Figure BDA0002222856890000024
is an article imThe mean and covariance matrices of the feature vectors of (a),and
Figure BDA0002222856890000026
is attribute akThe covariance matrix is a diagonal matrix defined as ∑ diag (σ), where ∑ diag (σ) is the mean of the eigenvectors and the covariance matrix
Figure BDA0002222856890000027
Is the element vector at the diagonal position of the covariance matrix.
202. Similarly, according to the behavior records of all users, the user ulIn context cnLower and article imThe interaction records between are modeled as:
Figure BDA0002222856890000028
wherein.
Figure BDA0002222856890000031
A vector x representing a gaussian distribution conforming to mean and covariance as mu and sigma respectively,
Figure BDA0002222856890000032
and
Figure BDA0002222856890000033
is user ulThe mean and covariance matrices of the feature vectors of (a),
Figure BDA0002222856890000034
is a context cnThe vector of the transformation of (a) is,
Figure BDA0002222856890000035
and
Figure BDA0002222856890000036
is an article imThe covariance matrix is a diagonal matrix defined as ∑ diag (σ), where ∑ diag (σ) is the mean of the eigenvectors and the covariance matrix
Figure BDA0002222856890000037
Is the element vector at the diagonal position of the covariance matrix.
202. According to the above formula, the following objective function O is established:
O＝Ouic+Oia
wherein: o isuicAnd OiaRespectively an interaction objective function and an attribute objective function. O isuicIs defined as follows:
Figure BDA0002222856890000038
wherein:
Figure BDA0002222856890000039
is the interval at which the positive (u, i, c) and negative (u, i', c) samples are separated. In the same way, OiaIs defined as follows:
Figure BDA00022228568900000310
wherein: ω is the interval separating the positive sample (i, a) and the negative sample (i, a'). By maximizing the objective function O, the parameter values in the model, including the mean of the feature vectors of the item i, can be learned
Figure BDA00022228568900000311
Sum covariance matrix
Figure BDA00022228568900000312
Attribute akMean of feature vectors ofSum covariance matrix
Figure BDA00022228568900000314
User ulMean of feature vectors of
Figure BDA00022228568900000315
Sum covariance matrix
Figure BDA00022228568900000316
And a translation vector v for context cc。
Wherein step 30 comprises:
301. according to the representation of the learned gaussian feature vector, the interest of the target user u in the item i under the context c can be represented by the integral of the product of the gaussian density functions, which is specifically as follows:
wherein the content of the first and second substances,
Figure BDA00022228568900000318
and
Figure BDA00022228568900000319
is the eigenvector mean and covariance matrix, v, of user ucIs the translation vector for the context c,
Figure BDA00022228568900000320
and
Figure BDA00022228568900000321
is an article imThe eigenvector mean and covariance matrix.
302. And sorting all candidate articles by using the calculation result of the formula, and recommending the first articles to the target user u.
The invention has the beneficial effects that:
1) and accurately modeling the dynamics of the user interest and the multi-dynamics of the article characteristics by utilizing a Gaussian embedding representation technology.
2) The context information of the user and the attributes of the articles are fully utilized, the recommendation accuracy is improved, and the user satisfaction is further improved.
Drawings
FIG. 1 is a schematic diagram of a Gaussian embedding representation technique according to the present invention.
FIG. 2 is a schematic diagram of a recommendation method according to the present invention.
Detailed Description
In order to more specifically describe the present invention, the following detailed description is provided for the technical solution of the present invention with reference to the accompanying drawings and the specific embodiments.
The context perception recommendation method based on the Gaussian embedded representation technology comprises the following steps:
(1) collect behavior records for all users and represent as a "user-item-context" triple set: r { (U, i, c) | user U has an interaction record with item i under context c }, U ═ U { (U, c) |, U { (U, i, c) |1，u2，...，u|U|Is the set of all users in the dataset, I ═ U |, I ═ I1，i2，...，i|I|Is the set of all | I | items. Collecting attribute data A of all the articles in the article set I, including article category and articleItem tag, wherein the set of attributes of item i is represented as
Figure BDA0002222856890000041
(2) Will article imAnd attribute a thereofkIs modeled as an integral of the product between corresponding Gaussian probability density functions Wherein
Figure BDA0002222856890000044
A vector x representing a gaussian distribution conforming to mean and covariance as mu and sigma respectively,
Figure BDA0002222856890000045
andis an article imThe mean and covariance matrices of the feature vectors of (a),
Figure BDA0002222856890000047
and
Figure BDA0002222856890000048
is attribute akThe covariance matrix is a diagonal matrix defined as ∑ diag (σ), where ∑ diag (σ) is the mean of the eigenvectors and the covariance matrix
Figure BDA0002222856890000051
Is a vector composed of elements at diagonal positions of the covariance matrix.
(3) According to the behavior records and context information of all users, user ulIn context cnLower and article imBetween as integrals of the products between Gaussian probability density functions
Figure BDA0002222856890000052
Wherein
Figure BDA0002222856890000053
And
Figure BDA0002222856890000054
is user ulThe mean and covariance matrices of the feature vectors of (a),
Figure BDA0002222856890000055
is a context cnThe vector of the transformation of (a) is,
Figure BDA0002222856890000056
andis an article imThe covariance matrix is a diagonal matrix defined as ∑ diag (σ), where
Figure BDA0002222856890000058
Is the element vector at the diagonal position of the covariance matrix.
(4) According to the above formula, the following objective function O is established:
wherein.
Figure BDA00022228568900000510
Is the interval at which the positive sample (u, i, c) and the negative sample (u, i ', c) are separated, and ω is the interval at which the positive sample (i, a) and the negative sample (i, a') are separated. By maximizing the objective function O, the conversion vectors of the items, attributes, gaussian feature vector representation of the user, and context in the model can be learned.
(5) According to the Gaussian feature vector representation obtained by learning, the interest of the target user u in the item i under the context c can be multiplied by a Gaussian density functionThe integral is expressed as
Figure BDA00022228568900000511
Figure BDA00022228568900000512
Wherein
Figure BDA00022228568900000513
And
Figure BDA00022228568900000514
is the mean and covariance matrix, v, of the eigenvectors of user ucIs the translation vector for the context c,
Figure BDA00022228568900000515
and
Figure BDA00022228568900000516
is the eigenvector mean and covariance matrix for item i.
(6) Sorting all candidate items by using the calculation result of the formula, and recommending the first items to the target user u.
Fig. 1 is a schematic diagram showing a gaussian embedding representation technique in the present embodiment. The technology models users, articles and attributes into eigenvectors conforming to Gaussian distribution, and models contexts into eigenvectors; on the basis, a probability density function is utilized to model a context-aware interaction behavior 'user-context-item', and a containing relation 'item-attribute' of the item and the attribute is modeled; finally, the Gaussian feature vector representation of each article, attribute, user and the conversion vector of the context are obtained.
Fig. 2 shows the framework of the context-aware recommendation method based on the gaussian embedded representation technique in this embodiment. The recommendation method is divided into two main modules: the online recommendation system comprises an offline preprocessing module and an online recommendation module. In the preprocessing module, firstly, data of all users including behavior records, context information and attributes of articles are obtained, and then Gaussian feature vector representation and context feature vectors of the users, the articles and the attributes are obtained by utilizing Gaussian embedded representation technology learning. In a recommending module, potential interest of a prediction target user on candidate items is represented and ranked based on the learned Gaussian feature vectors, and finally, a plurality of items ranked at the top are recommended to the user.
The embodiments described above are intended to facilitate one of ordinary skill in the art in understanding and using the invention. It will be readily apparent to those skilled in the art that various modifications to the above-described implementations may be made, and the generic principles described herein may be applied to other implementations without the use of inventive faculty. Therefore, the present invention is not limited to the above embodiments, and those skilled in the art should make improvements and modifications to the present invention based on the disclosure of the present invention within the protection scope of the present invention.

Claims (4)

1. The context perception recommendation method based on the Gaussian embedded representation technology comprises the following steps:
10. collecting context-aware behavior records of a user and attributes of an item;
20. learning the interest vector representation of the user and the feature vector representation of the article from the collected data by utilizing a Gaussian embedding representation technology;
30. and implementing recommendation based on the interest and the feature vector obtained by learning.
2. The context-aware recommendation method based on gaussian embedded representation technique according to claim 1, wherein step 10 comprises:
101. behavior records for all users are collected and represented as a "user-item-context" triple set: r { (U, i, c) | user U has an interaction record with item i under context c }, U ═ U { (U, c) |, U { (U, i, c) |1，u2，...，u|U|Is the set of all | U | users in the dataset, I ═ I1，i2，...，i|I|Is the set of all | I | items;
102. collecting attribute data A of all articles in the article set I, including article types and article labels, wherein the articlesThe attribute of i is represented as
Figure FDA0002222856880000011
3. The context-aware recommendation method based on gaussian embedded representation technique according to claim 1, wherein step 20 comprises:
201. item imAnd attribute a thereofkThe attribution relationship between them is modeled as the integral of the product of the gaussian probability density functions:
Figure FDA0002222856880000012
wherein:
Figure FDA0002222856880000021
a vector x representing a gaussian distribution conforming to mean and covariance as mu and sigma respectively,
Figure FDA0002222856880000022
and
Figure FDA0002222856880000023
is an article imThe mean and covariance matrices of the feature vectors of (a),and
Figure FDA0002222856880000025
is attribute akThe covariance matrix is a diagonal matrix defined as ∑ diag (σ), where ∑ diag (σ) is the mean of the eigenvectors and the covariance matrix
Figure FDA0002222856880000026
Is the element vector at the diagonal position of the covariance matrix;
202. similarly, according to the behavior records of all users, the user ulIn context cnLower and article imThe interaction records between are modeled as:
Figure FDA0002222856880000027
wherein:
Figure FDA0002222856880000028
and
Figure FDA0002222856880000029
is user ulThe mean and covariance matrices of the feature vectors of (a),
Figure FDA00022228568800000210
is a context cnThe vector of the transformation of (a) is,
Figure FDA00022228568800000211
and
Figure FDA00022228568800000212
is an article imThe mean value of the eigenvector and the covariance matrix are all diagonal matrixes;
202. according to the formula, the following objective function O is established, and the conversion vectors of the articles, the attributes, the Gaussian feature vector representation of the user and the context are obtained through learning by maximizing the objective function O:
O＝Ouic+Oia
wherein: o isuicAnd OiaRespectively an interaction objective function and an attribute objective function; o isuicIs defined as follows:
Figure FDA00022228568800000213
wherein:
Figure FDA00022228568800000214
is the interval at which the positive (u, i, c) and negative (u, i', c) samples are separated; in the same way, OiaIs defined as follows:
Figure FDA00022228568800000215
wherein: ω is the interval separating the positive sample (i, a) and the negative sample (i, a'); by maximizing the objective function O, learning to obtain parameter values in the model, including the mean value of the feature vectors of the articles i
Figure FDA00022228568800000216
Sum covariance matrix
Figure FDA00022228568800000217
Attribute akMean of feature vectors of
Figure FDA00022228568800000218
Sum covariance matrix
Figure FDA00022228568800000219
User ulMean of feature vectors of
Figure FDA00022228568800000220
Sum covariance matrix
Figure FDA00022228568800000221
And a translation vector v for context cc。
4. The context-aware recommendation method based on gaussian embedded representation technique according to claim 1, wherein step 30 comprises:
301. according to the representation of the learned Gaussian feature vector, the interest of the target user u in the article i under the context c is represented by the integral of the product of the Gaussian density function, which is specifically as follows:
wherein the content of the first and second substances,
Figure FDA0002222856880000032
and
Figure FDA0002222856880000033
is the eigenvector mean and covariance matrix, v, of user ucIs the translation vector for the context c,
Figure FDA0002222856880000034
andis an article imThe mean and covariance matrices of the feature vectors;
302. and sorting all candidate articles by using the calculation result of the formula, and recommending the first articles to the target user u.
CN201910940873.3A 2019-09-30 2019-09-30 Context perception recommendation method based on Gaussian embedded representation technology Active CN110717100B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910940873.3A CN110717100B (en) 2019-09-30 2019-09-30 Context perception recommendation method based on Gaussian embedded representation technology

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910940873.3A CN110717100B (en) 2019-09-30 2019-09-30 Context perception recommendation method based on Gaussian embedded representation technology

Publications (2)

Publication Number Publication Date
CN110717100A true CN110717100A (en) 2020-01-21
CN110717100B CN110717100B (en) 2021-09-28

Family

ID=69212102

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910940873.3A Active CN110717100B (en) 2019-09-30 2019-09-30 Context perception recommendation method based on Gaussian embedded representation technology

Country Status (1)

Country Link
CN (1) CN110717100B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112464100B (en) * 2020-12-14 2023-04-28 未来电视有限公司 Information recommendation model training method, information recommendation method, device and equipment

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100070871A1 (en) * 2008-09-12 2010-03-18 International Business Machines Corporation Extendable Recommender Framework for Web-Based Systems
CN106326483A (en) * 2016-08-31 2017-01-11 华南理工大学 Collaborative recommendation method with user context information aggregation
CN107506419A (en) * 2017-08-16 2017-12-22 桂林电子科技大学 A kind of recommendation method based on heterogeneous context-aware
CN108108399A (en) * 2017-12-05 2018-06-01 华南理工大学 A kind of improved Collaborative Filtering Recommendation Algorithm of Gaussian modeling

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100070871A1 (en) * 2008-09-12 2010-03-18 International Business Machines Corporation Extendable Recommender Framework for Web-Based Systems
CN106326483A (en) * 2016-08-31 2017-01-11 华南理工大学 Collaborative recommendation method with user context information aggregation
CN107506419A (en) * 2017-08-16 2017-12-22 桂林电子科技大学 A kind of recommendation method based on heterogeneous context-aware
CN108108399A (en) * 2017-12-05 2018-06-01 华南理工大学 A kind of improved Collaborative Filtering Recommendation Algorithm of Gaussian modeling

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
DONGJING WANG等: ""Learning Music Embedding with Metadata for Context Aware Recommendation"", 《PROCEEDINGS OF THE 2016 ACM ON INTERNATIONAL CONFERENCE ON MULTIMEDIA RETRIEVAL》 *
余平刚: ""基于变分自编码器的带属性网络表示学习与深度嵌入聚类"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112464100B (en) * 2020-12-14 2023-04-28 未来电视有限公司 Information recommendation model training method, information recommendation method, device and equipment

Also Published As

Publication number Publication date
CN110717100B (en) 2021-09-28

Similar Documents

Publication Publication Date Title
CN107506480B (en) Double-layer graph structure recommendation method based on comment mining and density clustering
CA2817103C (en) Learning tags for video annotation using latent subtags
Caicedo et al. Multimodal representation, indexing, automated annotation and retrieval of image collections via non-negative matrix factorization
CN111291261A (en) Cross-domain recommendation method integrating label and attention mechanism and implementation system thereof
CN110674407A (en) Hybrid recommendation method based on graph convolution neural network
CN108563690B (en) Collaborative filtering recommendation method based on object-oriented clustering
CN109947987B (en) Cross collaborative filtering recommendation method
CN106815310A (en) A kind of hierarchy clustering method and system to magnanimity document sets
CN107818183B (en) Three-stage combined party building video recommendation method based on feature similarity measurement
CN113158023A (en) Public digital life accurate classification service method based on mixed recommendation algorithm
CN112529638B (en) Service demand dynamic prediction method and system based on user classification and deep learning
CN103778206A (en) Method for providing network service resources
CN112052387A (en) Content recommendation method and device and computer readable storage medium
CN111125540A (en) Recommendation method integrating knowledge graph representation learning and bias matrix decomposition
CN111651678A (en) Knowledge graph-based personalized recommendation method
CN113761359A (en) Data packet recommendation method and device, electronic equipment and storage medium
CN110990670B (en) Growth incentive book recommendation method and recommendation system
WO2020147259A1 (en) User portait method and apparatus, readable storage medium, and terminal device
CN110795613A (en) Commodity searching method, device and system and electronic equipment
CN110083766B (en) Query recommendation method and device based on meta-path guiding embedding
CN110717100B (en) Context perception recommendation method based on Gaussian embedded representation technology
CN108153818B (en) Big data based clustering method
Wang et al. Task-dependent and query-dependent subspace learning for cross-modal retrieval
CN113343077A (en) Personalized recommendation method and system integrating user interest time sequence fluctuation
TW201243627A (en) Multi-label text categorization based on fuzzy similarity and k nearest neighbors

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
EE01 Entry into force of recordation of patent licensing contract
EE01 Entry into force of recordation of patent licensing contract

Application publication date: 20200121

Assignee: ZHEJIANG ANDA SYSTEM ENGINEERING Co.,Ltd.

Assignor: HANGZHOU DIANZI University

Contract record no.: X2022980022900

Denomination of invention: Context aware recommendation based on gaussian embedded representation

Granted publication date: 20210928

License type: Common License

Record date: 20221124