CN113434778B - Recommendation method based on regularization framework and attention mechanism - Google Patents
Recommendation method based on regularization framework and attention mechanism Download PDFInfo
- Publication number
- CN113434778B CN113434778B CN202110819327.1A CN202110819327A CN113434778B CN 113434778 B CN113434778 B CN 113434778B CN 202110819327 A CN202110819327 A CN 202110819327A CN 113434778 B CN113434778 B CN 113434778B
- Authority
- CN
- China
- Prior art keywords
- item
- sim
- value
- user
- score
- 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.)
- Active
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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/10—Office automation; Time management
- G06Q10/103—Workflow collaboration or project management
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/02—CAD in a network environment, e.g. collaborative CAD or distributed simulation
Abstract
A recommendation method based on a regularization frame and an attention mechanism comprises the steps of constructing a hypergraph network model, improving a similarity mode, determining a forgetting function, constructing the attention mechanism, and constructing and optimizing the regularization frame. The method has the advantages that the similarity relation between items, the scoring relation between users and the items and the marking relation between the items and labels are determined in the super-edge, the cold start problem in the recommendation method is relieved, the determination method of the similarity is improved, the most similar items are screened for the target items, the influence of time factors on the recommendation effect is considered, the real interest of the users is accurately represented, the attention mechanism is integrated into a regularization frame, the particularly relevant factors can be concentrated and other factors can be ignored, the problems of cold start, data sparseness and instantaneity in the recommendation method are relieved, the method has the advantages of high prediction and recommendation accuracy and the like, and can be used for network recommendation of commodities.
Description
Technical Field
The invention belongs to the technical field of networks, and particularly relates to commodity recommendation.
Background
With the development of the internet, information resources are explosively increased, and in the face of such huge data, it takes a lot of time for users to find the information needed by the users. In order to help users to find information, search engines appear, and since some users cannot accurately express their needs, recommendation technologies are generated, and become a focus of attention of many researchers. The recommendation system is beneficial to helping the user solve the problem of information overload, and items which are possibly interested by the user are screened from a large amount of information through the implicit characteristics and the explicit characteristics of the user. The recommendation system is used for associating users with information, on one hand, the users are helped to find valuable information for themselves, on the other hand, the information can be presented to interested users, and therefore the win-win situation of information consumers and information producers is achieved. At present, various recommendation methods are diversified, and can be mainly classified into 3 types, namely, a content-based recommendation method, a collaborative filtering recommendation method and a mixed recommendation method. The content-based recommendation method recommends items similar to the user's history likes to the user by analyzing attributes of the user and the items. Compared with a recommendation method based on content, the collaborative filtering recommendation method focuses more on researching the relationships between users and users, between items and between users and between items. These methods assume that the user will like the same or similar items to a large extent as users with similar hobbies. A hybrid recommendation is a combination of content-based recommendations and collaborative filtering recommendations. Although these recommendations are widely used, they all suffer from cold start problems due to new users and new items in the data set, and in addition, from real-time and sparse data.
Disclosure of Invention
The technical problem to be solved by the present invention is to overcome the above disadvantages of the prior art, and to provide a recommendation method based on a regularization framework and an attention mechanism, which has high utility value and small attenuation degree.
The technical scheme adopted for solving the problems comprises the following steps:
(1) Building hypergraph network model
1) Building of a super edge
All the items evaluated by each user are selected from the Movielens data set and connected by using the excess edges, and the number of the excess edges is the same as that of the users.
2) Determining weights for hyper-edges
Determining the weight W (e) of the excess edge according to equation (1) a )：
In the formula, delta (e) a ) Is the degree of the a-th over edge, δ (e) max ) Is the maximum degree of over-edge, δ (e) min ) Is the minimum degree of over-edge.
3) Establishing relevance of item and label
And establishing the relation between the item and the label by adopting a bipartite graph, calculating the marking times, and when the marking times of the label to the item is more than 2/3 of the total number of the marked items, having strong correlation between the item and the label.
(2) Improving similarity mode
The similarity Sim (i, j) of the items is determined by equation (2):
Sim(i,j)＝wSim cosine (i,j)+(1-w)[w 1 Sim 1 (i,j)+(1-w 1 )Sim 2 (i,j)] (2)
in the formula Sim cosine (i, j) is the cosine similarity of item i and item j, sim cosine (i, j) has a value in the range of [0, 1%]，Sim 1 (i, j) is the similarity of the attributes of item i and item j based on likes, sim 2 (i, j) is the dislike-based attribute similarity of item i and item j, sim 1 (i, j) and Sim 2 (i, j) has a value in the range of [0,1 ]]I and j are finite positive integers, w and w 1 Is 2 fusion factors, w 1 ∈[0,1]Determining the cosine similarity Sim of the item according to equation (3) cosine (i,j)：
D ij ＝{u 1 ,u 2 ,...,u p }
In the formula R ui Is the user u's score for item i, R uj Is the score of the user u on the item j, and p is a finite positive integer。
Determining the attribute similarity Sim of an item according to equation (4) 1 (i, j) and attribute similarity Sim 2 (i,j)：
In the formula, LA i,s Is the attraction of item i to the favorite attribute s, LA j,s Is the attraction of item j to the favorite attribute s,
is the average attraction of item i to the like attribute, based on the average value of the attraction>
Is the average attraction of item j to the like attribute, DLA i,s Is the attraction of item i to dislike property s, DLA j,s Is the attraction of item j to disliked attribute s, based on>
Is the average attraction of item i to the dislike attribute, based on the value of the item>
Is the average attraction of item j to dislike attributes, and k is the number of attributes, a finite positive integer.
(3) Determining a forgetting function
Determining a forgetting function f (t) according to equation (5) ui )：
In the formula t ui Represents the time of the user u's score on item i, t min Indicating the earliest time, t, in the user's historical access log max Representing the latest time in the user's historical access log.
(4) Construction of attention mechanism
Constructing an attention mechanism Att (u, i) according to the formula (6):
in the formula N i Is the nearest neighbor of item i, R ui Is the raw rating, R, of user u for item i uj Is the raw score of user u on item j, f (t) ui ) Is the forgetting degree of the user u to the item i, f (t) uj ) Is the degree of forgetting of item j by user u.
(5) Building and optimizing regularization framework
Construction and optimization of regularization framework Q according to equation (7) u (f,g)：
Where α, β, μ, σ are 4 parameters, α, β, μ, σ are positive numbers, w (e) is the weight of the excess edge e, h (i, e) is an element in the correlation matrix of the node and the excess edge, the value is 1 if the node i is in the excess edge e, otherwise 0, h (j, e) is an element in the correlation matrix of the node and the excess edge, the value is 1 if the node j is in the excess edge e, otherwise 0, δ (e) is the degree of the excess edge e, f is the degree of the node e, and f is the degree of the excess edge e i Is the score of item i, f j Is the score of item j, g l Is the score of label l, z ii Number of times item i is marked, z ll Number of times the item is marked for tag l, y i Is the initial score of item i, s l And optimizing the initial score of the label l by adopting a gradient descent method to obtain a prediction score and a recommendation list.
In the invention of (2) pairsIn formula (2) in which the improvement step is carried out in a similar manner, said w and w 1 Is 2 fusion factors, w is optimally 0.3 1 Is preferably 0.7.
In the step (5) of the present invention, the value ranges of α, β, μ, and σ are (0, 1), and α + β + μ + σ =1.
In the step (5), the value of alpha is optimally 0.2, the value of beta is optimally 0.07, the value of mu is optimally 0.7, and the value of sigma is optimally 0.03.
Because the invention adopts the hypergraph network model, the similarity relation between items, the scoring relation between users and items and the marking relation between items and labels are determined in the hyperedges, the cold start problem in the recommendation method is relieved, the determination method of the similarity is improved, the most similar items are screened for the target items, the influence of time factors on the recommendation effect is considered, the time information and the scoring information are combined, the dynamic change of the user interest is embodied, the real interest of the users can be accurately expressed, and the attention mechanism is integrated in the regularization frame, so that the invention can be focused on particularly relevant factors and neglect other factors. Compared with the prior art, the method has higher accuracy of prediction and recommendation, relieves the problems of cold start, data sparseness and instantaneity in the recommendation method to a certain extent, and can be used for network recommendation of commodities.
Drawings
FIG. 1 is a flowchart of example 1 of the present invention.
Detailed Description
The present invention will be described in further detail below with reference to the drawings and examples, but the present invention is not limited to the embodiments described below.
Example 1
Taking 100000 data in the Movielens-100k data set as an example, the recommendation method based on the regularization framework and attention mechanism of the present embodiment is composed of the following steps (see fig. 1):
(1) Constructing hypergraph network model
1) Building of a super edge
943 super edges are selected from 100000 data sets of Movielens-100k, all items evaluated by each user are selected from the data sets and connected by the super edges, and the number of the super edges is the same as that of the users u. The similarity relation between items, the grading relation between users and the items and the marking relation between the items and the labels are determined in the super edges, and the cold start problem in the recommendation method is relieved.
2) Determining weights for hyper-edges
Determining the weight W (e) of the excess edge according to equation (1) a )：
In the formula, delta (e) a ) Is the degree of the a-th over edge, δ (e) max ) Is the maximum degree of over-edging, δ (e) min ) In this embodiment, the maximum excess margin value is 686 and the minimum excess margin value is 5.
3) Establishing relevance of item and label
And establishing the relation between the item and the label by adopting a bipartite graph, calculating the marking times, and when the marking times of the label to the item is more than 2/3 of the total number of the marked items, having strong correlation between the item and the label.
(2) Improving similarity mode
The similarity Sim (i, j) of the items is determined by equation (2):
Sim(i,j)＝wSim cosine (i,j)+(1-w)[w 1 Sim 1 (i,j)+(1-w 1 )Sim 2 (i,j)] (2)
in the formula Sim cosine (i, j) is the cosine similarity of item i and item j, sim cosine (i, j) has a value in the range of [0,1 ]]，Sim 1 (i, j) item i and item j are based on likeness of the liked attributes, sim 2 (i, j) item i and item j are based on dislike attribute similarity, sim 1 (i, j) and Sim 2 (i, j) has a value in the range of [0,1 ]]I and j are finite positive integers, w and w 1 Is 2 fusion factors, w 1 ∈[0,1]In this embodiment, w is 0.3 1 Is 0.7. Determining cosine similarity S of items according to equation (3)im cosine (i,j)：
D ij ＝{u 1 ,u 2 ,...,u p }
In the formula R ui Is the user u's score for item i, R uj Is the user u's score for item j, and p is a finite positive integer.
Determining the attribute similarity Sim of an item according to equation (4) 1 (i, j) and attribute similarity Sim 2 (i,j)：
In the formula, LA i,s Is the attraction of item i to the favorite attribute s, LA j,s Is the attraction of item j to the favorite attribute s,
is the average attraction of item i to the like attribute, based on the value of the item>
Is the average attraction of item j to the like attribute, DLA i,s Is the attraction of item i to dislike property s, DLA j,s Is the attraction of item j to the disliked attribute s, <' >>
Is the average attraction of item i to the dislike attribute, based on the value of the item>
Is the average attraction of the item j to the dislike attribute, k is the number of the attributes and is a finite positive integer, and k in this embodiment takes the value of 26. And the similarity mode is improved, so that the most similar items can be screened for the target items.
(3) Determining a forgetting function
Determining a forgetting function f (t) according to equation (5) ui )：
In the formula t ui Represents the time of the user u's score on item i, t min Indicating the earliest time, t, in the user's historical access records max Represents the latest time in the user's historical access record, t of this embodiment min The value is 874724727,t max The value is 893286638. The influence of time factors on the recommendation effect is considered, dynamic change of user interest is reflected, and the real interest of the user is accurately expressed.
(4) Construction of attention mechanism
Constructing an attention mechanism Att (u, i) according to a formula (6):
in the formula N i Is the nearest neighbor of item i, R ui Is the raw rating, R, of user u for item i uj Is the raw score of user u on item j, f (t) ui ) Is the forgetting degree of the user u to the item i, f (t) uj ) Is the degree of forgetting of item j by user u.
(5) Building and optimizing regularization framework
Construction and optimization of regularization framework Q according to equation (7) u (f,g)：
Where α, β, μ, and σ are 4 parameters, α, β, μ, and σ are positive numbers, α in this embodiment is 0.2, β is 0.1, μ is 0.5, σ is 0.2, w (e) is a weight of a super edge e, h (i, e) is an element in an association matrix of a node and the super edge, if the node i is in the super edge e, the value is 1, otherwise, 0, h (j, e) is an element in the association matrix of the node and the super edge, if the node j is in the super edge e, the value is 1, otherwise, δ (e) is a degree of the super edge e, f, μ, and σ are positive numbers, f (i, e), and δ (i, e) is a positive number i Is the score of item i, f j Is the score of item j, g l Is the score of label l, z ii Number of times item i is marked, z ll Number of times the item is marked for tag l, y i Is the initial score of item i, s l And optimizing the initial score of the label l by adopting a gradient descent method to obtain a prediction score and a recommendation list. And an attention mechanism is integrated into the regularization frame, so that the regularization frame can focus on particularly relevant factors and ignore other factors, and a better recommendation effect can be achieved.
Example 2
Taking 100000 data in the Movielens-100k data set as an example, the recommendation method based on the regularization framework and attention mechanism of the present embodiment comprises the following steps:
(1) Building hypergraph network model
This procedure is the same as in example 1.
(2) Improving similarity mode
In formula (2) in this step, w and w 1 Is 2 fusion factors, w 1 ∈[0,1]In this embodiment, w is 0 1 Is 0.
The other steps of this step are the same as in example 1.
(3) Determining a forgetting function
This procedure is the same as in example 1.
(4) Construction of attention mechanism
This procedure is the same as in example 1.
(5) Building and optimizing regularization framework
Constructed and superior according to the formula (7)Regularization framework Q u (f,g)：
Where α, β, μ, and σ are 4 parameters, α, β, μ, and σ are positive numbers, α in this embodiment takes a value of 0.4, β takes a value of 0.17, μ takes a value of 0.4, σ takes a value of 0.03, w (e) is a weight of a super edge e, h (i, e) is an element in a correlation matrix of a node and a super edge, if the node i is in the super edge e, the value is 1, otherwise 0, h (j, e) is an element in the correlation matrix of the node and the super edge, if the node j is in the super edge e, the value is 1, otherwise 0, δ (e) is a degree of the super edge e, f (j, σ) is a positive number, h (i, g (i, e) is a positive number of the node and (e), and w (i, g) is a positive number of the value of the node and the super edge e i Is the score of item i, f j Is the score of item j, g l Is the score of label l, z ii Number of times item i is marked, z ll Number of times item is marked for tag l, y i Is an initial score, s, of item i l And optimizing the initial score of the label l by adopting a gradient descent method to obtain a prediction score and a recommendation list.
The other steps of this procedure were the same as in example 1.
Example 3
Taking 100000 data in the Movielens-100k data set as an example, the recommendation method based on the regularization framework and attention mechanism of the present embodiment comprises the following steps:
(1) Building hypergraph network model
This procedure is the same as in example 1.
(2) Improving similarity mode
In formula (2) in this step, w and w 1 Is 2 fusion factors, w 1 ∈[0,1]In this embodiment, w is 1,w 1 Is 1.
The other steps of this step are the same as in example 1.
(3) Determining a forgetting function
This procedure is the same as in example 1.
(4) Build attention mechanism
This procedure is the same as in example 1.
(5) Building and optimizing regularization framework
Construction and optimization of regularization framework Q according to equation (7) u (f,g)：
Where α, β, μ, and σ are 4 parameters, α, β, μ, and σ are positive numbers, α in this embodiment takes a value of 0.25, β takes a value of 0.07, μ takes a value of 0.65, σ takes a value of 0.03, w (e) is a weight of a super edge e, h (i, e) is an element in a correlation matrix of a node and a super edge, if the node i is in the super edge e, the value is 1, otherwise 0, h (j, e) is an element in the correlation matrix of the node and the super edge, if the node j is in the super edge e, the value is 1, otherwise 0, δ (e) is a degree of the super edge e, f (j, σ) is a positive number, h (i, g (i, e) is a positive number of the node, and δ (i, g) is 0.07 i Is the score of item i, f j Is the score of item j, g l Is the score of label l, z ii Number of times item i is marked, z ll Number of times the item is marked for tag l, y i Is an initial score, s, of item i l And optimizing the initial score of the label l by adopting a gradient descent method to obtain a prediction score and a recommendation list. Incorporating an attention mechanism into the regularization framework to focus on particularly relevant factors while ignoring other factors, canBetter recommendation effect can be achieved.
Claims (3)
1. A recommendation method based on a regularization framework and an attention mechanism is characterized by comprising the following steps of:
(1) Building hypergraph network model
1) Building of a super edge
Selecting all the items evaluated by each user from the Movielens data set, and connecting the items by using the excess edges, wherein the number of the excess edges is the same as that of the users;
2) Determining weights for hyper-edges
Determining the weight W (e) of the overcenter according to equation (1) a )：
In the formula, delta (e) a ) Is the degree of the a-th over edge, δ (e) max ) Is the maximum degree of over-edge, δ (e) min ) The degree of the minimum excess edge;
3) Establishing relevance of item and label
Establishing the relation between the item and the label by adopting a bipartite graph, calculating the marking times, and when the marking times of the label to the item is more than 2/3 of the total marked number of the item, having strong correlation between the item and the label;
(2) Improving similarity mode
The similarity Sim (i, j) of the items is determined by equation (2):
Sim(i,j)＝wSim cosine (i,j)+(1-w)[w 1 Sim 1 (i,j)+(1-w 1 )Sim 2 (i,j)] (2)
in the formula Sim cosine (i, j) is the cosine similarity of item i and item j, sim cosine (i, j) has a value in the range of [0,1 ]]，Sim 1 (i, j) is the similarity of the attributes of item i and item j based on likes, sim 2 (i, j) is the dislike-based attribute similarity of item i and item j, sim 1 (i, j) and Sim 2 (i, j) has a value in the range of [0,1 ]]I and j are finite positive integers, w and w 1 Is a fusion factor of 2, and the fusion factor is a fusion factor of 2,w,w 1 ∈[0,1]determining the cosine similarity Sim of the item according to equation (3) cosine (i,j)：
D ij ＝{u 1 ,u 2 ,...,u p }
In the formula R ui Is the user u's score for item i, R uj Is the score of the user u on the item j, and p is a limited positive integer;
determining the attribute similarity Sim of an item according to equation (4) 1 (i, j) and attribute similarity Sim 2 (i,j)：
In the formula, LA i,s Is the attraction of item i to the favorite attribute s, LA j,s Is the attraction of item j to the favorite attribute s,
is the average attraction of item i to the like attribute,
is the average attraction of item j to the like attribute, DLA i,s Is the attraction of item i to dislike property s, DLA j,s Is the attraction of item j to disliked property s,
is the average attraction of item i to the dislike property,
is the average attraction of item j to dislike attributes, and k is the number of attributes, a finite positive integer;
(3) Determining a forgetting function
Determining a forgetting function f (t) according to equation (5) ui )：
In the formula t ui Represents the time of the user u's score on item i, t min Indicating the earliest time, t, in the user's historical access log max Representing the latest time in the user's historical access records;
(4) Construction of attention mechanism
Constructing an attention mechanism Att (u, i) according to a formula (6):
in the formula N i Is the nearest neighbor of item i, R ui Is the raw rating, R, of user u for item i uj Is the raw score of user u on item j, f (t) ui ) Is the forgetting degree of the user u to the item i, f (t) uj ) Is the degree of forgetting of item j by user u;
(5) Building and optimizing regularization framework
Construction and optimization of regularization framework Q according to equation (7) u (f,g)：
In the formula, alpha, beta, mu and sigma are 4 parameters, and the value ranges of alpha, beta, mu and sigmaA circumference of (0, 1), and α + β + μ + σ =1, w (e) is the weight of the super-edge e, h (i, e) is an element in the incidence matrix of the node and the super-edge, the value is 1 if the node i is in the super-edge e, otherwise 0, h (j, e) is the element in the correlation matrix of the node and the super edge, if node j is in super edge e, the value is 1, otherwise 0, δ (e) is the degree of super edge e, f i Is the score of item i, f j Is the score of item j, g l Is the score of label l, z ii Number of times item i is marked, z ll Number of times the item is marked for tag l, y i Is an initial score, s, of item i l And optimizing the initial score of the label l by adopting a gradient descent method to obtain a prediction score and a recommendation list.
2. The regularization framework and attention mechanism based recommendation method as claimed in claim 1, wherein: in the formula (2) in which the similarity pattern improvement step is performed, w and w are 1 Is 2 fusion factors, w takes the value of 0.3 1 Is 0.7.
3. The regularization framework and attention mechanism based recommendation method according to claim 1, wherein: and (5) the value of alpha is 0.2, the value of beta is 0.07, the value of mu is 0.7, and the value of sigma is 0.03.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110819327.1A CN113434778B (en) | 2021-07-20 | 2021-07-20 | Recommendation method based on regularization framework and attention mechanism |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110819327.1A CN113434778B (en) | 2021-07-20 | 2021-07-20 | Recommendation method based on regularization framework and attention mechanism |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113434778A CN113434778A (en) | 2021-09-24 |
| CN113434778B true CN113434778B (en) | 2023-03-24 |
Family
ID=77761129
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110819327.1A Active CN113434778B (en) | 2021-07-20 | 2021-07-20 | Recommendation method based on regularization framework and attention mechanism |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113434778B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114117142A (en) * | 2021-12-02 | 2022-03-01 | 南京邮电大学 | Label perception recommendation method based on attention mechanism and hypergraph convolution |
| CN114707075B (en) * | 2022-06-06 | 2022-10-25 | 荣耀终端有限公司 | Cold start recommendation method and device |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109670121A (en) * | 2018-12-18 | 2019-04-23 | 辽宁工程技术大学 | Project level and feature level depth Collaborative Filtering Recommendation Algorithm based on attention mechanism |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2015016784A1 (en) * | 2013-08-01 | 2015-02-05 | National University Of Singapore | A method and apparatus for tracking microblog messages for relevancy to an entity identifiable by an associated text and an image |
| CN109241424B (en) * | 2018-08-29 | 2019-08-27 | 陕西师范大学 | A kind of recommended method |
| US11443346B2 (en) * | 2019-10-14 | 2022-09-13 | Visa International Service Association | Group item recommendations for ephemeral groups based on mutual information maximization |
| AU2020101604A4 (en) * | 2020-07-31 | 2020-09-10 | The University of Xinjiang | A Recommendation with Item Cooccurrence based on Metric Factorization |
| CN112613602A (en) * | 2020-12-25 | 2021-04-06 | 神行太保智能科技(苏州)有限公司 | Recommendation method and system based on knowledge-aware hypergraph neural network |
-
2021
- 2021-07-20 CN CN202110819327.1A patent/CN113434778B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109670121A (en) * | 2018-12-18 | 2019-04-23 | 辽宁工程技术大学 | Project level and feature level depth Collaborative Filtering Recommendation Algorithm based on attention mechanism |
Non-Patent Citations (1)
| Title |
|---|
| 基于双向条目注意网络的推荐系统;赵永建等;《广东工业大学学报》;20200714(第04期);31-38 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113434778A (en) | 2021-09-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110162706B (en) | Personalized recommendation method and system based on interactive data clustering | |
| CN109241405B (en) | Learning resource collaborative filtering recommendation method and system based on knowledge association | |
| CN104835072B (en) | Method and system for compatibility scoring of users in a social network | |
| CN106156127B (en) | Method and device for selecting data content to push to terminal | |
| CN103678618B (en) | Web service recommendation method based on socializing network platform | |
| CN103186539B (en) | A kind of method and system determining user group, information inquiry and recommendation | |
| CN100465954C (en) | Reinforced clustering of multi-type data objects for search term suggestion | |
| CN113434778B (en) | Recommendation method based on regularization framework and attention mechanism | |
| CN108717407B (en) | Entity vector determination method and device, and information retrieval method and device | |
| US20100293174A1 (en) | Query classification | |
| CN103678635A (en) | Network music aggregation recommendation method based on label digraphs | |
| CN104572733B (en) | The method and device of user interest labeling | |
| Chen | RETRACTED ARTICLE: Research on personalized recommendation algorithm based on user preference in mobile e-commerce | |
| CN109597899B (en) | Optimization method of media personalized recommendation system | |
| CN109816015B (en) | Recommendation method and system based on material data | |
| Stanescu et al. | A hybrid recommender system: User profiling from keywords and ratings | |
| CN108874916A (en) | A kind of stacked combination collaborative filtering recommending method | |
| CN108109058A (en) | A kind of single classification collaborative filtering method for merging personal traits and article tag | |
| Tran et al. | A comparison study for job recommendation | |
| US9020863B2 (en) | Information processing device, information processing method, and program | |
| CN108389113B (en) | Collaborative filtering recommendation method and system | |
| CN113591947A (en) | Power data clustering method and device based on power consumption behaviors and storage medium | |
| Liang et al. | Personalized recommender systems integrating social tags and item taxonomy | |
| Qi et al. | An inverse collaborative filtering approach for cold-start problem in web service recommendation | |
| CN108710648B (en) | Collaborative filtering recommendation method based on S-type improved similarity |
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 |