CN109918615B - Multi-mode recommendation method and device - Google Patents
Multi-mode recommendation method and device Download PDFInfo
- Publication number
- CN109918615B CN109918615B CN201811587985.7A CN201811587985A CN109918615B CN 109918615 B CN109918615 B CN 109918615B CN 201811587985 A CN201811587985 A CN 201811587985A CN 109918615 B CN109918615 B CN 109918615B
- Authority
- CN
- China
- Prior art keywords
- tensor
- matrix
- negative
- user
- sigma
- 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
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to the field of automatic recommendation systems, in particular to a multi-modal recommendation method, which comprises the following steps: constructing a multi-modal user space based on the attribute information of the user; constructing a multi-modal item space based on the attribute information of the item; obtaining a scoring tensor between a user and an item based on the multi-modal user space and the multi-modal item space; carrying out non-negative orthogonal tensor singular value decomposition on the scoring tensor to obtain an approximate tensor of the scoring tensor; and recommending the corresponding articles for the user based on the scoring results in the approximate tensor, and further adopting non-negative limitation to ensure that the finally obtained recommending scoring results are all non-negative values, so that the recommending effect is improved.
Description
Technical Field
The invention relates to the field of automatic recommendation systems, in particular to a multi-mode recommendation method and device.
Background
The emergence of the internet brings a large amount of data information nowadays, and the demand of users in the information era is met, however, the amount of information on the internet is greatly increased along with the rapid development of the network, so that users cannot obtain the part of information which is really useful for the users when facing a large amount of information, and the use efficiency of the information is reduced on the contrary, thereby causing the problem of information overload.
In order to solve the problem of information overload, a recommendation system is adopted to quickly recommend information suitable for a user, the recommendation system analyzes and calculates by analyzing interest preference of the user, and guides the user to find own information demand based on interest points of the user, so that the existing recommendation system not only can provide personalized service for the user, but also can establish close relation with the user.
At present, a tensor decomposition algorithm is often adopted to determine the incidence relation between data objects, and is often adopted in a recommendation system, and specific tensor decomposition algorithms include an SVD (singular value decomposition) algorithm and an HOSVD (high-order singular value decomposition) algorithm, but at present, the non-negative characteristic of a scoring system is often ignored by the recommendation algorithm based on tensor decomposition, that is, negative numbers exist in obtained scoring results, and the negative numbers are meaningless numerical values and are often discarded, so that the recommendation effect is influenced.
Therefore, how to avoid meaningless values in the scoring result and obtain more meaningful values to improve the recommendation effect is a technical problem to be solved urgently at present.
Disclosure of Invention
In view of the above, the present invention has been made to provide a multimodal recommendation method and apparatus that overcomes or at least partially solves the above problems.
In a first aspect, the present invention provides a multi-modal recommendation method, including:
constructing a multi-modal user space based on the attribute information of the user;
constructing a multi-modal item space based on the attribute information of the item;
obtaining a scoring tensor between a user and an item based on the multi-modal user space and the multi-modal item space;
carrying out non-negative orthogonal tensor singular value decomposition on the scoring tensor to obtain an approximate tensor of the scoring tensor;
and recommending the adaptive items for the user based on the scoring result in the approximate tensor.
Further, the obtaining a score tensor between the user and the item based on the multi-modal user space and the multi-modal item space specifically includes:
based on the multi-modal user space
The multi-modal item space
Obtaining a score tensor between a user and an item
Wherein, I 1 ，I 2 ，…，I M Attribute information of the user, J, respectively 1 ，J 2 ，…，J N Respectively, attribute information of the article.
Further, the attribute information of the user specifically includes at least one of a physical condition, a mood condition, a gender, an age, and a preference of the user; the attribute information of the article specifically includes at least one of a place of origin, a size, a type, and a use of the article.
Further, the performing non-negative orthogonal tensor singular value decomposition on the score tensor to obtain an approximate tensor of the score tensor specifically includes:
tensor to be scored
Homomorphic mapping to obtain low-order matrix
Performing non-negative matrix decomposition on the matrix TM, specifically: the matrix TM = UM x sigma M x VM, wherein UM >0, VM >0 and UM are first orthogonal matrixes after decomposition, VM is a second orthogonal matrix after decomposition, and sigma M is a non-negative diagonal matrix after decomposition;
respectively intercepting the first orthogonal matrix UM, the second orthogonal matrix VM and the non-negative diagonal matrix sigma M to obtain a corresponding first low-dimensional matrix UM ', a corresponding second low-dimensional matrix VM ' and a corresponding truncated diagonal matrix sigma M ';
mapping the first low-dimensional matrix UM 'into a first non-negative feature tensor U', mapping the second low-dimensional matrix VM 'into a second non-negative feature tensor V', and mapping the truncated diagonal matrix sigma M 'into a non-negative diagonal tensor sigma';
performing modular multiplication on the first non-negative feature tensor U ', the second non-negative feature tensor V ' and the non-negative diagonal tensor Sigma ' to obtain an approximate tensor of the scoring tensor
Further, the performing non-negative orthogonal tensor singular value decomposition on the score tensor to obtain an approximate tensor of the score tensor specifically includes:
tensor to be scored
Homomorphic mapping to obtain low-order matrix
Performing non-negative matrix factorization on the matrix TM, specifically: the matrix TM = UM x sigma M x VM, wherein UM >0, VM >0, UM is a decomposed first orthogonal matrix, VM is a decomposed second orthogonal matrix, and sigma M is a decomposed non-negative diagonal matrix;
respectively intercepting the first orthogonal matrix UM, the second orthogonal matrix VM and the non-negative diagonal matrix sigma M to obtain a corresponding first low-dimensional matrix UM ', a corresponding second low-dimensional matrix VM ' and a corresponding truncated diagonal matrix sigma M ';
multiplying the first low-dimensional matrix UM ', the second low-dimensional matrix VM ' and the truncated diagonal matrix sigma M ' to obtain an approximate matrix
Obtaining an approximate tensor of the score tensor by homomorphic reverse mapping the approximate matrix
In a second aspect, the present invention provides a multi-modal recommendation device, including:
the first building module is used for building a multi-modal user space based on the attribute information of the user;
the second construction module is used for constructing a multi-mode article space based on the attribute information of the articles;
a score tensor obtaining module for obtaining a score tensor between the user and the item based on the multi-modal user space and the multi-modal item space;
the approximate tensor obtaining module is used for carrying out nonnegative orthogonal tensor singular value decomposition on the scoring tensor to obtain an approximate tensor of the scoring tensor;
and the recommending module is used for recommending the adaptive articles for the user based on the scoring result in the approximate tensor.
Further, the score tensor obtaining module is specifically configured to obtain the score tensor based on the multi-modal user space
The multi-modal item space
Obtaining a scoring tensor between a user and an item
Wherein, I 1 ，I 2 ，…，I M Attribute information of the users, respectively, J 1 ，J 2 ，…，J N Respectively, attribute information of the article.
Further, the attribute information of the user specifically includes at least one of a physical condition, a mood condition, a gender, an age, and a preference of the user; the attribute information of the article specifically includes at least one of a place of production, a size, a type, and an application of the article.
In a third aspect, the present invention further provides a computer device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the multi-modal recommendation method when executing the program.
In a fourth aspect, the present invention further provides a computer-readable storage medium, on which a computer program is stored, which when executed by a processor, implements the steps of the above multi-modal recommendation method.
One or more technical solutions in the embodiments of the present invention have at least the following technical effects or advantages:
the invention provides a multi-modal recommendation method, which comprises the following steps: constructing a multi-mode user space based on the attribute information of the user, constructing a multi-mode article space based on the attribute information of the article, and acquiring a score tensor between the user and the article based on the multi-mode user space and the multi-mode article space; carrying out non-negative orthogonal tensor singular value decomposition on the scoring tensor to obtain an approximate tensor of the scoring tensor; based on the scoring result in the approximate tensor, the adaptive articles are recommended for the user, the problem that the scoring result obtained through a recommending algorithm in the prior art has negative number and influences the recommending effect is solved, and then non-negative limitation is adopted, so that the finally obtained recommending scoring result is guaranteed to be non-negative, and the recommending effect is improved.
Drawings
Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the invention. Also, like reference numerals are used to refer to like parts throughout the drawings. In the drawings:
FIG. 1 shows a flow chart of the steps of a multi-modal recommendation method in an embodiment of the invention;
FIG. 2 is a schematic flow chart illustrating a first tensor decomposition in an embodiment of the present invention;
FIG. 3 is a schematic flow chart illustrating a second tensor decomposition in an embodiment of the present invention;
FIG. 4 is a block diagram of a multi-modal recommender in accordance with an embodiment of the present invention;
fig. 5 shows a schematic structural diagram of a computer device in an embodiment of the present invention.
Detailed Description
Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
Example one
An embodiment of the present invention provides a multi-modal recommendation method, as shown in fig. 1, including: s101, constructing a multi-modal user space based on attribute information of a user; s102, constructing a multi-mode article space based on the attribute information of the articles; s103, obtaining a score tensor between the user and the item based on the multi-mode user space and the multi-mode item space; s104, performing non-negative orthogonal tensor singular value decomposition on the scoring tensor to obtain an approximate tensor of the scoring tensor; and S105, recommending an adaptive item for the user based on the scoring result in the approximate tensor.
In a specific embodiment, in S101, S102, a multi-modal user space and a multi-modal item space are first constructed. Specifically, a multi-modal user space is constructed based on attribute information of a user, wherein the attribute information of the user specifically comprises at least one of physical condition, mood condition, gender, age and preference of the user; and constructing a multi-modal item space based on the attribute information of the item, wherein the attribute information of the item specifically comprises at least one of the origin, the size, the type and the application of the item. S101 and S102 are not sequentially determined, that is, may be performed simultaneously, and are not specifically limited in the embodiment of the present invention.
The multi-dimensional features of the user are fully described through the multi-modal user attribute information, and the multi-dimensional features of the article are fully described through the multi-modal article attribute information.
Next, in S103, a score tensor between the user and the item is obtained based on the multi-modal user space and the multi-modal item space. In particular, based on the multi-modal user space
Multi-modal item space
Obtaining users and itemsScore tensor of (a) between
Wherein, I 1 ，I 2 ，…，I M Attribute information of the users, respectively, J 1 ，J 2 ，…，J N Respectively, attribute information of the article.
Due to the scoring tensor
The score result in (1) is sparse and null values exist, so the score tensor needs to be decomposed to eliminate these effects.
Specifically, by executing S104, the non-negative orthogonal tensor singular value decomposition is performed on the score tensor, and the approximate tensor of the score tensor is obtained.
The specific process of singular value decomposition of the non-negative orthogonal tensor can adopt two modes.
First decomposition, as shown in fig. 2:
s201, scoring tensor
Homomorphic mapping to obtain low-order matrix
S202 is then performed.
S202, carrying out non-negative matrix decomposition on the matrix TM, specifically: the matrix TM = UM x sigma M x VM, wherein UM >0, VM >0 and UM are first orthogonal matrixes after decomposition, VM is a second orthogonal matrix after decomposition, and sigma M is a non-negative diagonal matrix after decomposition; s203 is then performed.
S203, respectively intercepting the first orthogonal matrix UM, the second orthogonal matrix VM and the nonnegative diagonal matrix sigma M to obtain a corresponding first low-dimensional matrix UM ', a corresponding second low-dimensional matrix VM ' and a corresponding truncated diagonal matrix sigma M '; s204 is then performed.
S204, adopting homomorphic reverse mapping to map the first low-dimensional matrix UM 'into a first non-negative feature tensor U', map the second low-dimensional matrix VM 'into a second non-negative feature tensor V', and map the truncated diagonal matrix sigma M 'into a non-negative diagonal sigma'; s205 is then executed.
S205, performing modular multiplication on the first non-negative feature tensor U ', the second non-negative feature tensor V ' and the non-negative diagonal tensor Sigma ' to obtain an approximate tensor of the scoring tensor
The approximate tensor thus obtained presents scoring results that eliminate the situation where the data is sparse and empty without tensor decomposition.
Second decomposition, as shown in fig. 3:
s301, scoring tensor
Homomorphic mapping to obtain low-order matrix
S302 is then performed.
S302, carrying out non-negative matrix decomposition on the matrix TM, specifically: the matrix TM = UM x sigma M x VM, wherein UM >0, VM >0 and UM are decomposed first orthogonal matrixes, VM is decomposed second orthogonal matrixes, and sigma M is decomposed non-negative diagonal matrixes; s303 is then performed.
S303, respectively intercepting the first orthogonal matrix UM, the second orthogonal matrix VM and the nonnegative diagonal matrix sigma M to obtain a corresponding first low-dimensional matrix UM ', a corresponding second low-dimensional matrix VM ' and a corresponding truncated diagonal matrix sigma M '; s304 is then performed.
S304, multiplying the first low-dimensional matrix UM ', the second low-dimensional matrix VM ' and the truncated diagonal matrix sigma M ' to obtain an approximate matrix TM ' = UM ' × sigma M ' × VM '; s305 is then performed.
S305, the approximation matrix is mapped reversely through the homomorphism, and the approximate tensor of the scoring tensor is obtained
In the second tensor decomposition, firstly, the matrix obtained after decomposition is intercepted, the intercepted matrix is multiplied to obtain an approximate matrix, and finally, the approximate matrix is subjected to homomorphic reverse mapping to obtain an approximate tensor. In the first tensor decomposition, firstly, the matrix obtained after decomposition is intercepted, the intercepted matrix is respectively subjected to homomorphic reverse mapping to obtain tensors corresponding to the matrix after decomposition, and the tensors are subjected to modular multiplication to finally obtain an approximate tensor. The final result obtained by both tensor decompositions is the same.
After the approximate tensor is obtained, S105 is executed, and based on the scoring result in the approximate tensor, an adaptive item is recommended for the user.
In a specific implementation manner, the items corresponding to the score results larger than the preset threshold in the score results in the approximation tensor can be recommended to the corresponding users, and in the recommendation process, the items can be recommended according to the values of the score results from large to small, wherein the items can be entity items suitable for the users, such as clothes, food, and the like, and can also be news or advertisements and the like push texts conforming to the interests of the users, which is not described in detail in the embodiment of the present invention.
Therefore, by adopting the technical scheme of the invention, the tensor is decomposed and under the condition of non-negative constraint, the condition that negative numbers appear in the scoring tensor is avoided, the scoring results in the scoring tensor are all meaningful numerical values, and the recommendation effect is improved.
One or more technical solutions in the embodiments of the present invention have at least the following technical effects or advantages:
the invention provides a multi-modal recommendation method, which comprises the following steps: constructing a multi-mode user space based on the attribute information of the user, constructing a multi-mode article space based on the attribute information of the article, and acquiring a score tensor between the user and the article based on the multi-mode user space and the multi-mode article space; carrying out non-negative orthogonal tensor singular value decomposition on the scoring tensor to obtain an approximate tensor of the scoring tensor; based on the scoring result in the approximate tensor, the adaptive article is recommended for the user, the problem that in the prior art, the recommending effect is influenced due to the fact that the scoring result obtained through a recommending algorithm is negative is solved, and therefore non-negative limitation is adopted, the fact that the finally obtained recommending scoring result is non-negative is guaranteed, and the recommending effect is improved.
Example two
Based on the same inventive concept, an embodiment of the present invention further provides a multi-modal recommendation apparatus, as shown in fig. 4, including:
a first constructing module 401, configured to construct a multi-modal user space based on attribute information of a user;
a second construction module 402 for constructing a multi-modal item space based on attribute information of the item;
a score tensor obtaining module 403, configured to obtain a score tensor between a user and an item based on the multi-modal user space and the multi-modal item space;
an approximate tensor obtaining module 404, configured to perform non-negative orthogonal tensor singular value decomposition on the score tensor to obtain an approximate tensor of the score tensor;
and a recommending module 405, configured to recommend an appropriate item for the user based on the scoring result in the approximate tensor.
Preferably, the score tensor obtaining module 403 is specifically configured to base the multi-modal user space on
The multi-modal item space
Obtaining a scoring tensor between a user and an item
Wherein, I 1 ，I 2 ，…，I M Attribute information of the user, J, respectively 1 ，J 2 ，…，J N Respectively, attribute information of the article.
Preferably, the attribute information of the user specifically includes at least one of a physical condition, a mood condition, a gender, an age, and a preference of the user; the attribute information of the article specifically includes at least one of a place of origin, a size, a type, and a use of the article.
Preferably, the approximate tensor obtaining module 404 specifically includes:
a first mapping unit for tensor of score
Homomorphic mapping to obtain low-order matrix
A first decomposition unit, configured to perform non-negative matrix decomposition on the matrix TM, specifically: the matrix TM = UM x sigma M x VM, wherein UM >0, VM >0 and UM are first orthogonal matrixes after decomposition, VM is a second orthogonal matrix after decomposition, and sigma M is a non-negative diagonal matrix after decomposition;
the first interception unit is used for respectively intercepting the first orthogonal matrix UM, the second orthogonal matrix VM and the non-negative diagonal matrix sigma M to obtain a corresponding first low-dimensional matrix UM ', a corresponding second low-dimensional matrix VM ' and an intercepted diagonal matrix sigma M ';
a homomorphic direction mapping unit, configured to map the first low-dimensional matrix UM 'into a first non-negative feature tensor U', map the second low-dimensional matrix VM 'into a second non-negative feature tensor V', and map the truncated diagonal matrix Σ M 'into a non-negative diagonal tensor Σ'
A first approximate tensor obtaining unit, configured to perform modular multiplication on the first non-negative feature tensor U ', the second non-negative feature tensor V ', and the non-negative diagonal tensor Σ ', to obtain an approximate tensor of the score tensor
Preferably, the approximate tensor obtaining module 404 specifically includes:
a second mapping unit for mapping the score tensor
Homomorphic mapping to obtain low-order matrix
The second decomposition unit is configured to perform non-negative matrix decomposition on the matrix TM, specifically: the matrix TM = UM x sigma M x VM, wherein UM >0, VM >0 and UM are decomposed first orthogonal matrixes and VM is decomposed second orthogonal matrixes, and sigma M is decomposed non-negative diagonal matrixes;
the second interception unit is used for respectively intercepting the first orthogonal matrix UM, the second orthogonal matrix VM and the non-negative diagonal matrix sigma M to obtain a corresponding first low-dimensional matrix UM ', a corresponding second low-dimensional matrix VM ' and a corresponding truncated diagonal matrix sigma M ';
an approximate matrix obtaining unit, configured to obtain an approximate matrix TM ' = UM ' × Σ M ' × VM ', by multiplying the first low-dimensional matrix UM ', the second low-dimensional matrix VM ', and the truncated diagonal matrix Σ M ';
a second approximate tensor obtaining unit, configured to obtain an approximate tensor of the score tensor by homomorphic reverse mapping the approximate matrix
EXAMPLE III
Based on the same inventive concept, the embodiment of the present invention further provides a computer device, which includes a memory 504, a processor 502 and a computer program stored on the memory 504 and capable of running on the processor 502, wherein the processor 502 implements the steps of a multi-modal recommendation method when executing the program.
Where in fig. 5 a bus architecture (represented by bus 500) is shown, bus 500 may include any number of interconnected buses and bridges, and bus 500 links together various circuits including one or more processors, represented by processor 502, and memory, represented by memory 504. The bus 500 may also link together various other circuits such as peripherals, voltage regulators, power management circuits, and the like, which are well known in the art, and therefore, will not be described any further herein. A bus interface 506 provides an interface between the bus 500 and the receiver 501 and transmitter 503. The receiver 501 and the transmitter 503 may be the same element, i.e. a transceiver, providing a means for communicating with various other apparatus over a transmission medium. The processor 502 is responsible for managing the bus 500 and general processing, and the memory 504 may be used for storing data used by the processor 502 in performing operations.
Example four
Based on the same inventive concept, the embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, and the program, when executed by a processor, implements the steps of a multi-modal recommendation method.
While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including preferred embodiments and all such alterations and modifications as fall within the scope of the invention.
It will be apparent to those skilled in the art that various changes and modifications may be made in the present invention without departing from the spirit and scope of the invention. Thus, if such modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include such modifications and variations.
Claims (7)
1. A multi-modal recommendation method, comprising:
constructing a multi-modal user space based on the attribute information of the user;
constructing a multi-modal item space based on the attribute information of the item;
obtaining a scoring tensor between a user and an item based on the multi-modal user space and the multi-modal item space;
carrying out non-negative orthogonal tensor singular value decomposition on the scoring tensor to obtain an approximate tensor of the scoring tensor;
recommending adaptive articles for the user based on the scoring result in the approximate tensor;
the obtaining a score tensor between the user and the item based on the multi-modal user space and the multi-modal item space specifically includes:
based on the multi-modal user space
The multi-modal item space
Obtaining a scoring tensor between a user and an item
Wherein, I 1 ，I 2 ，L，I M Attribute information of the users, respectively, J 1 ，J 2 ，L，J N Respectively are attribute information of the articles;
the performing non-negative orthogonal tensor singular value decomposition on the score tensor to obtain an approximate tensor of the score tensor specifically includes:
tensor to be scored
Homomorphic mapping to obtain low-order matrix
Performing non-negative matrix decomposition on the matrix TM, specifically: the matrix TM = UM x Σ M × VM, wherein UM>0，VM>0,UM is a decomposed first orthogonal matrix, VM is a decomposed second orthogonal matrix, and sigma M is a decomposed non-negative diagonal matrix;
respectively intercepting the first orthogonal matrix UM, the second orthogonal matrix VM and the non-negative diagonal matrix sigma M to obtain a corresponding first low-dimensional matrix UM ', a corresponding second low-dimensional matrix VM ' and a corresponding truncated diagonal matrix sigma M ';
mapping the first low-dimensional matrix UM 'into a first non-negative feature tensor U', mapping the second low-dimensional matrix VM 'into a second non-negative feature tensor V', and mapping the truncated diagonal matrix sigma M 'into a non-negative diagonal tensor sigma';
performing modular multiplication on the first non-negative feature tensor U ', the second non-negative feature tensor V ' and the non-negative diagonal tensor Sigma ' to obtain an approximate tensor of the scoring tensor
2. The method according to claim 1, wherein the attribute information of the user specifically includes at least one of a physical condition, a mood condition, a gender, an age, and a preference of the user; the attribute information of the article specifically includes at least one of a place of origin, a size, a type, and a use of the article.
3. The method of claim 1, wherein the performing non-orthonormal tensor singular value decomposition on the score tensor to obtain an approximate tensor of the score tensor, further comprises:
tensor to be scored
Homomorphic mapping to obtain low-order matrix
Performing non-negative matrix decomposition on the matrix TM, specifically: the matrix TM = UM x sigma M x VM, wherein UM >0, VM >0 and UM are first orthogonal matrixes after decomposition, VM is a second orthogonal matrix after decomposition, and sigma M is a non-negative diagonal matrix after decomposition;
respectively intercepting the first orthogonal matrix UM, the second orthogonal matrix VM and the non-negative diagonal matrix sigma M to obtain a corresponding first low-dimensional matrix UM ', a corresponding second low-dimensional matrix VM ' and a corresponding truncated diagonal matrix sigma M ';
multiplying the first low-dimensional matrix UM ', the second low-dimensional matrix VM ' and the truncated diagonal matrix sigma M ' to obtain an approximate matrix TM ' = UM ' × sigma M ' × VM ';
obtaining an approximate tensor of the scoring tensor by homomorphic reverse mapping the approximate matrix
4. A multimodal recommendation apparatus, comprising:
the first building module is used for building a multi-modal user space based on the attribute information of the user;
the second building module is used for building a multi-mode article space based on the attribute information of the articles;
a score tensor obtaining module for obtaining a score tensor between the user and the item based on the multi-modal user space and the multi-modal item space;
the approximate tensor obtaining module is used for carrying out nonnegative orthogonal tensor singular value decomposition on the scoring tensor to obtain an approximate tensor of the scoring tensor;
the recommending module is used for recommending the adaptive articles for the user based on the scoring result in the approximate tensor;
the score tensor obtaining module is specifically configured to obtain a score based on the multi-modal user space
The multi-modal item space
Obtaining a scoring tensor between a user and an item
Wherein, I 1 ，I 2 ，L，I M Attribute information of the users, respectively, J 1 ，J 2 ，L，J N Respectively are attribute information of the article;
the approximate tensor obtaining module is used for obtaining the score tensor
Homomorphic mapping to obtain low-order matrix
Performing non-negative matrix decomposition on the matrix TM, specifically: the matrix TM = UM x Σ M × VM, wherein UM>0，VM>0,UM is a decomposed first orthogonal matrix, VM is a decomposed second orthogonal matrix, and sigma M is a decomposed non-negative diagonal matrix; respectively intercepting the first orthogonal matrix UM, the second orthogonal matrix VM and the non-negative diagonal matrix sigma M to obtain a corresponding first low-dimensional matrix UM ', a corresponding second low-dimensional matrix VM ' and a corresponding truncated diagonal matrix sigma M '; mapping the first low-dimensional matrix UM 'into a first non-negative feature tensor U', mapping the second low-dimensional matrix VM 'into a second non-negative feature tensor V', and mapping the truncated diagonal matrix sigma M 'into a non-negative diagonal tensor sigma'; performing modular multiplication on the first non-negative feature tensor U ', the second non-negative feature tensor V ' and the non-negative diagonal tensor Sigma ' to obtain an approximate tensor of the scoring tensor
5. The apparatus of claim 4, wherein the attribute information of the user specifically includes at least one of a physical condition, a mood condition, a gender, an age, and a preference of the user; the attribute information of the article specifically includes at least one of a place of origin, a size, a type, and a use of the article.
6. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the method steps of any of claims 1-3 when executing the program.
7. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the method steps of any one of claims 1 to 3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811587985.7A CN109918615B (en) | 2018-12-25 | 2018-12-25 | Multi-mode recommendation method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811587985.7A CN109918615B (en) | 2018-12-25 | 2018-12-25 | Multi-mode recommendation method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109918615A CN109918615A (en) | 2019-06-21 |
| CN109918615B true CN109918615B (en) | 2022-12-27 |
Family
ID=66959929
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811587985.7A Active CN109918615B (en) | 2018-12-25 | 2018-12-25 | Multi-mode recommendation method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109918615B (en) |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2018128708A (en) * | 2017-02-06 | 2018-08-16 | 日本電信電話株式会社 | Tensor factor decomposition processing apparatus, tensor factor decomposition processing method and tensor factor decomposition processing program |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8346708B2 (en) * | 2009-01-22 | 2013-01-01 | Nec Laboratories America, Inc. | Social network analysis with prior knowledge and non-negative tensor factorization |
| US8452770B2 (en) * | 2010-07-15 | 2013-05-28 | Xerox Corporation | Constrained nonnegative tensor factorization for clustering |
-
2018
- 2018-12-25 CN CN201811587985.7A patent/CN109918615B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2018128708A (en) * | 2017-02-06 | 2018-08-16 | 日本電信電話株式会社 | Tensor factor decomposition processing apparatus, tensor factor decomposition processing method and tensor factor decomposition processing program |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109918615A (en) | 2019-06-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108491529B (en) | Information recommendation method and device | |
| KR102411921B1 (en) | A method for calculating relevance, an apparatus for calculating relevance, a data query apparatus, and a non-transitory computer-readable storage medium | |
| CN108287864B (en) | Interest group dividing method, device, medium and computing equipment | |
| CN109062914B (en) | User recommendation method and device, storage medium and server | |
| US8032535B2 (en) | Personalized web search ranking | |
| CA3047353A1 (en) | Learning document embeddings with convolutional neural network architectures | |
| WO2023134086A1 (en) | Convolutional neural network model pruning method and apparatus, and electronic device and storage medium | |
| CN111652378B (en) | Learning to select vocabulary for category features | |
| US11651255B2 (en) | Method and apparatus for object preference prediction, and computer readable medium | |
| US20220261591A1 (en) | Data processing method and apparatus | |
| CN111639970A (en) | Method for determining price of article based on image recognition and related equipment | |
| KR101635283B1 (en) | Method for analyzing data based on matrix factorization model and apparatus therefor | |
| US20220114644A1 (en) | Recommendation system with sparse feature encoding | |
| US9836450B2 (en) | Methods and systems for providing universal portability in machine learning | |
| CN110085292B (en) | Medicine recommendation method and device and computer-readable storage medium | |
| Kumar et al. | Parameterization reduction using soft set theory for better decision making | |
| CN114329004A (en) | Digital fingerprint generation method, digital fingerprint generation device, data push method, data push device and storage medium | |
| CN109918615B (en) | Multi-mode recommendation method and device | |
| CN111832782B (en) | Method and device for determining physical distribution properties of articles | |
| CN109886299B (en) | User portrait method and device, readable storage medium and terminal equipment | |
| CN115906863B (en) | Emotion analysis method, device, equipment and storage medium based on contrast learning | |
| CN104834958B (en) | A kind of method and apparatus judged the step of answer | |
| EP4398128A1 (en) | Recommendation method and related device | |
| CN116049567A (en) | Collaborative filtering-based fault inspection recommendation method and system | |
| CN114707638A (en) | Model training method, model training device, object recognition method, object recognition device, object recognition medium and product |
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 |