A tensor-based hybrid recommendation method and system

Abstract

The present invention provides a tensor-based hybrid recommendation method and system, belonging to the field of data processing technology, comprising: converting user-side information, project-side information, and time-side information into discrete data information; then inputting the information into user, project, and time tensor neural networks for feature extraction to obtain user latent factor matrices, project latent factor matrices, and time latent factor matrices; performing a paradigm multivariate decomposition on a third-order rating tensor to extract the latent factor matrices, thereby obtaining user latent factor matrices, project latent factor matrices, and time latent factor matrices; after fusing the two user latent factor matrices, project latent factor matrices, and time latent factor matrices, the fused factor matrices are restored using the inverse operation of the paradigm multivariate decomposition to obtain a final approximate rating matrix. The present invention, through a tensor-based hybrid recommendation algorithm, fully considers both linear and nonlinear information, thereby improving the accuracy of recommendation results.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a tensor-based hybrid recommendation method and system. Background Art

[0002] In the era of big data, application services are filled with massive amounts of information, making it difficult for users to effectively mine the high-quality information they need. The recommendation system can effectively solve the problem of "information overload" and accurately recommend potential product information of interest to users based on their preferences.

[0003] Traditional recommendation systems are mainly divided into three categories: collaborative filtering algorithm (CF), content-based recommendation algorithm (CB) and hybrid recommendation algorithm. Among them, the essential idea of ​​the hybrid recommendation algorithm is to combine multiple (two or more) recommendation algorithms to overcome the problems that may exist in a single recommendation algorithm, hoping to achieve the effect of improving the performance of the recommendation system by combining multiple recommendation algorithms. There are many categories of hybrid recommendation systems, such as weighted hybrid recommendation system, feature combination recommendation system, cascade hybrid recommendation system, switching hybrid recommendation system, feature enhancement recommendation system, meta-level hybrid recommendation and mixed hybrid recommendation system. Although the above methods have a place in today's applications, they generally have defects that are difficult to optimize:

[0004] 1) Data sparsity. Traditional recommendation algorithms still suffer from data sparsity. Data sparsity refers to the situation where a user only rates a small fraction of the total number of items. This paper proposes to alleviate this data sparsity problem by employing tensor decomposition.

[0005] 2) There is no unified modeling of high-order, multi-dimensional, and multi-source heterogeneous nonlinear information. Most existing recommendation systems do not fully utilize the potential high-order, multi-dimensional tensor information in nonlinear information, but only process and analyze linear information to derive recommendations for users. Summary of the Invention

[0006] The present invention provides a tensor-based hybrid recommendation method and system to solve the common cold start problem and data sparsity problem in the hybrid recommendation system in the prior art, which results in low recommendation accuracy.

[0007] In a first aspect, the present invention provides a tensor-based hybrid recommendation method, comprising:

[0008] Collect user-side information sets, project-side information sets, and time-side information sets;

[0009] Converting the user-side information set, the item-side information set, and the time-side information set into user characteristic information, item characteristic information, and time characteristic information, respectively;

[0010] Inputting the user feature information, the item feature information, and the time feature information into a user tensor neural network, an item tensor neural network, and a time tensor neural network, respectively, for feature extraction to obtain a first user latent factor matrix, a first item latent factor matrix, and a first time latent factor matrix;

[0011] Performing a normalized multivariate decomposition on a third-order scoring tensor formed by the user feature information, the item feature information, and the time feature information to obtain a second user latent factor matrix, a second item latent factor matrix, and a second time latent factor matrix;

[0012] Fusing the first user latent factor matrix with the second user latent factor matrix to obtain a fused user latent factor matrix, fusing the first item latent factor matrix with the second item latent factor matrix to obtain a fused item latent factor matrix, and fusing the first time latent factor matrix with the second time latent factor matrix to obtain a fused time latent factor matrix;

[0013] Calculating the fused user latent factor matrix, the fused project latent factor matrix, and the fused time latent factor matrix based on an inverse operation of the normal form multivariate decomposition to obtain an approximate rating matrix;

[0014] The approximate rating matrix is ​​used to output user recommendation results.

[0015] According to a tensor-based hybrid recommendation method provided by the present invention, the user-side information set includes user ID, gender, age and occupation, the project-side information set includes project ID, project type and project title, and the time-side information set includes time, user ID and project ID.

[0016] According to a tensor-based hybrid recommendation method provided by the present invention, the user feature information, the item feature information, and the time feature information are respectively input into a user tensor neural network, an item tensor neural network, and a time tensor neural network for feature extraction to obtain a first user latent factor matrix, a first item latent factor matrix, and a first time latent factor matrix, including:

[0017] The user feature information Input the user tensor neural network to obtain the first user potential factor matrix U tnn :

[0018]

[0019] User tensor weight tensor in neural network

[0020] Among them, U represents the number of users, Represents multiple user-side information in the user-side information set. The subscripts u1, u2, ..., and uM are the user-side information dimension identifiers respectively. n represents the number of tensor neural network layers. represents the activation functions of different user tensor neural network layers, and represents the offset tensor of the neural network layer at different user tensors, The superscript 1 represents the user type, O∈[1,n-1], where O represents the symbolic parameter that distinguishes different user tensor neural network layers;

[0021] The project feature information Input the project tensor neural network to obtain the first project potential factor matrix I tnn :

[0022]

[0023] Project tensor neural network weight tensor

[0024] Where I represents the number of items, Represents multiple project side information in the project side information set. The subscripts i1, i2, ..., iL are project side information dimension identifiers respectively. n represents the number of tensor neural network layers. represents the activation functions of the different project tensor neural network layers, and The offset tensor representing the different project tensor neural network layers, The superscript 2 represents the item type, Z∈[1,n-1], where Z represents the symbolic parameter that distinguishes different item tensor neural network layers;

[0025] The time characteristic information Input the time tensor neural network to obtain the first time potential factor matrix T tnn :

[0026]

[0027] Weight Tensor of Time Tensor Neural Network

[0028] Where T represents the number of times, Represents multiple time-side information in the time-side information set. The subscripts t1, t2, ..., tP are time-side information dimension identifiers respectively. n represents the number of tensor neural network layers. represents the activation functions of the different time tensor neural network layers, and represents the offset tensor of the neural network layer at different time tensors, The superscript 3 represents the time type, Q∈[1,n-1], where Q represents the symbolic parameter that distinguishes different time tensor neural network layers.

[0029] According to a tensor-based hybrid recommendation method provided by the present invention, a third-order rating tensor formed by the user feature information, the item feature information, and the time feature information is subjected to a normalized multivariate decomposition to obtain a second user latent factor matrix, a second item latent factor matrix, and a second time latent factor matrix, including:

[0030] For the third-order score tensor R o ∈R U×I×T Performing paradigm multivariate decomposition, we get:

[0031]

[0032] in, To reconstruct the approximate rating tensor, R cp is the rank of the normal form multivariate decomposition, ° represents the outer product operation, λ is the weight vector, λ r represents the rth element of the weight vector λ, U cp is the second user latent factor matrix, I cp is the latent factor matrix of the second item, T cp is the second time latent factor matrix, r is the user’s latent factor vector, i r is the project latent factor vector, t r is the temporal latent factor vector.

[0033] According to a tensor-based hybrid recommendation method provided by the present invention, the first user latent factor matrix and the second user latent factor matrix are fused to obtain a fused user latent factor matrix, the first item latent factor matrix and the second item latent factor matrix are fused to obtain a fused item latent factor matrix, and the first time latent factor matrix and the second time latent factor matrix are fused to obtain a fused time latent factor matrix, including:

[0034]

[0035]

[0036]

[0037] Among them, U cpnn To integrate the user potential factor matrix, I cpnn is the latent factor matrix of the fusion project, T cpnn is the fusion time latent factor matrix, It is a fusion operation.

[0038] According to a tensor-based hybrid recommendation method provided by the present invention, the fused user latent factor matrix, the fused item latent factor matrix, and the fused time latent factor matrix are calculated based on the inverse operation of the normal form multivariate decomposition to obtain an approximate rating matrix, including:

[0039]

[0040] in, is the approximate rating matrix, R cpnn for rank.

[0041] In a second aspect, the present invention further provides a tensor-based hybrid recommendation system, comprising:

[0042] A collection module, used to collect user-side information sets, project-side information sets, and time-side information sets;

[0043] A conversion module, configured to convert the user-side information set, the item-side information set, and the time-side information set into user characteristic information, item characteristic information, and time characteristic information, respectively;

[0044] a first processing module, configured to input the user feature information, the item feature information, and the time feature information into a user tensor neural network, an item tensor neural network, and a time tensor neural network, respectively, for feature extraction to obtain a first user latent factor matrix, a first item latent factor matrix, and a first time latent factor matrix;

[0045] a second processing module, configured to perform a normalized multivariate decomposition on a third-order rating tensor formed by the user feature information, the item feature information, and the time feature information to obtain a second user latent factor matrix, a second item latent factor matrix, and a second time latent factor matrix;

[0046] a fusion module, configured to fuse the first user latent factor matrix with the second user latent factor matrix to obtain a fused user latent factor matrix, fuse the first project latent factor matrix with the second project latent factor matrix to obtain a fused project latent factor matrix, and fuse the first time latent factor matrix with the second time latent factor matrix to obtain a fused time latent factor matrix;

[0047] a restoration module, configured to calculate the fused user latent factor matrix, the fused project latent factor matrix, and the fused time latent factor matrix based on an inverse operation of a normal form multivariate decomposition to obtain an approximate rating matrix;

[0048] The recommendation module is used to output user recommendation results using the approximate rating matrix.

[0049] In a third aspect, the present invention also provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the tensor-based hybrid recommendation method as described above is implemented.

[0050] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the tensor-based hybrid recommendation methods described above.

[0051] In a fifth aspect, the present invention further provides a computer program product, comprising a computer program, which, when executed by a processor, implements any of the tensor-based hybrid recommendation methods described above.

[0052] The tensor-based hybrid recommendation method and system provided by the present invention fully consider linear information and nonlinear information through a tensor-based hybrid recommendation algorithm, effectively solve the cold start problem and data sparsity problem, thereby improving the accuracy of the recommendation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0054] Figure 1 1 is a flow chart of the tensor-based hybrid recommendation method provided by the present invention;

[0055] Figure 2 Schematic diagram of the structure of the tensor-based hybrid recommendation system provided by the present invention;

[0056] Figure 3 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0057] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0058] In view of the many shortcomings of existing hybrid recommendation methods, and considering that nonlinear information contains a large amount of usable information between users and items, between users and users, and between items and items, the present invention proposes a hybrid recommendation algorithm of tensor decomposition and tensor neural network. By making full use of multi-source heterogeneous nonlinear information through tensor decomposition and tensor neural network, nonlinear information and linear information are uniformly modeled, thereby improving the recommendation accuracy of the recommendation system.

[0059] Figure 1 is a flow chart of a tensor-based hybrid recommendation method provided by an embodiment of the present invention, such as Figure 1 As shown, including:

[0060] Step 100: Collect user-side information set, project-side information set, and time-side information set;

[0061] Step 200: Convert the user-side information set, the item-side information set, and the time-side information set into user feature information, item feature information, and time feature information, respectively;

[0062] Step 300: Inputting the user feature information, the item feature information, and the time feature information into a user tensor neural network, an item tensor neural network, and a time tensor neural network, respectively, for feature extraction to obtain a first user latent factor matrix, a first item latent factor matrix, and a first time latent factor matrix;

[0063] Step 400: Performing a normalized multivariate decomposition on the third-order rating tensor formed by the user feature information, the item feature information, and the time feature information to obtain a second user latent factor matrix, a second item latent factor matrix, and a second time latent factor matrix;

[0064] Step 500: Fusing the first user latent factor matrix with the second user latent factor matrix to obtain a fused user latent factor matrix, fusing the first item latent factor matrix with the second item latent factor matrix to obtain a fused item latent factor matrix, and fusing the first time latent factor matrix with the second time latent factor matrix to obtain a fused time latent factor matrix.

[0065] Step 600: Calculate the fused user latent factor matrix, the fused item latent factor matrix, and the fused time latent factor matrix based on the inverse operation of the normal form multivariate decomposition to obtain an approximate rating matrix;

[0066] Step 700: Output user recommendation results using the approximate rating matrix.

[0067] Specifically, the embodiment of the present invention converts the collected user-side information, item-side information, and time-side information into corresponding discrete data information, namely, user feature information, item feature information, and time feature information, respectively, by using an embedding layer.

[0068] Further feature extraction is performed from two dimensions. First, the obtained user feature information, item feature information and time feature information are input into the user tensor neural network, item tensor neural network and time tensor neural network respectively in the embedding layer to obtain the first user latent factor matrix, the first item latent factor matrix and the first time latent factor matrix; then the third-order rating tensor formed by the user feature information, item feature information and time feature information is subjected to canonical polyadic decomposition (CP) to obtain the second user latent factor matrix, the second item latent factor matrix and the second time latent factor matrix.

[0069] Furthermore, the first user latent factor matrix, the first project latent factor matrix and the first time latent factor matrix extracted by the tensor neural network are fused with the second user latent factor matrix, the second project latent factor matrix and the second time latent factor matrix obtained by CP decomposition, respectively, to obtain a fused user latent factor matrix, a fused project latent factor matrix and a fused time latent factor matrix.

[0070] The fused latent factor matrix is ​​then restored through the inverse operation of CP decomposition to obtain an approximate rating matrix, which can be used to make recommendations and output user recommendation results.

[0071] The present invention adopts a tensor-based hybrid recommendation algorithm, fully considers linear information and nonlinear information, effectively solves the cold start problem and data sparsity problem, and thus improves the accuracy of recommendation results.

[0072] Based on the above embodiment, the user side information set includes user ID, gender, age and occupation, the project side information set includes project ID, project type and project title, and the time side information set includes time, user ID and project ID.

[0073] Based on the above embodiment, the embodiment of the present invention uses a tensor neural network to perform feature extraction on nonlinear linearity to obtain the corresponding user latent factor matrix, project latent factor matrix and time latent factor matrix.

[0074] Specifically, the user feature information, the project feature information, and the time feature information are respectively input into the user tensor neural network, the project tensor neural network, and the time tensor neural network for feature extraction to obtain a first user latent factor matrix, a first project latent factor matrix, and a first time latent factor matrix, including:

[0075] User feature information Input the user tensor neural network to obtain the first user potential factor matrix U tnn :

[0076]

[0077] User tensor weight tensor in neural network

[0078] Where I represents the number of users, Represents multiple user-side information in the user-side information set. The subscripts u1, u2, ..., and uM are the user-side information dimension identifiers respectively. n represents the number of tensor neural network layers. represents the activation functions of different user tensor neural network layers, and represents the offset tensor of the neural network layer at different user tensors, The superscript 1 represents the user type, O∈[1,n-1], where O represents the symbolic parameter that distinguishes different user tensor neural network layers;

[0079] Project feature information Input the item tensor neural network to obtain the first item potential factor matrix I tnn :

[0080]

[0081] Project tensor neural network weight tensor

[0082] Where I represents the number of items, Represents multiple project side information in the project side information set. The subscripts i1, i2, ..., iL are project side information dimension identifiers respectively. n represents the number of tensor neural network layers. represents the activation functions of the different project tensor neural network layers, and The offset tensor representing the different project tensor neural network layers, The superscript 2 represents the item type, Z∈[1,n-1], where Z represents the symbolic parameter that distinguishes different item tensor neural network layers;

[0083] Time feature information Input the time tensor neural network to obtain the first time potential factor matrix T tnn :

[0084]

[0085] Weight Tensor of Time Tensor Neural Network

[0086] Where T represents the number of times, Represents multiple time-side information in the time-side information set. The subscripts t1, t2, ..., tP are time-side information dimension identifiers respectively. n represents the number of tensor neural network layers. represents the activation functions of the different time tensor neural network layers, and represents the offset tensor of the neural network layer at different time tensors, The superscript 3 represents the time type, Q∈[1,n-1], where Q represents the symbolic parameter that distinguishes different time tensor neural network layers.

[0087] Based on the above embodiment, the embodiment of the present invention uses CP decomposition to extract the user latent factor matrix, item latent factor matrix and time latent factor matrix of the rating tensor (linear information).

[0088] It should be noted that CP decomposition is a commonly used high-order tensor decomposition method. Its basic idea is to represent an N-th order tensor as the sum of several rank-1 tensors.

[0089] Specifically, the third-order rating tensor formed by user feature information, item feature information, and time feature information is subjected to a paradigm multivariate decomposition to obtain a second user latent factor matrix, a second item latent factor matrix, and a second time latent factor matrix, including:

[0090] For the third-order scoring tensor R o ∈R U×I×T Performing paradigm multivariate decomposition, we get:

[0091]

[0092] in, To reconstruct the approximate rating tensor, R cpis the rank of the normal form multivariate decomposition, ° represents the outer product operation, λ is the weight vector, λ r represents the rth element of the weight vector λ, U cp is the second user latent factor matrix, I cp is the latent factor matrix of the second item, T cp is the second time latent factor matrix, r is the user’s latent factor vector, i r is the project latent factor vector, t r is the temporal latent factor vector.

[0093] Based on the above embodiment, the embodiment of the present invention fuses the feature information obtained by CP decomposition and the feature information obtained by the tensor neural network.

[0094] Specifically, the first user latent factor matrix and the second user latent factor matrix are fused to obtain a fused user latent factor matrix, the first project latent factor matrix and the second project latent factor matrix are fused to obtain a fused project latent factor matrix, and the first time latent factor matrix and the second time latent factor matrix are fused to obtain a fused time latent factor matrix, including:

[0095]

[0096]

[0097]

[0098] Among them, U cpnn To integrate the user potential factor matrix, I cpnn is the latent factor matrix of the fusion project, T cpnn is the fusion time latent factor matrix, It is a fusion operation.

[0099] Based on the above embodiment, the embodiment of the present invention uses the inverse operation of CP decomposition to restore and reconstruct the mixing factor matrix to obtain the final approximate score tensor.

[0100] Specifically, based on the inverse operation of the paradigm multivariate decomposition, the fused user latent factor matrix, the fused item latent factor matrix, and the fused time latent factor matrix are calculated to obtain an approximate rating matrix, including:

[0101]

[0102] in, is the approximate rating matrix, R cpnn for rank.

[0103] Last use Provide recommendations to users.

[0104] The tensor-based hybrid recommendation system provided by the present invention is described below. The tensor-based hybrid recommendation system described below and the tensor-based hybrid recommendation method described above can be referenced to each other.

[0105] Figure 2 is a schematic diagram of the structure of a tensor-based hybrid recommendation system provided by an embodiment of the present invention. Figure 2 As shown, it includes: an acquisition module 21, a conversion module 22, a first processing module 23, a second processing module 24, a fusion module 25, a restoration module 26 and a recommendation module 27, wherein:

[0106] The acquisition module 21 is used to collect user-side information sets, project-side information sets and time-side information sets; the conversion module 22 is used to convert the user-side information sets, the project-side information sets and the time-side information sets into user feature information, project feature information and time feature information respectively; the first processing module 23 is used to input the user feature information, the project feature information and the time feature information into the user tensor neural network, the project tensor neural network and the time tensor neural network respectively for feature extraction, and obtain the first user latent factor matrix, the first project latent factor matrix and the first time latent factor matrix; the second processing module 24 is used to perform a paradigm multivariate decomposition on the third-order scoring tensor formed by the user feature information, the project feature information and the time feature information, and obtain a second user latent factor matrix, a second project latent factor matrix, and a second time latent factor matrix; a fusion module 25 is used to fuse the first user latent factor matrix with the second user latent factor matrix to obtain a fused user latent factor matrix, fuse the first project latent factor matrix with the second project latent factor matrix to obtain a fused project latent factor matrix, and fuse the first time latent factor matrix with the second time latent factor matrix to obtain a fused time latent factor matrix; a restoration module 26 is used to calculate the fused user latent factor matrix, the fused project latent factor matrix, and the fused time latent factor matrix based on the inverse operation of the normal form multivariate decomposition to obtain an approximate rating matrix; a recommendation module 27 is used to output user recommendation results using the approximate rating matrix.

[0107] Figure 3 An example of a physical structure diagram of an electronic device is shown below. Figure 3As shown, the electronic device may include: a processor (processor) 310, a communication interface (Communications Interface) 320, a memory (memory) 330 and a communication bus 340, wherein the processor 310, the communication interface 320, and the memory 330 communicate with each other through the communication bus 340. The processor 310 can call the logic instructions in the memory 330 to execute the tensor-based hybrid recommendation method, which includes: collecting a user-side information set, a project-side information set and a time-side information set; respectively converting the user-side information set, the project-side information set and the time-side information set into user feature information, project feature information and time feature information; respectively inputting the user feature information, the project feature information and the time feature information into the user tensor neural network, the project tensor neural network and the time tensor neural network for feature extraction to obtain a first user latent factor matrix, a first project latent factor matrix and a first time latent factor matrix; and performing feature extraction on the three matrixes formed by the user feature information, the project feature information and the time feature information. The first-order rating tensor is subjected to a normal form multivariate decomposition to obtain a second user latent factor matrix, a second project latent factor matrix, and a second time latent factor matrix; the first user latent factor matrix is ​​fused with the second user latent factor matrix to obtain a fused user latent factor matrix, the first project latent factor matrix is ​​fused with the second project latent factor matrix to obtain a fused project latent factor matrix, and the first time latent factor matrix is ​​fused with the second time latent factor matrix to obtain a fused time latent factor matrix; the fused user latent factor matrix, the fused project latent factor matrix, and the fused time latent factor matrix are calculated based on the inverse operation of the normal form multivariate decomposition to obtain an approximate rating matrix; and the approximate rating matrix is ​​used to output the user recommendation result.

[0108] In addition, the logic instructions in the above-mentioned memory 330 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0109] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the tensor-based hybrid recommendation method provided by the above methods, which includes: collecting a user-side information set, a project-side information set, and a time-side information set; converting the user-side information set, the project-side information set, and the time-side information set into user feature information, project feature information, and time feature information respectively; inputting the user feature information, the project feature information, and the time feature information into a user tensor neural network, a project tensor neural network, and a time tensor neural network for feature extraction, and obtaining a first user latent factor matrix, a first project latent factor matrix, and a first time latent factor matrix. matrix; perform a normal form multivariate decomposition on the third-order rating tensor formed by the user feature information, the project feature information and the time feature information to obtain a second user latent factor matrix, a second project latent factor matrix and a second time latent factor matrix; fuse the first user latent factor matrix with the second user latent factor matrix to obtain a fused user latent factor matrix, fuse the first project latent factor matrix with the second project latent factor matrix to obtain a fused project latent factor matrix, and fuse the first time latent factor matrix with the second time latent factor matrix to obtain a fused time latent factor matrix; calculate the fused user latent factor matrix, the fused project latent factor matrix and the fused time latent factor matrix based on the inverse operation of the normal form multivariate decomposition to obtain an approximate rating matrix; use the approximate rating matrix to output the user recommendation result.

[0110] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented by a processor to execute the tensor-based hybrid recommendation method provided by the above-mentioned methods, the method comprising: collecting a user-side information set, an item-side information set and a time-side information set; converting the user-side information set, the item-side information set and the time-side information set into user feature information, item feature information and time feature information respectively; inputting the user feature information, the item feature information and the time feature information into a user tensor neural network, an item tensor neural network and a time tensor neural network respectively for feature extraction to obtain a first user latent factor matrix, a first item latent factor matrix and a first time latent factor matrix; extracting the user feature information, the item feature information and the time feature information from a ... The third-order rating tensor formed by the target feature information and the time feature information is subjected to a paradigm multivariate decomposition to obtain a second user latent factor matrix, a second project latent factor matrix and a second time latent factor matrix; the first user latent factor matrix is ​​fused with the second user latent factor matrix to obtain a fused user latent factor matrix, the first project latent factor matrix is ​​fused with the second project latent factor matrix to obtain a fused project latent factor matrix, and the first time latent factor matrix is ​​fused with the second time latent factor matrix to obtain a fused time latent factor matrix; the fused user latent factor matrix, the fused project latent factor matrix and the fused time latent factor matrix are calculated based on the inverse operation of the paradigm multivariate decomposition to obtain an approximate rating matrix; and the approximate rating matrix is ​​used to output the user recommendation result.

[0111] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0112] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0113] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A tensor-based hybrid recommendation method, characterized in that: include: Collect user-side information sets, project-side information sets, and time-side information sets; Converting the user-side information set, the item-side information set, and the time-side information set into user characteristic information, item characteristic information, and time characteristic information, respectively; Inputting the user feature information, the item feature information, and the time feature information into a user tensor neural network, an item tensor neural network, and a time tensor neural network, respectively, for feature extraction to obtain a first user latent factor matrix, a first item latent factor matrix, and a first time latent factor matrix; Performing a normalized multivariate decomposition on a third-order scoring tensor formed by the user feature information, the item feature information, and the time feature information to obtain a second user latent factor matrix, a second item latent factor matrix, and a second time latent factor matrix; Fusing the first user latent factor matrix with the second user latent factor matrix to obtain a fused user latent factor matrix, fusing the first item latent factor matrix with the second item latent factor matrix to obtain a fused item latent factor matrix, and fusing the first time latent factor matrix with the second time latent factor matrix to obtain a fused time latent factor matrix; Calculating the fused user latent factor matrix, the fused project latent factor matrix, and the fused time latent factor matrix based on an inverse operation of the normal form multivariate decomposition to obtain an approximate rating matrix; The approximate rating matrix is ​​used to output user recommendation results.

2. The tensor-based hybrid recommendation method according to claim 1, characterized in that The user-side information set includes user ID, gender, age and occupation, the project-side information set includes project ID, project type and project title, and the time-side information set includes time, user ID and project ID.

3. The tensor-based hybrid recommendation method according to claim 1, characterized in that The user feature information, the item feature information, and the time feature information are respectively input into a user tensor neural network, an item tensor neural network, and a time tensor neural network for feature extraction to obtain a first user latent factor matrix, a first item latent factor matrix, and a first time latent factor matrix, including: The user feature information Input the user tensor neural network to obtain the first user potential factor matrix U tnn : User tensor weight tensor in neural network Where U represents the number of users, Represents multiple user-side information in the user-side information set. The subscripts u1, u2, ..., and uM are the user-side information dimension identifiers respectively. n represents the number of tensor neural network layers. represents the activation functions of different user tensor neural network layers, and represents the offset tensor of the neural network layer at different user tensors, The superscript 1 represents the user type, O∈[1,n-1], where O represents the symbolic parameter that distinguishes different user tensor neural network layers; The project feature information Input the project tensor neural network to obtain the first project potential factor matrix I tnn : Project tensor neural network weight tensor Where I represents the number of items, Represents multiple project side information in the project side information set. The subscripts i1, i2, ..., iL are project side information dimension identifiers respectively. n represents the number of tensor neural network layers. represents the activation functions of the different project tensor neural network layers, and The offset tensor representing the different project tensor neural network layers, The superscript 2 represents the item type, Z∈[1,n-1], where Z represents the symbolic parameter that distinguishes different item tensor neural network layers; The time characteristic information Input the time tensor neural network to obtain the first time potential factor matrix T tnn : Weight Tensor of Time Tensor Neural Network Where T represents the number of time periods. Represents multiple time-side information in the time-side information set. The subscripts t1, t2, ..., tP are time-side information dimension identifiers respectively. n represents the number of tensor neural network layers. represents the activation functions of the different time tensor neural network layers, and represents the offset tensor of the neural network layer at different time tensors, The superscript 3 represents the time type, Q∈[1,n-1], where Q represents the symbolic parameter that distinguishes different time tensor neural network layers.

4. The tensor-based hybrid recommendation method according to claim 3, characterized in that: Performing a normalized multivariate decomposition on the third-order rating tensor formed by the user feature information, the item feature information, and the time feature information to obtain a second user latent factor matrix, a second item latent factor matrix, and a second time latent factor matrix, including: For the third-order score tensor R o ∈R U×I×T Performing paradigm multivariate decomposition, we get: in, To reconstruct the approximate rating tensor, R cp is the rank of the normal form multivariate decomposition, represents the outer product operation, λ is the weight vector, λ r represents the rth element of the weight vector λ, U cp is the second user latent factor matrix, I cp is the latent factor matrix of the second item, T cp is the second time latent factor matrix, u r is the user’s latent factor vector, i r is the project latent factor vector, t r is the temporal latent factor vector.

5. The tensor-based hybrid recommendation method according to claim 4, characterized in that: The method comprises: fusing the first user latent factor matrix with the second user latent factor matrix to obtain a fused user latent factor matrix; fusing the first item latent factor matrix with the second item latent factor matrix to obtain a fused item latent factor matrix; and fusing the first time latent factor matrix with the second time latent factor matrix to obtain a fused time latent factor matrix. Among them, U cpnn To fuse the user latent factor matrix, I cpnn is the latent factor matrix of the fusion project, T cpnn is the fusion time latent factor matrix, It is a fusion operation.

6. The tensor-based hybrid recommendation method according to claim 5, characterized in that: The fused user latent factor matrix, the fused item latent factor matrix, and the fused time latent factor matrix are calculated based on the inverse operation of the normal form multivariate decomposition to obtain an approximate rating matrix, including: in, is the approximate rating matrix, R cpnn for rank.

7. A tensor-based hybrid recommendation system, characterized in that include: A collection module, used to collect user-side information sets, project-side information sets, and time-side information sets; A conversion module, configured to convert the user-side information set, the item-side information set, and the time-side information set into user characteristic information, item characteristic information, and time characteristic information, respectively; a first processing module, configured to input the user feature information, the item feature information, and the time feature information into a user tensor neural network, an item tensor neural network, and a time tensor neural network, respectively, for feature extraction to obtain a first user latent factor matrix, a first item latent factor matrix, and a first time latent factor matrix; a second processing module, configured to perform a normalized multivariate decomposition on a third-order rating tensor formed by the user feature information, the item feature information, and the time feature information to obtain a second user latent factor matrix, a second item latent factor matrix, and a second time latent factor matrix; a fusion module, configured to fuse the first user latent factor matrix with the second user latent factor matrix to obtain a fused user latent factor matrix, fuse the first project latent factor matrix with the second project latent factor matrix to obtain a fused project latent factor matrix, and fuse the first time latent factor matrix with the second time latent factor matrix to obtain a fused time latent factor matrix; a restoration module, configured to calculate the fused user latent factor matrix, the fused project latent factor matrix, and the fused time latent factor matrix based on an inverse operation of a normal form multivariate decomposition to obtain an approximate rating matrix; The recommendation module is used to output user recommendation results using the approximate rating matrix.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the tensor-based hybrid recommendation method according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the tensor-based hybrid recommendation method according to any one of claims 1 to 6 is implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the tensor-based hybrid recommendation method according to any one of claims 1 to 6 is implemented.

Images