CN113051463B - Project pushing method and system - Google Patents
Project pushing method and system Download PDFInfo
- Publication number
- CN113051463B CN113051463B CN201911365828.6A CN201911365828A CN113051463B CN 113051463 B CN113051463 B CN 113051463B CN 201911365828 A CN201911365828 A CN 201911365828A CN 113051463 B CN113051463 B CN 113051463B
- Authority
- CN
- China
- Prior art keywords
- matrix
- dimension
- item
- tensor
- user
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a project pushing method and a project pushing system, which belong to the technical field of data service, wherein the project pushing method comprises the following steps: establishing tensors according to user data of a plurality of dimensions related to the item to be recommended; constructing a plurality of matrixes according to the relation among the dimensions of the tensor; combining the tensor and the matrixes to synchronously decompose so as to jointly form a cost function; and pushing the target item in the items to be recommended to the terminal corresponding to the target user according to the cost function and the data of the target user. The invention provides a multi-source-oriented high-order tensor and matrix joint synchronous factor decomposition mechanism, which can effectively relieve the problems of cold start and data sparseness of recommendation, thereby improving the accuracy of recommendation.
Description
Technical Field
The present invention relates to the field of data service technologies, and in particular, to a method and a system for pushing a project.
Background
Context-aware based recommendations often involve high-dimensional data (especially above 3 dimensions), face serious high-dimensional data sparsity problems, and generating relatively accurate recommendations with few non-zero user preferences is a difficulty of research.
Disclosure of Invention
In view of the above, the present invention provides a method and a system for pushing items, which are used for solving the problem of low recommendation accuracy in the current high-dimensional data recommendation technology.
In order to solve the above technical problems, in a first aspect, the present invention provides an item pushing method, including:
establishing tensors according to user data of a plurality of dimensions related to the item to be recommended;
constructing a plurality of matrixes according to the relation among the dimensions of the tensor;
combining the tensor and the matrixes to synchronously decompose so as to jointly form a cost function;
and pushing the target item in the items to be recommended to the terminal corresponding to the target user according to the cost function and the data of the target user.
Optionally, the cost function includes a weighted least squares error term of the tensor decomposition and a weighted least squares error term of the plurality of matrix factorizations.
Optionally, the cost function further includes a regularization term, where the regularization term includes at least one of a first regularization term and a second regularization term, where the first regularization term is a regularization term for smoothing user behavior in a continuous first dimension, the first dimension is one of the plurality of dimensions, and the second regularization term is a regularization term for preventing overfitting.
Optionally, in a case where the plurality of dimensions includes a user dimension, a user scenario dimension, and an item dimension, and the user scenario dimension includes a first dimension, a second dimension, and a third dimension, the plurality of matrices includes at least one of the following matrices:
matrix D εR t×e ；
Matrix Y εR l×e ；
Matrix Z εR l×l ；
Matrix H E R c×e ；
Matrix S epsilon R e×e ；
Where t is the first dimension, e is the item dimension, l is the second dimension, and c is the third dimension.
Optionally, the cost function is:
wherein X is the tensor, X ε R u×t×l×c×e U is the user dimension;
W∈R u×t×l×c×e for indicating missing items in the tensor X,
a weighted least squares error term decomposed for the tensor X;
matrix U epsilon R u×k Matrix T epsilon R t×k Matrix L epsilon R l×k Matrix C ε R c×k Matrix E E R e×k Five matrices decomposed for the tensor X, k being a factor number;
respectively a matrix Y, a matrix Z, a matrix S, a matrix H and a matrix D factorized weighted least square error term;
λ 1 -λ 6 is a parameter;
"|| I' is the matrix norm is used to determine the matrix norm," omicron "is the outer product.
Optionally, the plurality of dimensions further includes an item context dimension, and the factored term associated with the item in the cost function includes an item context utility value.
Optionally, in a case where the plurality of dimensions further includes an item scenario dimension, the cost function is:
wherein X is the tensor, X ε R u×t×l×c×e U is the user dimension;
W∈R u×t×l×c×e for indicating missing items in the tensor X,
a weighted least squares error term decomposed for the tensor X;
B m ∈R u×t×l×c×e the method comprises the steps that the influence value of different states of an item class M scene on the tensor X is given, and the item is provided with the class M scene;
matrix U epsilon R u×k Matrix T epsilon R t×k Matrix L epsilon R l×k Matrix C ε R c×k Matrix E E R e×k Five matrices decomposed for the tensor X, k being a factor number;
respectively a matrix Y, a matrix Z, a matrix S, a matrix H and a matrix D factorized weighted least square error term;
λ 1 -λ 6 is a parameter;
"|| I' is the matrix norm is used to determine the matrix norm," omicron "is the outer product.
In a second aspect, the present invention further provides an item pushing system, including:
the tensor establishing module is used for establishing tensors according to user data of a plurality of dimensions related to the item to be recommended;
the matrix construction module is used for constructing a plurality of matrixes according to the relation among the dimensions of the tensor;
the decomposition module is used for synchronously decomposing the tensor and the matrixes to jointly form a cost function;
and the pushing module is used for pushing the target item in the items to be recommended to the terminal corresponding to the target user according to the cost function and the data of the target user.
Optionally, the cost function includes a weighted least squares error term of the tensor decomposition and a weighted least squares error term of the plurality of matrix factorizations.
Optionally, the cost function further includes a regularization term, where the regularization term includes at least one of a first regularization term and a second regularization term, where the first regularization term is a regularization term for smoothing user behavior in a continuous first dimension, the first dimension is one of the plurality of dimensions, and the second regularization term is a regularization term for preventing overfitting.
Optionally, in a case where the plurality of dimensions includes a user dimension, a user scenario dimension, and an item dimension, and the user scenario dimension includes a first dimension, a second dimension, and a third dimension, the plurality of matrices includes at least one of the following matrices:
matrix D εR t×e ；
Matrix Y εR l×e ；
Matrix Z εR l×l ；
Matrix H E R c×e ；
Matrix S epsilon R e×e ；
Where t is the first dimension, e is the item dimension, l is the second dimension, and c is the third dimension.
Optionally, the cost function is:
wherein X is the tensor, X ε R u×t×l×c×e U is the user dimension;
W∈R u×t×l×c×e for indicating missing items in the tensor X,
a weighted least squares error term decomposed for the tensor X;
matrix U epsilon R u×k Matrix T epsilon R t×k Matrix L epsilon R l×k Matrix C ε R c×k Matrix E E R e×k Five matrices decomposed for the tensor X, k being a factor number;
respectively a matrix Y, a matrix Z, a matrix S, a matrix H and a matrix D factorized weighted least square error term;
λ 1 -λ 6 is a parameter;
"|| I' is the matrix norm is used to determine the matrix norm," omicron "is the outer product.
Optionally, the plurality of dimensions further includes an item context dimension, and the factored term associated with the item in the cost function includes an item context utility value.
Optionally, in a case where the plurality of dimensions further includes an item scenario dimension, the cost function is:
wherein X is the tensor, X ε R u×t×l×c×e U is the user dimension;
W∈R u×t×l×c×e for indicating missing items in the tensor X,
a weighted least squares error term decomposed for the tensor X;
B m ∈R u×t×l×c×e the method comprises the steps that the influence value of different states of an item class M scene on the tensor X is given, and the item is provided with the class M scene;
matrix U epsilon R u×k Matrix T epsilon R t×k Matrix L epsilon R l×k Matrix C ε R c×k Matrix E E R e×k Five matrices decomposed for the tensor X, k being a factor number;
respectively a matrix Y, a matrix Z, a matrix S, a matrix H and a matrix D factorized weighted least square error term;
λ 1 -λ 6 is a parameter;
"|| I' is the matrix norm is used to determine the matrix norm," omicron "is the outer product.
In a third aspect, the present invention also provides an item push system, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor; the processor, when executing the computer program, implements the steps of any of the item pushing methods described above.
In a fourth aspect, the present invention also provides a computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of any of the item pushing methods described above.
The technical scheme of the invention has the following beneficial effects:
the embodiment of the invention provides a multi-source (multi-dimension) oriented high-order tensor and matrix joint synchronization factor decomposition mechanism, which can effectively relieve the problems of recommended cold start and data sparseness, thereby improving the accuracy of recommendation and further improving the satisfaction of users.
Drawings
Fig. 1 is a flow chart of a method for pushing items according to a first embodiment of the invention;
fig. 2 is a schematic structural diagram of an item pushing system in a second embodiment of the present invention;
fig. 3 is a schematic structural diagram of an item pushing system according to a third embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. It will be apparent that the described embodiments are some, but not all, embodiments of the invention. All other embodiments, which are obtained by a person skilled in the art based on the described embodiments of the invention, fall within the scope of protection of the invention.
In recent years, the internet of things gradually goes into our lives, which is another information industry revolution following computers, the internet and mobile communication networks. The internet of things is an expansion of the internet to physical space, has all the characteristics of the internet, and causes quality change by a variable quantity. The internet of things is not only a network with all things connected, but also a link for mutual communication and connection between the physical world and the virtual world, so that objects in the physical world can be automatically accepted by the virtual world, and information in the virtual world can be automatically understood by the physical world, thereby realizing seamless integration of the physical world and the virtual world. The Internet of things brings people into the 'big data' era, the latest report shows that the number of Internet of things devices worldwide in 2015 can reach 49 hundred million, 30% more in 2014, and the number of Internet of things devices can be expected to reach 250 hundred million in 2020. The Internet of things rapidly develops and simultaneously produces massive multi-mode data, in addition, research reports predict that the data will expand by 10 times as much as the current data in 2020, and climb from 4.4 megaGB to 44 megaGB. In summary, the problem of information overload and knowledge shortage in the environment of the internet of things is becoming more serious.
The internet of things changes the daily life and communication modes of people, and aims at physical entities and massive polymorphic data in the billions, so that the internet of things information can be fully utilized, the internet of things recommendation technology is urgently required to be developed, the requirement of people for acquiring personalized internet of things information in real time is met, and the problems of information overload and knowledge shortage of the internet of things are effectively solved. Although some existing information retrieval technologies can be transplanted to certain application scenes of the internet of things through improvement or expansion, the requirements of new mechanisms and new methods recommended by the internet of things are urgent.
Physical entity recommendations provide physical entity filtering and recommendation services by modeling user interests or predicting user behavior. Scenes are any information describing the status of an entity, wherein an entity may be a person, an object, a place, or a related object interacting with a user and an application (including the user and the application itself), and the scenes are classified into two types of static scenes and dynamic scenes according to the change situation. The recommendation system based on context awareness introduces context information into the recommendation system, has the advantages of universality and individuation, and can further improve recommendation accuracy and user satisfaction.
Recommendation techniques based on context awareness are divided into: content-based recommendations, collaborative filtering-based recommendations, and hybrid recommendations. Among them, in collaborative filtering algorithms, matrix decomposition has gradually replaced traditional neighbor-based collaborative filtering algorithms to become research hotspots and mainstream models. This is mainly due to the higher prediction accuracy of matrix decomposition and the ability to alleviate the data sparseness problem. Matrix factorization reconstructs a scoring matrix by mapping users (items) into a low-dimensional space and estimating user (item) potential feature vectors by fitting preference scores. With the development of higher-order statistics, a higher-order tensor model is used for analyzing multidimensional data, and a higher-order tensor decomposition technology becomes a new hot spot in the application of a recommendation system.
The recommendation of physical entities is mainly applied in the field of Location Based Social Networks (LBSN), and the targets recommended to users mainly include: location (e.g., venue, area, etc.), user (e.g., expert, friend, etc.), and activity.
Recommendation techniques based on context awareness generally involve high-dimensional data, which, as described above, has serious data sparsity problems. Although tensor decomposition technology can alleviate sparsity problem to a certain extent, how to introduce a multisource data fusion mechanism into the recommendation generation process on the premise of not affecting recommendation instantaneity so as to overcome recommendation cold start and high-dimensional data sparsity and further solve recommendation accuracy problem is still a problem to be solved. In order to solve the technical problems, the present invention provides a method for pushing items, which is described in detail below.
Referring to fig. 1, fig. 1 is a flowchart of an item pushing method according to an embodiment of the invention, including the following steps:
step 11: establishing tensors according to user data of a plurality of dimensions related to the item to be recommended;
step 12: constructing a plurality of matrixes according to the relation among the dimensions of the tensor;
step 13: combining the tensor and the matrixes to synchronously decompose so as to jointly form a cost function;
step 14: and pushing the target item in the items to be recommended to the terminal corresponding to the target user according to the cost function and the data of the target user. And recommending target items to the target user according to the cost function and the data of the target user.
Wherein the tensor may also be referred to as a utility tensor, the matrix may also be referred to as a utility matrix, and the cost function is a supervised learning cost function. The item to be recommended may be a tourist attraction and the target item may be a specific tourist attraction, such as a mountain. The item to be recommended can also be food, and the target item can be a specific store.
The embodiment of the invention provides a multi-source (multi-dimension) oriented high-order tensor and matrix joint synchronization factor decomposition mechanism, which can effectively relieve the problems of recommended cold start and data sparseness, thereby improving the accuracy of recommendation and further improving the satisfaction of users.
The above item recommendation method is exemplified below.
Alternatively, the user data of multiple dimensions related to the item to be recommended may be extracted from information of multiple different sources. The heterogeneous source oriented user preference extraction process may be: based on the user characteristic model, a quantitative analysis method is adopted to calculate the preference of the user for a plurality of items of the item to be recommended under a plurality of different dimensions. For example, the preference of a user for a physical entity at a certain time, a certain location, and a certain emotion can be implicitly calculated using quantitative analysis methods, wherein time, location, emotion are three different dimensions of a user scenario, which of course may also include other dimensions, such as physical conditions, surrounding environment, etc., whereAnd will not be described in detail. After extracting preferences of a plurality of items to be recommended of a user in a plurality of different dimensions, a high-order preference utility tensor, namely, the tensor is obtained. For example, if the user's preferences for a physical entity at a certain time, location, and emotion are extracted, then a "user-time-location-emotion-physical entity" higher order preference utility tensor, denoted as XεR, may be obtained u×t×l×c×e Where u is the user dimension, t is the time dimension, l is the location dimension, c is the emotion dimension, and e is the physical entity dimension.
In addition, to alleviate the problems of cold start and tensor data sparseness, a number of utility matrices may be constructed according to the relationships between the dimensions of the tensor to supplement the utility tensor. For example, if the tensor is a "user-time-location-emotion-physical entity" higher order preference utility tensor, then the following several utility matrices may be constructed: "location-physical entity" matrix Y εR l×e "position-position" similarity matrix Z ε R l×l "physical entity-physical entity" relevance matrix S ε R e×e "emotion-physical entity" relevance matrix H ε R c×e And a "time-physical entity" correlation matrix D εR t×e 。
After the matrixes are constructed, a method of combining high-order tensors and matrixes and synchronizing factor decomposition can be adopted, and the tensors and the matrixes are combined and synchronously decomposed to jointly form a cost function of supervised learning. Optionally, the cost function includes a weighted least squares error term of the tensor decomposition and a weighted least squares error term of the plurality of matrix factorizations.
Further optionally, the cost function further includes a regularization term including at least one of a first regularization term for smoothing user behavior in a first dimension of the succession, the first dimension being one of the plurality of dimensions, and a second regularization term for preventing overfitting. For example, the first dimension may be a time dimension.
Optionally, in a case where the plurality of dimensions includes a user dimension, a user scenario dimension, and an item dimension, and the user scenario dimension includes a first dimension, a second dimension, and a third dimension, the plurality of matrices includes at least one of the following matrices:
matrix D εR t×e ；
Matrix Y εR l×e ；
Matrix Z εR l×l ；
Matrix H E R c×e ；
Matrix S epsilon R e×e ；
Where t is the first dimension, e is the item dimension, l is the second dimension, and c is the third dimension.
For example, the first dimension may be a time dimension, the second dimension may be a location dimension, and the third dimension may be an emotion dimension.
Optionally, the cost function is:
wherein X is the tensor, X ε R u×t×l×c×e U is the user dimension;
W∈R u×t×l×c×e for indicating missing items in the tensor X,
a weighted least squares error term decomposed for the tensor X;
matrix U epsilon R u×k Matrix T epsilon R t×k Matrix L epsilon R l×k Matrix C ε R c×k Matrix E E R e×k Five matrices decomposed for the tensor X, k being a factor number;
respectively a matrix Y, a matrix Z, a matrix S, a matrix H and a matrix D factorized weighted least square error term;
λ 1 -λ 6 is a parameter;
"|| I' is the matrix norm is used to determine the matrix norm," omicron "is the outer product.
In addition, in the case of the optical fiber,
if x u,t,l,c,e is missing means: if x u,t,l,c,e Deletion.
In the embodiment of the invention, a multidimensional collaborative filtering method is adopted to generate the recommendation. The first dimension, the second dimension, and the third dimension are merged as three dimensions of the user scenario in one utility tensor and a plurality of utility matrices. In other embodiments, the user context dimension may be extended according to the actual application requirements. Based on tensor decomposition technology, the user scenario dimension is integrated into the recommendation process, so that the association relationship among the user, the scenario and the project can be effectively mined.
The above process of constructing a cost function, i.e. the modeling process, is also referred to as the cost function, i.e. the model being constructed.
Optionally, the plurality of dimensions further includes an item context dimension, and the factored term associated with the item in the cost function includes an item context utility value.
Although the user scene is integrated into the recommendation process, the association relationship among the user, the scene and the project can be effectively mined. However, some recommendation generation of items requires attention to real-time scenarios of both the user and the item, e.g. in case the item is a physical entity, the scenario of the physical entity cannot be ignored, especially its dynamic scenario. Therefore, the embodiment of the invention also provides a multi-dimensional collaborative recommendation technology based on double-sided context awareness, namely the multiple dimensions not only comprise the user context dimension but also comprise the project context dimension, or the embodiment of the invention fuses the real-time context of the user and the project into the recommendation generation process.
Alternatively, after the cost function is obtained, a random gradient descent algorithm or other gradient descent algorithm may be used to train the model (i.e., the cost function) such that the cost function falls within an acceptable threshold, e.g., less than a preset threshold. And substituting the scenes related to the target user into the trained model, so as to generate the target item for pushing to the user.
Optionally, in a case where the plurality of dimensions further includes an item scenario dimension, the cost function is:
wherein X is the tensor, X ε R u×t×l×c×e U is the user dimension;
W∈R u×t×l×c×e for indicating missing items in the tensor X,
a weighted least squares error term decomposed for the tensor X;
B m ∈R u×t×l×c×e the method comprises the steps that the influence value of different states of an item class M scene on the tensor X is given, and the item is provided with the class M scene;
matrix U epsilon R u×k Matrix T epsilon R t×k Matrix L epsilon R l×k Matrix C ε R c×k Matrix E E R e×k Five matrices decomposed for the tensor X, k being a factor number;
respectively a matrix Y, a matrix Z, a matrix S, a matrix H and a matrix D factorized weighted least square error term;
λ 1 -λ 6 is a parameter;
"|| I' is the matrix norm is used to determine the matrix norm," omicron "is the outer product.
In the cost function, scenes of the project, such as real-time attributes of physical entities, are merged.
Wherein the state set of the m-th scene is {0,1,2, … z m A state of 0 indicates that the scene is unknown. For example, if the class m scenario is a location, then the different states of the class m scenario refer to different locations.
In the embodiment of the invention, the position relation between the user and the project can be integrated into the recommendation process as in the recommendation scheme facing to the physical entity, thereby realizing static project recommendation. Other context information of both the user and the item except the location, in particular dynamic context information of both, such as emotion, surrounding environment, etc., can also be combined, so that personalized recommendations are generated for the user in real time, for example, target items are recommended to the user, where the scene may meet the interest demands of the user.
Referring to fig. 2, fig. 2 is a schematic structural diagram of an item pushing system according to a second embodiment of the present invention, where the item pushing system 20 includes:
a tensor establishing module 21, configured to establish tensors according to user data of multiple dimensions related to the item to be recommended;
a matrix construction module 22, configured to construct a plurality of matrices according to the relationships among the dimensions of the tensor;
a decomposition module 23, configured to combine the tensor and the matrices to perform decomposition synchronously, so as to jointly form a cost function;
and the pushing module 24 is configured to push the target item in the items to be recommended to the terminal corresponding to the target user according to the cost function and the data of the target user.
The embodiment of the invention provides a multi-source-oriented high-order tensor and matrix joint synchronization factor decomposition mechanism, which can effectively relieve the problems of cold start and data sparseness of recommendation, thereby improving the accuracy of recommendation and further improving the satisfaction degree of users.
Optionally, the cost function includes a weighted least squares error term of the tensor decomposition and a weighted least squares error term of the plurality of matrix factorizations.
Optionally, the cost function further includes a regularization term, where the regularization term includes at least one of a first regularization term and a second regularization term, where the first regularization term is a regularization term for smoothing user behavior in a continuous first dimension, the first dimension is one of the plurality of dimensions, and the second regularization term is a regularization term for preventing overfitting.
Optionally, in a case where the plurality of dimensions includes a user dimension, a user scenario dimension, and an item dimension, and the user scenario dimension includes a first dimension, a second dimension, and a third dimension, the plurality of matrices includes at least one of the following matrices:
matrix D εR t×e ；
Matrix Y εR l×e ；
Matrix Z εR l×l ；
Matrix H E R c×e ；
Matrix S epsilon R e×e ；
Where t is the first dimension, e is the item dimension, l is the second dimension, and c is the third dimension.
Optionally, the cost function is:
wherein X is the tensor, X ε R u×t×l×c×e U is the user dimension;
W∈R u×t×l×c×e for indicating missing items in the tensor X,
a weighted least squares error term decomposed for the tensor X;
matrix U epsilon R u×k Matrix T epsilon R t×k Matrix L epsilon R l×k Matrix C ε R c×k Matrix E E R e×k Five matrices decomposed for the tensor X, k being a factor number;
respectively a matrix Y, a matrix Z, a matrix S, a matrix H and a matrix D factorized weighted least square error term;
λ 1 -λ 6 is a parameter;
"|| I' is the matrix norm is used to determine the matrix norm," omicron "is the outer product.
Optionally, the plurality of dimensions further includes an item context dimension, and the factored term associated with the item in the cost function includes an item context utility value.
Optionally, in a case where the plurality of dimensions further includes an item scenario dimension, the cost function is:
wherein X is the tensor, X ε R u×t×l×c×e U is the user dimension;
W∈R u×t×l×c×e for indicating missing items in the tensor X,
a weighted least squares error term decomposed for the tensor X;
B m ∈R u×t×l×c×e the method comprises the steps that the influence value of different states of an item class M scene on the tensor X is given, and the item is provided with the class M scene;
matrix U epsilon R u×k Matrix T epsilon R t×k Matrix L epsilon R l×k Matrix C ε R c×k Matrix E E R e×k Five matrices decomposed for the tensor X, k being a factor number;
respectively a matrix Y, a matrix Z, a matrix S, a matrix H and a matrix D factorized weighted least square error term;
λ 1 -λ 6 is a parameter;
"|| I' is the matrix norm is used to determine the matrix norm," omicron "is the outer product.
The embodiment of the present invention is a product embodiment corresponding to the first embodiment of the above method, so that the detailed description thereof will be omitted herein.
Referring to fig. 3, fig. 3 is a schematic structural diagram of an item pushing system according to a third embodiment of the present invention, where the item pushing system 30 includes a processor 31, a memory 32, and a computer program stored in the memory 32 and capable of running on the processor 31; the processor 31, when executing the computer program, implements the following steps:
establishing tensors according to user data of a plurality of dimensions related to the item to be recommended;
constructing a plurality of matrixes according to the relation among the dimensions of the tensor;
combining the tensor and the matrixes to synchronously decompose so as to jointly form a cost function;
and pushing the target item in the items to be recommended to the terminal corresponding to the target user according to the cost function and the data of the target user.
The embodiment of the invention provides a multi-source-oriented high-order tensor and matrix joint synchronization factor decomposition mechanism, which can effectively relieve the problems of cold start and data sparseness of recommendation, thereby improving the accuracy of recommendation and further improving the satisfaction degree of users.
Optionally, the cost function includes a weighted least squares error term of the tensor decomposition and a weighted least squares error term of the plurality of matrix factorizations.
Optionally, the cost function further includes a regularization term, where the regularization term includes at least one of a first regularization term and a second regularization term, where the first regularization term is a regularization term for smoothing user behavior in a continuous first dimension, the first dimension is one of the plurality of dimensions, and the second regularization term is a regularization term for preventing overfitting.
Optionally, in a case where the plurality of dimensions includes a user dimension, a user scenario dimension, and an item dimension, and the user scenario dimension includes a first dimension, a second dimension, and a third dimension, the plurality of matrices includes at least one of the following matrices:
matrix D εR t×e ；
Matrix Y εR l×e ；
Matrix Z εR l×l ；
Matrix H E R c×e ；
Matrix S epsilon R e×e ；
Where t is the first dimension, e is the item dimension, l is the second dimension, and c is the third dimension.
Optionally, the cost function is:
wherein X is the tensor, X ε R u×t×l×c×e U is the user dimension;
W∈R u×t×l×c×e for indicating missing items in the tensor X,
a weighted least squares error term decomposed for the tensor X;
matrix U epsilon R u×k Matrix T epsilon R t×k Matrix L epsilon R l×k Matrix C ε R c×k Matrix E E R e×k Five matrices decomposed for the tensor X, k being a factor number;
respectively a matrix Y, a matrix Z, a matrix S, a matrix H and a matrix D factorized weighted least square error term;
λ 1 -λ 6 is a parameter;
"|| I' is the matrix norm is used to determine the matrix norm," omicron "is the outer product.
Optionally, the plurality of dimensions further includes an item context dimension, and the factored term associated with the item in the cost function includes an item context utility value.
Optionally, in a case where the plurality of dimensions further includes an item scenario dimension, the cost function is:
wherein X is the tensor, X ε R u×t×l×c×e U is the user dimension;
W∈R u×t×l×c×e for indicating missing items in the tensor X,
a weighted least squares error term decomposed for the tensor X;
B m ∈R u×t×l×c×e the method comprises the steps that the influence value of different states of an item class M scene on the tensor X is given, and the item is provided with the class M scene;
matrix U epsilon R u×k Matrix T epsilon R t×k Matrix L epsilon R l×k Matrix C ε R c×k Matrix E E R e×k Five matrices decomposed for the tensor X, k being a factor number;
respectively a matrix Y, a matrix Z, a matrix S, a matrix H and a matrix D factorized weighted least square error term;
λ 1 -λ 6 is a parameter;
"|| I' is the matrix norm is used to determine the matrix norm," omicron "is the outer product.
The specific working process of the embodiment of the present invention is the same as that of the first embodiment of the method, so that the detailed description thereof will be omitted herein.
A fourth embodiment of the present invention provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of any one of the item pushing methods of the first embodiment. For details, reference is made to the description of the method steps in the corresponding embodiments above.
The terminal in the embodiment of the invention can be a wireless terminal or a wired terminal, and the wireless terminal can be a device for providing voice and/or other service data connectivity for a user, a handheld device with a wireless connection function or other processing devices connected to a wireless modem. A wireless terminal may communicate with one or more core networks via a radio access network (Radio Access Network, RAN for short), which may be mobile terminals such as mobile phones (or "cellular" phones) and computers with mobile terminals, e.g., portable, pocket, hand-held, computer-built-in or vehicle-mounted mobile devices that exchange voice and/or data with the radio access network. Such as personal communication services (Personal Communication Service, PCS) phones, cordless phones, session initiation protocol (Session Initiation Protocol, SIP) phones, wireless local loop (Wireless Local Loop, WLL) stations, personal digital assistants (Personal Digital Assistant, PDA) and the like. A wireless Terminal may also be referred to as a system, subscriber Unit (Subscriber Unit), subscriber Station (Subscriber Station), mobile Station (Mobile Station), remote Station (Remote Station), remote Terminal (Remote Terminal), access Terminal (Access Terminal), user Terminal (User Terminal), user Agent (User Agent), terminal (User Device or User Equipment), without limitation.
Such computer-readable storage media, including both non-transitory and non-transitory, removable and non-removable media, may be implemented in any method or technology for information storage. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of storage media for a computer include, but are not limited to, phase change memory (PRAM), static Random Access Memory (SRAM), dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium, which can be used to store information that can be accessed by a computing device.
While the foregoing is directed to the preferred embodiments of the present invention, it will be appreciated by those skilled in the art that various modifications and adaptations can be made without departing from the principles of the present invention, and such modifications and adaptations are intended to be comprehended within the scope of the present invention.
Claims (5)
1. An item pushing method is characterized by comprising the following steps:
establishing tensors according to user data of a plurality of dimensions related to the item to be recommended, wherein the tensors are utility tensors, and acquiring high-order preference utility tensors after extracting preferences of a plurality of items of the item to be recommended of a user under a plurality of different dimensions;
constructing a plurality of matrixes according to the relation among the dimensions of the tensor, wherein the matrixes are utility matrixes;
after a plurality of utility matrixes are constructed, a method of combining high-order tensors and matrixes and synchronizing factor decomposition is adopted, and the tensors and the matrixes are combined and synchronously decomposed to jointly form a cost function of supervised learning; the cost function includes a weighted least squares error term for the tensor decomposition and a weighted least squares error term for the plurality of matrix factorization;
pushing target items in the items to be recommended to terminals corresponding to the target users according to the cost function and the data of the target users;
in the case where the plurality of dimensions includes a user dimension, a user context dimension, and an item dimension, and the user context dimension includes a first dimension, a second dimension, and a third dimension, the plurality of matrices includes at least one of the following matrices:
matrix D εR t×e ；
Matrix Y εR l×e ；
Matrix Z εR l×l ；
Matrix H E R c×e ；
Matrix S epsilon R e×e ；
Wherein t is a first dimension, e is a project dimension, l is a second dimension, and c is a third dimension;
the plurality of dimensions further comprise a project scenario dimension, and the factored term related to the project in the cost function comprises a project scenario utility value;
in the case where the plurality of dimensions further includes an item context dimension, the cost function is:
wherein X is the tensor, X ε R u×t×l×c×e U is the user dimension;
W∈R u×t×l×c×e for indicating missing items in the tensor X,
a weighted least squares error term decomposed for the tensor X;
B m ∈R u×t×l×c×e the method comprises the steps that the influence value of different states of an item class M scene on the tensor X is given, and the item is provided with the class M scene;
matrix U epsilon R u×k Matrix T epsilon R t×k Matrix L epsilon R l×k Matrix C ε R c×k Matrix E E R e×k Five matrices decomposed for the tensor X, k being a factor number;
respectively a matrix Y, a matrix Z, a matrix S, a matrix H and a matrix D factorized weighted least square error term;
λ 1 -λ 6 is a parameter;
2. The method of claim 1, wherein the cost function further comprises a regularization term comprising at least one of a first regularization term for smoothing user behavior in a first continuous dimension, the first dimension being one of the plurality of dimensions, and a second regularization term for preventing overfitting.
3. An item pushing system, comprising:
the tensor establishing module is used for establishing tensors according to user data of a plurality of dimensions related to the item to be recommended, wherein the tensors are utility tensors, and after extracting preferences of a plurality of items of the item to be recommended of a user in a plurality of different dimensions, the tensors acquire high-order preference utility tensors;
the matrix construction module is used for constructing a plurality of matrixes according to the relation among the dimensions of the tensor, wherein the matrixes are utility matrixes;
the decomposition module is used for adopting a method of combining high-order tensors and matrixes to synchronously decompose the tensors and the matrixes after constructing a plurality of utility matrixes, and forming a cost function of supervised learning together; the cost function includes a weighted least squares error term for the tensor decomposition and a weighted least squares error term for the plurality of matrix factorization;
the pushing module is used for pushing the target item in the items to be recommended to the terminal corresponding to the target user according to the cost function and the data of the target user;
in the case where the plurality of dimensions includes a user dimension, a user context dimension, and an item dimension, and the user context dimension includes a first dimension, a second dimension, and a third dimension, the plurality of matrices includes at least one of the following matrices:
matrix D εR t×e ；
Matrix Y εR l×e ；
Matrix Z εR l×l ；
Matrix H E R c×e ；
Matrix S epsilon R e×e ；
Wherein t is a first dimension, e is a project dimension, l is a second dimension, and c is a third dimension;
the plurality of dimensions further comprise a project scenario dimension, and the factored term related to the project in the cost function comprises a project scenario utility value;
in the case where the plurality of dimensions further includes an item context dimension, the cost function is:
wherein X is the tensor, X ε R u×t×l×c×e U is the user dimension;
W∈R u×t×l×c×e for indicating missing items in the tensor X,
a weighted least squares error term decomposed for the tensor X;
B m ∈R u×t×l×c×e the method comprises the steps that the influence value of different states of an item class M scene on the tensor X is given, and the item is provided with the class M scene;
matrix U epsilon R u×k Matrix T epsilon R t×k Matrix L epsilon R l×k Matrix C ε R c×k Matrix E E R e×k Five matrices decomposed for the tensor X, k being a factor number;
respectively a matrix Y, a matrix Z, a matrix S, a matrix H and a matrix D factorized weighted least square error term;
λ 1 -λ 6 is a parameter;
4. An item push system comprising a memory, a processor, and a computer program stored on the memory and executable on the processor; -characterized in that the processor, when executing the computer program, implements the steps of the item pushing method according to any of claims 1 to 2.
5. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the item pushing method according to any of claims 1 to 2.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911365828.6A CN113051463B (en) | 2019-12-26 | 2019-12-26 | Project pushing method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911365828.6A CN113051463B (en) | 2019-12-26 | 2019-12-26 | Project pushing method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113051463A CN113051463A (en) | 2021-06-29 |
| CN113051463B true CN113051463B (en) | 2023-07-07 |
Family
ID=76505229
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911365828.6A Active CN113051463B (en) | 2019-12-26 | 2019-12-26 | Project pushing method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113051463B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104933595A (en) * | 2015-05-22 | 2015-09-23 | 齐鲁工业大学 | Collaborative filtering recommendation method based on Markov prediction model |
| CN105866756A (en) * | 2016-03-29 | 2016-08-17 | 西安电子科技大学 | Staring imaging method of uniform area array emitting radar based on tensor compression perception |
| CN110263232A (en) * | 2019-05-21 | 2019-09-20 | 安徽建筑大学 | A kind of mixed recommendation method based on range study and deep learning |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2012126741A2 (en) * | 2011-03-18 | 2012-09-27 | Telefonica, S.A. | Method for context-aware recommendations based on implicit user feedback |
| CN104331411B (en) * | 2014-09-19 | 2018-01-09 | 华为技术有限公司 | The method and apparatus of recommended project |
| CN105550950A (en) * | 2015-11-20 | 2016-05-04 | 广东工业大学 | Location-based service travel recommendation method |
| CN106651549B (en) * | 2017-01-09 | 2019-10-01 | 山东大学 | A kind of personalized automobile recommended method and system merging supply and demand chain |
| CN107679242B (en) * | 2017-10-30 | 2018-07-27 | 河海大学 | Merge the label recommendation method that multiple information sources Coupling Tensor is decomposed |
| CN108521586B (en) * | 2018-03-20 | 2020-01-14 | 西北大学 | IPTV television program personalized recommendation method giving consideration to time context and implicit feedback |
-
2019
- 2019-12-26 CN CN201911365828.6A patent/CN113051463B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104933595A (en) * | 2015-05-22 | 2015-09-23 | 齐鲁工业大学 | Collaborative filtering recommendation method based on Markov prediction model |
| CN105866756A (en) * | 2016-03-29 | 2016-08-17 | 西安电子科技大学 | Staring imaging method of uniform area array emitting radar based on tensor compression perception |
| CN110263232A (en) * | 2019-05-21 | 2019-09-20 | 安徽建筑大学 | A kind of mixed recommendation method based on range study and deep learning |
Non-Patent Citations (1)
| Title |
|---|
| 张宏丽 等. 利用最近邻域推荐且结合情境感知的个性化推荐算法.《计算机科学》.2019,第46卷(第4期),235-240. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113051463A (en) | 2021-06-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Kumari et al. | Multimedia big data computing and Internet of Things applications: A taxonomy and process model | |
| CN105320957B (en) | Classifier training method and device | |
| US20190122093A1 (en) | Artificial intelligence system for supporting communication | |
| KR101645149B1 (en) | Data synchronization | |
| CN103678647B (en) | A kind of method and system for realizing information recommendation | |
| EP2741473B1 (en) | Human-machine interaction data processing method and apparatus | |
| US20160117574A1 (en) | Tagging Personal Photos with Deep Networks | |
| CN103793465A (en) | Cloud computing based real-time mass user behavior analyzing method and system | |
| US20120102018A1 (en) | Ranking Model Adaptation for Domain-Specific Search | |
| CN113379176B (en) | Method, device, equipment and readable storage medium for detecting abnormal data of telecommunication network | |
| Adams et al. | Sensing and using social context | |
| CN110119478B (en) | Similarity-based item recommendation method combining multiple user feedback data | |
| US11768871B2 (en) | Systems and methods for contextualizing computer vision generated tags using natural language processing | |
| US20200118033A1 (en) | Method for approximate k-nearest-neighbor search on parallel hardware accelerators | |
| CN109241442B (en) | Project recommendation method based on predictive value filling, readable storage medium and terminal | |
| US20230169402A1 (en) | Collaborative machine learning | |
| CN108389113B (en) | Collaborative filtering recommendation method and system | |
| CN110851609A (en) | Representation learning method and device | |
| CN113641797A (en) | Data processing method, device, equipment, storage medium and computer program product | |
| CN108885637A (en) | Personage is mild-natured related | |
| CN115186177A (en) | Social information recommendation method and device, computer equipment and storage medium | |
| CN113051463B (en) | Project pushing method and system | |
| Guo et al. | PicPick: a generic data selection framework for mobile crowd photography | |
| Arnaboldi et al. | Droidopppathfinder: A context and social-aware path recommender system based on opportunistic sensing | |
| Lau et al. | Enhancing collaborative filtering of learning resources with semantically-enhanced social tags |
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 |