CN112749345B - K neighbor matrix decomposition recommendation method based on neural network - Google Patents
K neighbor matrix decomposition recommendation method based on neural network Download PDFInfo
- Publication number
- CN112749345B CN112749345B CN202110176380.4A CN202110176380A CN112749345B CN 112749345 B CN112749345 B CN 112749345B CN 202110176380 A CN202110176380 A CN 202110176380A CN 112749345 B CN112749345 B CN 112749345B
- Authority
- CN
- China
- Prior art keywords
- matrix
- items
- neural network
- user
- scoring
- 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
- 239000011159 matrix material Substances 0.000 title claims abstract description 100
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 44
- 238000000034 method Methods 0.000 title claims abstract description 28
- 238000000354 decomposition reaction Methods 0.000 title claims abstract description 25
- 239000013598 vector Substances 0.000 claims description 39
- 230000006870 function Effects 0.000 claims description 19
- 238000004364 calculation method Methods 0.000 claims description 11
- 210000002569 neuron Anatomy 0.000 claims description 10
- 238000010606 normalization Methods 0.000 claims description 10
- 210000004205 output neuron Anatomy 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 8
- 238000013507 mapping Methods 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 6
- 230000011218 segmentation Effects 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 5
- 230000009466 transformation Effects 0.000 claims description 4
- 238000011156 evaluation Methods 0.000 claims description 3
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 210000002364 input neuron Anatomy 0.000 claims description 3
- 230000009471 action Effects 0.000 description 5
- 238000001914 filtration Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000006399 behavior Effects 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/906—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/088—Non-supervised learning, e.g. competitive learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Software Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a K neighbor matrix decomposition recommendation method based on a neural network, which comprises the following steps: acquiring a data set from a website, selecting a plurality of users and a plurality of items in the data set, and forming an initial matrix according to the plurality of users and the plurality of items; using RBF neural network to predict and fill partial blank values of the initial matrix so as to reduce the sparseness of the initial matrix; calculating K nearest neighbor users in the initial matrix obtained in the step S2 by using a KNN algorithm for the target user, similarly finding K nearest neighbor items for the target item, and constructing a scoring matrix according to the K nearest neighbor users and the K nearest neighbor items; performing matrix decomposition on the scoring matrix, wherein the U matrix represents the characteristics of a target user, the V matrix represents the characteristics of a target item, extracting new implicit characteristics and increasing the explanatory property of the decomposition; multiplying the determinant of the scoring matrix by 1/k as a scoring criterion, and giving a recommendation threshold value to determine whether to give a recommendation.
Description
Technical Field
The invention relates to the field of recommendation systems, in particular to a K neighbor matrix decomposition recommendation method based on a neural network.
Background
With the continuous increase of online information quantity, online active users are continuously increased, the users cannot accurately grasp information on the Internet, so that the user needs to be met, the experience of surfing the Internet is poor, and the recommendation system becomes an effective strategy for overcoming information overload. The recommendation system extracts items (such as information, services, articles and the like) interested by the user from the massive data through a recommendation algorithm according to the requirements, interests and the like of the user, and recommends the results to the user in the form of a personalized list.
At present, the recommendation system algorithm is roughly divided into three types, namely a recommendation system based on content, a recommendation system based on collaborative filtering and a mixed recommendation system. The difficulty of the content-based recommendation system is that the content information of the user is acquired and filtered, the difficulty of the hybrid recommendation system is that the information content of the user is combined with the entity information of the user, and the relation between the information content of the user and the entity information of the user is expressed as the problem of structured data, so that collaborative filtering becomes the most mainstream recommendation system algorithm at present.
In order to solve the problems of matrix sparsity and expansibility in a collaborative filtering algorithm, when high-quality information and edge information of a user cannot be acquired, a machine is enabled to learn and predict behavior characteristics of the user from an incomplete information matrix, a K nearest neighbor matrix decomposition algorithm based on a neural network is provided, and the method has important significance in mining the relation between the user and the project under the condition of less information quantity.
Disclosure of Invention
The invention aims to provide a K neighbor matrix decomposition recommendation method based on a neural network, which is characterized in that a deletion matrix is complemented by using the current popular neural network according to a public data set provided by a website, similar users and similar items are clustered by using a clustering technology, and finally the items favored by the users are recommended to the corresponding users in a recommendation list mode.
In order to achieve the above purpose, the technical scheme adopted by the invention is as follows:
a K-nearest neighbor matrix decomposition recommendation method based on a neural network comprises the following steps:
step S1, acquiring a data set from a website, selecting a plurality of users and a plurality of items in the data set, and forming an initial matrix according to the plurality of users and the plurality of items;
s2, predicting and filling partial blank values of the initial matrix by using an RBF neural network so as to reduce the sparsity of the initial matrix;
s3, calculating the initial matrix obtained in the S2 by using a KNN algorithm for the target user, finding K nearest neighbor users in the initial matrix, and similarly finding K nearest neighbor items for the target item, and constructing a scoring matrix according to the K nearest neighbor users and the K nearest neighbor items;
s4, carrying out matrix decomposition on the scoring matrix, wherein a U matrix represents the characteristics of a target user, a V matrix represents the characteristics of a target item and the latent meaning characteristics U, V, so that the explanatory property of decomposition is improved;
and S5, multiplying determinant of the scoring matrix by 1/k as scoring standard, giving a recommendation threshold value, and judging whether recommendation is given.
Further, the method further comprises a preprocessing step for the initial matrix, the preprocessing step comprising:
deleting the users with the user sparsity of more than 99.5% from the plurality of users, wherein the calculation formula of the user sparsity a is as follows:
a=1-the user evaluates the excessive number of items/total number of items;
and deleting the items without any score in the plurality of items.
Further, the method further comprises normalization processing for the initial matrix, wherein the normalization processing is used for uniformly mapping the scoring values to the intervals of [0,1], and the normalization processing is specifically implemented by the following calculation formula:
wherein y represents the initial item score value, y min Lower bound score value, y, representing raw project data max Representing the upper bound score value of the raw data, and Y represents the normalized data.
Further, the step S2 specifically includes:
s201, normalizing the user vectors with a plurality of common scoring items, and then using the normalized user vectors as the input of the RBF neural network, and determining the number of input layers of the neural network; the RBF neural network trains the large-scale data fast, selects the local optimum instead of the global optimum, and prevents the overfitting;
s202, randomly selecting a batch of center nodes, adopting an unsupervised gradient descent method, and updating the center nodes through negative feedback to finally determine the number of the center nodes of the hidden layer;
s203, the hidden layer neuron kernel function is a Gaussian function, the input information is subjected to space mapping transformation, and the output layer uses a linear weighting function;
s204, performing inverse normalization processing on the output of the neural network to obtain a prediction score in the range of [0,5 ];
s205, filling the blank values by using the trained neural network, filling the unpredictable blank values with the mean value, and reducing the sparseness of a new scoring matrix to be close to 0.
Further, the center, width and adjustment weight parameters are adaptively adjusted to the optimal values through learning, and the iterative calculation is as follows:
W kj (t) is the adjustment weight between the kth output neuron and the jth hidden layer neuron at the time of the t-th iterative computation;
C pq (t) is the central component of the p-th hidden layer neuron for the q-th input neuron at the time of the t-th iterative computation;
d ij (t) is ANDCenter C pq (t) a corresponding radial width;
η is a learning factor;
e is an RBF neural network evaluation function:
wherein y is lk A desired output value for the kth output neuron at the ith input sample; o (O) lk Is the network output value of the kth output neuron at the ith input sample.
Further, the step S3 includes:
grouping the new scoring matrix into user vectors according to row segmentation, dividing the user vectors into item vectors according to column segmentation, solving the similarity between a new target user i vector and the user vectors, solving the similarity between a new target item j and the target user, and respectively solving the first K values for the similarity user vectors and the item vectors in sequence, wherein the K user vectors and the K item vectors form a new scoring matrix; wherein the similarity calculation method is shown as the formula (2):
wherein x is i Representing the element values in the target user i vector, y i Representing the corresponding element values in the user vector, sim represents the calculated similarity value.
Further, the step S4 includes the following specific steps:
SVD carries out matrix decomposition on the new K-dimensional matrix M to obtain a decomposed matrix SR, wherein U represents the overall lingering semantic features of the target user, V represents the overall lingering semantic features of the target item, elements in the sigma are singular values of the matrix, and new KR (K x K) =U (K x r) sigma (r x r) V (r x K) is continuously constructed according to the first r singular values in the sigma, so that the dimension of the lingering semantic matrix is reduced, and the calculated amount is reduced.
Further, the step S5 includes the following specific steps:
multiplying the value of determinant of the scoring matrix by 1/k to be used as the predictive score of target user I on target item j, setting a threshold value c=4, adding items with predictive scores greater than or equal to 4 into a recommendation set Si, wherein Si comprises elements which are scoring items greater than or equal to 4 in all scores of I users, sorting the items in the Si set according to the score size, and recommending the items of the front Top5 to the user I, wherein the user I belongs to a set I= { I1, I2, I3, …, I943}.
Compared with the prior art, the invention has at least one of the following advantages:
when the high-quality content information and the edge information of the user cannot be acquired, the invention can provide interesting movie items for different users according to single scores of the users, but is not limited to movie recommendation, such as music recommendation, electronic book recommendation, news recommendation and the like.
The sparsity of the initial scoring matrix (initial matrix for short) is effectively reduced by utilizing the strong learning ability of the neural network, and the recommendation result of the final result is greatly influenced.
The new users and the new projects are introduced and scoring prediction is carried out, the expansibility of a recommendation system is increased, and the recommendation system can recommend the new users and the new projects which are continuously added more and more accurately.
Drawings
FIG. 1 is a general flow chart of a recommendation method for K-nearest neighbor matrix factorization based on a neural network in an embodiment of the invention;
fig. 2 is a schematic diagram of an RBF neural network according to an embodiment of the invention.
Detailed Description
The invention is described in further detail below with reference to figures 1-2 and the detailed description. The advantages and features of the present invention will become more apparent from the following description. It should be noted that the drawings are in a very simplified form and are all to a non-precise scale, merely for the purpose of facilitating and clearly aiding in the description of embodiments of the invention. For a better understanding of the invention with objects, features and advantages, refer to the drawings. It should be understood that the structures, proportions, sizes, etc. shown in the drawings are for illustration purposes only and should not be construed as limiting the invention to the extent that any modifications, changes in the proportions, or adjustments of the sizes of structures, proportions, or otherwise, used in the practice of the invention, are included in the spirit and scope of the invention which is otherwise, without departing from the spirit or essential characteristics thereof.
It is noted that in the present invention, relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, neural network-based K-nearest neighbor matrix factorization recommendation method, article, or field device that comprises a series of elements includes not only those elements, but also other elements not explicitly listed, or also elements inherent to such a process, neural network-based K-nearest neighbor matrix factorization recommendation method, article, or field device. Without further limitation, the element defined by the statement "comprising one … …" does not exclude that there are also additional identical elements in the process comprising said element, the neural network based K-nearest neighbor matrix factorization recommendation method, the article or the field device.
Referring to fig. 1-2, the method for K-nearest neighbor matrix decomposition recommendation based on a neural network provided in this embodiment includes:
step S1, acquiring a data set from a website, selecting 943 users and 11508 items in the data set, and forming an initial matrix according to the plurality of users and the plurality of items;
s2, predicting and filling partial blank values of the initial matrix by using an RBF neural network so as to reduce the sparsity of the initial matrix;
s3, calculating the initial matrix obtained in the S2 by using a KNN algorithm for the target user, finding K nearest neighbor users in the initial matrix, and similarly finding K nearest neighbor items for the target item, and constructing a scoring matrix according to the K nearest neighbor users and the K nearest neighbor items;
s4, SVD carries out matrix singular value decomposition on the scoring matrix, wherein a U matrix represents the characteristics of a target user, a V matrix represents the characteristics of a target item and the latent meaning characteristics U, V, so that the interpretation of the decomposition is improved;
and S5, multiplying determinant of the scoring matrix by 1/k as scoring standard, giving a recommendation threshold value, and judging whether recommendation is given.
In this embodiment, in step S3, the magnitude of the parameter K is determined specifically using the magnitude of the Mean Square Error (MSE), and the MSE is calculated as follows:
wherein x is i The actual item is scored for the purpose of,scoring criteria for step S5;
in this embodiment, the method further includes a preprocessing step for the initial matrix, where the preprocessing step includes: deleting nonsensical rows and columns, and deleting users with the user sparsity of more than 99.5% from the plurality of users, wherein the user sparsity a is calculated as follows because the users with the sparsity of too high have no reference value and the final result is not affected by discarding the users with the sparsity of too high as reference value:
a=1-the user evaluates the excessive number of items/total number of items;
and deleting the items without any score in the plurality of items.
In this embodiment, the method further includes normalization processing for the initial matrix, where the normalization processing is used to map the score value onto the interval of [0,1] in a unified manner, specifically by the following calculation formula:
wherein y represents the initial item score value, y min Lower bound score value, y, representing raw project data max Representing the upper bound score value of the raw data, and Y represents the normalized data.
In this embodiment, the step S2 specifically includes:
s201, normalizing the user vectors with a plurality of common scoring items, and then using the normalized user vectors as the input of the RBF neural network, and determining the number of input layers of the neural network; the RBF neural network trains the large-scale data fast, selects the local optimum instead of the global optimum, and prevents the overfitting;
s202, randomly selecting a batch of center nodes, adopting an unsupervised gradient descent method, and updating the center nodes through negative feedback to finally determine the number of the center nodes of the hidden layer;
s203, the hidden layer neuron kernel function (action function) is a Gaussian function, and the input information is subjected to transformation of space mapping;
s204, performing inverse normalization processing on the output of the neural network to obtain a prediction score in the range of [0,5 ];
s205, filling the blank values by using the trained neural network, filling the unpredictable blank values with the mean value, and reducing the sparseness of a new scoring matrix to be close to 0.
In this embodiment, the center, width and adjustment weight parameters are all adaptively adjusted to the optimal values by learning, and the iterative calculation is as follows:
W kj (t) is the adjustment weight between the kth output neuron and the jth hidden layer neuron at the time of the t-th iterative computation;
C pq (t) is the central component of the p-th hidden layer neuron for the q-th input neuron at the time of the t-th iterative computation;
d ij (t) is with the center C pq (t) a corresponding width;
η is a learning factor, the learning factor is cycled to set an initial value of η=0.001, cycled 100 times, each stride of 0.001, and a final value of η is determined;
e is an RBF neural network evaluation function:
wherein y is lk A desired output value for the kth output neuron at the ith input sample; o (O) lk Is the network output value of the kth output neuron at the ith input sample.
In this embodiment, in step S203, the hidden layer neuron kernel function (function) is a gaussian function, and performs a transformation of spatial mapping on the input information; the data may be distinguishable in a low-dimensional space and indistinguishable in a high-dimensional space, the Gaussian kernel is essentially a measure of the similarity between samples, and in a space depicting the similarity, the similar samples are better clustered together, and then linearly separable, and the Gaussian function is as follows:
wherein Xc is the kernel function center, sigma is the width parameter of the function, the radial action range of the function is controlled, and yi is the final network output.
In this embodiment, the step S3 includes: grouping the new scoring matrix into user vectors according to row segmentation, dividing the user vectors into item vectors according to column segmentation, solving the similarity between a new target user i vector and the user vectors, solving the similarity between a new target item j and the target user, and respectively solving the first K values for the similarity user vectors and the item vectors in sequence, wherein the K user vectors and the K item vectors form a new scoring matrix; wherein the similarity calculation method is shown as the formula (2):
wherein x is i Representing the element values in the target user i vector, y i Representing the corresponding element values in the user vector, sim represents the calculated similarity value.
In this embodiment, the step S4 includes the following specific steps: SVD carries out matrix decomposition on the new K-dimensional matrix M to obtain a decomposed matrix SR, wherein U represents the overall lingering features of the target user, V represents the overall lingering features of the target item, elements in the sigma are singular values of the matrix, and new KR (K x K) =U (K x r) sigma (r x r) V (r x K) is continuously constructed according to the first r (r < K) singular values in the sigma, so that the dimension of the lingering matrix is reduced, and the calculated amount is reduced.
In this embodiment, the step S5 includes the following specific steps: multiplying the value of determinant of the scoring matrix by 1/k to be used as the predictive score of target user I on target item j, setting a threshold value c=4, adding items with predictive scores greater than or equal to 4 into a recommendation set Si, wherein Si comprises elements which are scoring items greater than or equal to 4 in all scores of I users, sorting the items in the Si set according to the score size, and recommending the items of the front Top5 to the user I, wherein the user I belongs to a set I= { I1, I2, I3, …, I943}.
The present document provides movie items of interest to different users based on only a single score of the user, but is not limited to movie recommendations such as music recommendation, e-book recommendation, news recommendation, etc.
While the present invention has been described in detail through the foregoing description of the preferred embodiment, it should be understood that the foregoing description is not to be considered as limiting the invention. Many modifications and substitutions of the present invention will become apparent to those of ordinary skill in the art upon reading the foregoing. Accordingly, the scope of the invention should be limited only by the attached claims.
Claims (7)
1. The K neighbor matrix decomposition recommendation method based on the neural network is characterized by comprising the following steps of:
step S1, acquiring a data set from a website, selecting a plurality of users and a plurality of items in the data set, and forming an initial matrix according to the plurality of users and the plurality of items;
s2, performing prediction filling on partial blank values of the initial matrix by using an RBF neural network to reduce the sparsity of the initial matrix,
the step S2 specifically includes:
s201, normalizing the user vectors with a plurality of common scoring items, and then using the normalized user vectors as the input of the RBF neural network, and determining the number of input layers of the neural network; the RBF neural network trains the large-scale data fast, selects the local optimum instead of the global optimum, and prevents the overfitting;
s202, randomly selecting a batch of center nodes, adopting an unsupervised gradient descent method, and updating the center nodes through negative feedback to finally determine the number of the center nodes of the hidden layer;
s203, the hidden layer neuron kernel function is a Gaussian function, the input information is subjected to space mapping transformation, and the output layer uses a linear weighting function;
s204, performing inverse normalization processing on the output of the neural network to obtain a prediction score in the range of [0,5 ];
s205, filling the blank values by using a trained neural network, filling the unpredictable blank values with a mean value, and reducing the sparseness of a new scoring matrix to be close to 0;
s3, calculating the initial matrix obtained in the S2 by using a KNN algorithm for the target user, finding K nearest neighbor users in the initial matrix, and similarly finding K nearest neighbor items for the target item, and constructing a scoring matrix according to the K nearest neighbor users and the K nearest neighbor items;
s4, performing matrix decomposition on the scoring matrix, wherein a U matrix represents the characteristics of a target user, a V matrix represents the characteristics of a target item, and the latent meaning characteristics U, V increase the explanatory property of the decomposition;
and S5, multiplying determinant of the scoring matrix by 1/k as scoring standard, giving a recommendation threshold value, and judging whether recommendation is given.
2. The neural network-based K-nearest neighbor matrix decomposition recommendation method of claim 1, further comprising a preprocessing step of the initial matrix, the preprocessing step comprising:
deleting the users with the user sparsity of more than 99.5% from the plurality of users, wherein the calculation formula of the user sparsity a is as follows:
a=1-the user evaluates the excessive number of items/total number of items;
and deleting the items without any score in the plurality of items.
3. The neural network-based K-nearest neighbor matrix decomposition recommendation method according to claim 1 or 2, further comprising a normalization process for the initial matrix, wherein the normalization process is used for uniformly mapping the scoring values to the intervals of [0,1], specifically by a calculation formula as follows:
wherein y represents the initial item score value, y min Lower bound score value, y, representing raw project data max Representing the upper bound score value of the raw data, and Y represents the normalized data.
4. The K nearest neighbor matrix factorization recommendation method based on a neural network of claim 3,
the center, width and adjustment weight parameters are adaptively adjusted to the optimal values through learning, and the iterative calculation is as follows:
W kj (t) is the adjustment weight between the kth output neuron and the jth hidden layer neuron at the time of the t-th iterative computation;
C pq (t) is the central component of the p-th hidden layer neuron for the q-th input neuron at the time of the t-th iterative computation;
d ij (t) is with the center C pq (t) a corresponding radial width;
η is a learning factor;
e is an RBF neural network evaluation function:
wherein y is lk A desired output value for the kth output neuron at the ith input sample; o (O) lk Is the network output value of the kth output neuron at the ith input sample.
5. The neural network-based K-nearest neighbor matrix decomposition recommendation method of claim 1, wherein said step S3 comprises:
grouping the new scoring matrix into user vectors according to row segmentation, dividing the user vectors into item vectors according to column segmentation, solving the similarity between a new target user i vector and the user vectors, solving the similarity between a new target item j and the target user, and respectively solving the first K values for the similarity user vectors and the item vectors in sequence, wherein the K user vectors and the K item vectors form a new scoring matrix; wherein the similarity calculation method is shown as the formula (2):
wherein x is i Representing the element values in the target user i vector, y i Representing the corresponding element values in the user vector, sim represents the calculated similarity value.
6. The neural network-based K-nearest neighbor matrix decomposition recommendation method of claim 1, wherein said step S4 comprises the specific steps of:
SVD carries out matrix decomposition on the new K-dimensional matrix M to obtain a decomposed matrix SR, wherein U represents the overall lingering semantic features of the target user, V represents the overall lingering semantic features of the target item, elements in the sigma are singular values of the matrix, and new KR (K x K) =U (K x r) sigma (r x r) V (r x K) is continuously constructed according to the first r singular values in the sigma, so that the dimension of the lingering semantic matrix is reduced, and the calculated amount is reduced.
7. The neural network-based K-nearest neighbor matrix decomposition recommendation method of claim 1, wherein said step S5 comprises the specific steps of:
multiplying the value of determinant of the scoring matrix by 1/k to be used as the predictive score of target user I on target item j, setting a threshold value c=4, adding items with predictive scores greater than or equal to 4 into a recommendation set Si, wherein Si comprises elements which are scoring items greater than or equal to 4 in all scores of I users, sorting the items in the Si set according to the score size, and recommending the items of the front Top5 to the user I, wherein the user I belongs to a set I= { I1, I2, I3, …, I943}.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110176380.4A CN112749345B (en) | 2021-02-09 | 2021-02-09 | K neighbor matrix decomposition recommendation method based on neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110176380.4A CN112749345B (en) | 2021-02-09 | 2021-02-09 | K neighbor matrix decomposition recommendation method based on neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112749345A CN112749345A (en) | 2021-05-04 |
| CN112749345B true CN112749345B (en) | 2023-11-14 |
Family
ID=75651970
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110176380.4A Active CN112749345B (en) | 2021-02-09 | 2021-02-09 | K neighbor matrix decomposition recommendation method based on neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112749345B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113536140B (en) * | 2021-09-16 | 2022-02-11 | 广东众聚人工智能科技有限公司 | Real-time recommendation method and system based on small sample data |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103106535A (en) * | 2013-02-21 | 2013-05-15 | 电子科技大学 | Method for solving collaborative filtering recommendation data sparsity based on neural network |
| CN108038629A (en) * | 2017-12-30 | 2018-05-15 | 北京工业大学 | A kind of optimization method based on collaborative filtering |
| CN108090229A (en) * | 2018-01-10 | 2018-05-29 | 广东工业大学 | A kind of method and apparatus that rating matrix is determined based on convolutional neural networks |
| CN108320187A (en) * | 2018-02-02 | 2018-07-24 | 合肥工业大学 | A kind of recommendation method based on depth social networks |
| CN109902201A (en) * | 2019-03-08 | 2019-06-18 | 天津理工大学 | A kind of recommended method based on CNN and BP neural network |
| CN110162706A (en) * | 2019-05-22 | 2019-08-23 | 南京邮电大学 | A kind of personalized recommendation method and system based on interaction data cluster |
| CN111598627A (en) * | 2020-05-26 | 2020-08-28 | 揭阳职业技术学院 | Personalized advertisement pushing method for elevator media terminal |
| CN111882381A (en) * | 2020-06-30 | 2020-11-03 | 桂林电子科技大学 | Travel recommendation method based on collaborative memory network |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10803386B2 (en) * | 2018-02-09 | 2020-10-13 | Twitter, Inc. | Matching cross domain user affinity with co-embeddings |
-
2021
- 2021-02-09 CN CN202110176380.4A patent/CN112749345B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103106535A (en) * | 2013-02-21 | 2013-05-15 | 电子科技大学 | Method for solving collaborative filtering recommendation data sparsity based on neural network |
| CN108038629A (en) * | 2017-12-30 | 2018-05-15 | 北京工业大学 | A kind of optimization method based on collaborative filtering |
| CN108090229A (en) * | 2018-01-10 | 2018-05-29 | 广东工业大学 | A kind of method and apparatus that rating matrix is determined based on convolutional neural networks |
| CN108320187A (en) * | 2018-02-02 | 2018-07-24 | 合肥工业大学 | A kind of recommendation method based on depth social networks |
| CN109902201A (en) * | 2019-03-08 | 2019-06-18 | 天津理工大学 | A kind of recommended method based on CNN and BP neural network |
| CN110162706A (en) * | 2019-05-22 | 2019-08-23 | 南京邮电大学 | A kind of personalized recommendation method and system based on interaction data cluster |
| CN111598627A (en) * | 2020-05-26 | 2020-08-28 | 揭阳职业技术学院 | Personalized advertisement pushing method for elevator media terminal |
| CN111882381A (en) * | 2020-06-30 | 2020-11-03 | 桂林电子科技大学 | Travel recommendation method based on collaborative memory network |
Non-Patent Citations (3)
| Title |
|---|
| ZhaoWei Qu 等.Converged Recommendation System Based on RNN and BP Neural Networks.《2018 IEEE International Conference on Big Data and Smart Computing (BigComp)》.2018,308-313. * |
| 孙坤.基于改进协同过滤的个性化推荐方法研究.《中国优秀硕士学位论文全文数据库信息科技辑》.2020,(第2期),I138-2237. * |
| 张锋 等.使用BP神经网络缓解协同过滤推荐算法的稀疏性问题.《计算机研究与发展》.2006,667-672. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112749345A (en) | 2021-05-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110363282B (en) | Network node label active learning method and system based on graph convolution network | |
| CN105224872B (en) | A kind of user's anomaly detection method based on neural network clustering | |
| Taneja et al. | Modeling user preferences using neural networks and tensor factorization model | |
| CN108897791B (en) | Image retrieval method based on depth convolution characteristics and semantic similarity measurement | |
| CN112967088A (en) | Marketing activity prediction model structure and prediction method based on knowledge distillation | |
| Navgaran et al. | Evolutionary based matrix factorization method for collaborative filtering systems | |
| CN112529638B (en) | Service demand dynamic prediction method and system based on user classification and deep learning | |
| Tiwari | Introduction to machine learning | |
| EP3920100A1 (en) | Adaptive pointwise-pairwise learning to rank | |
| Deodhar et al. | A framework for simultaneous co-clustering and learning from complex data | |
| CN110727872A (en) | Method and device for mining ambiguous selection behavior based on implicit feedback | |
| CN113343077A (en) | Personalized recommendation method and system integrating user interest time sequence fluctuation | |
| CN112749345B (en) | K neighbor matrix decomposition recommendation method based on neural network | |
| Chen et al. | Poverty/investment slow distribution effect analysis based on Hopfield neural network | |
| CN113221003A (en) | Mixed filtering recommendation method and system based on dual theory | |
| Luo et al. | Addressing imbalance in weakly supervised multi-label learning | |
| CN109885758A (en) | A kind of recommended method of the novel random walk based on bigraph (bipartite graph) | |
| Chen et al. | Gaussian mixture embedding of multiple node roles in networks | |
| Shen et al. | On image classification: Correlation vs causality | |
| Liwei | Predictive analysis of machine learning error classification based on bayesian network | |
| Liang et al. | Incremental deep forest for multi-label data streams learning | |
| CN112464098A (en) | Recommendation system article prediction method based on similarity pairwise ranking | |
| CN115098787B (en) | Article recommendation method based on cosine ranking loss and virtual edge map neural network | |
| Li et al. | Prediction model of social trust relationship in network environment based on organizational behavior theory | |
| CN115292587B (en) | Recommendation method and system based on knowledge distillation and causal reasoning |
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 |