CN115408621B - Interest point recommendation method considering auxiliary information characteristic linear and nonlinear interaction - Google Patents

Interest point recommendation method considering auxiliary information characteristic linear and nonlinear interaction Download PDF

Info

Publication number
CN115408621B
CN115408621B CN202210968531.4A CN202210968531A CN115408621B CN 115408621 B CN115408621 B CN 115408621B CN 202210968531 A CN202210968531 A CN 202210968531A CN 115408621 B CN115408621 B CN 115408621B
Authority
CN
China
Prior art keywords
poi
user
information
auxiliary information
linear
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
Application number
CN202210968531.4A
Other languages
Chinese (zh)
Other versions
CN115408621A (en
Inventor
李晓燕
徐胜华
姜涛
刘纪平
王勇
罗安
车向红
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Chinese Academy of Surveying and Mapping
Original Assignee
Chinese Academy of Surveying and Mapping
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Chinese Academy of Surveying and Mapping filed Critical Chinese Academy of Surveying and Mapping
Priority to CN202210968531.4A priority Critical patent/CN115408621B/en
Publication of CN115408621A publication Critical patent/CN115408621A/en
Application granted granted Critical
Publication of CN115408621B publication Critical patent/CN115408621B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06F16/9536—Search customisation based on social or collaborative filtering
    • 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
    • G06F16/9537—Spatial or temporal dependent retrieval, e.g. spatiotemporal queries
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention relates to the technical field of computers, in particular to a point of interest recommendation method considering the linear and nonlinear interaction of auxiliary information features, which comprises POI auxiliary information, user auxiliary information, auxiliary information construction based on a convolutional attention mechanism and a neural matrix decomposition model. And learning linear and nonlinear interaction relations between the user and potential feature vectors of the POI by using a neural matrix decomposition model, and calculating a preference score of the user to the POI.

Description

Interest point recommendation method considering auxiliary information characteristic linear and nonlinear interaction
Technical Field
The invention relates to the technical field of computers, in particular to a point of interest recommendation method considering linear and nonlinear interaction of auxiliary information features.
Background
Point of interest (POI) recommendations are becoming a research focus in the fields of geographic information systems, web information retrieval, and the like, as one of important services of location-based social networks. POI recommendation is various life patterns and personal preferences of a user hidden after large-scale data mining, so as to help the user to find a most interesting personalized place, and further enrich the life experience of the user. Meanwhile, related service providers are helped to provide intelligent, precise and personalized services for potential users, and the use viscosity of the users on the social network service platform based on the position is greatly enhanced, so that the economic benefit is improved.
However, in the prior art, a single matrix decomposition or a deep neural network is mostly used, which cannot effectively capture a complex structure of interaction between a user and a POI, and cannot effectively deal with the implicit feedback problem. In addition, in order to alleviate the problem of data sparseness, the current technology mainly introduces auxiliary information of users and POIs, and the auxiliary information is often judged to have the same value, which enables some valuable information to be resolved.
Therefore, in order to solve the above problems, the present application provides a point of interest recommendation method considering the linear and nonlinear interaction of auxiliary information features, which utilizes location clustering and a TF-IDF algorithm to mine and construct auxiliary information of a user and a POI from user sign-in data, enriches the expressions of potential feature vectors of the user and the POI, learns potential representations from the auxiliary information of the user and the POI by utilizing a convolutional neural network, introduces an attention mechanism to distinguish the importance of the constructed auxiliary information of the user and the POI, constructs a neural matrix decomposition model, learns the linear and nonlinear interaction between the user and the POI, and calculates a preference score of the user to the POI.
Disclosure of Invention
The invention aims to fill the blank of the prior art, provides a point of interest recommendation method considering the linear and nonlinear interaction of auxiliary information features, utilizes position clustering and TF-IDF algorithm to mine and construct the auxiliary information of a user and a POI from user sign-in data, enriches the expression of potential feature vectors of the user and the POI, utilizes a convolutional neural network to learn potential representation from the auxiliary information of the user and the POI, introduces an attention mechanism to distinguish the importance of the constructed auxiliary information of the user and the POI, constructs a neural matrix decomposition model, learns the linear and nonlinear interaction between the user and the POI, and calculates the preference score of the user to the POI. In order to achieve the purpose, the invention provides a point of interest recommendation method considering the linear and nonlinear interaction of auxiliary information characteristics, which mainly comprises POI auxiliary information, user auxiliary information, auxiliary information construction based on a convolution attention mechanism and a neural matrix decomposition model;
the POI auxiliary information comprises a POI category, a POI popularity and a POI belonging area;
the POI category is inherent information;
the POI popularity is the popularity of the POI to the user, and the check-in data in the POI is evaluated through check-in data set, wherein the check-in data not only contains information about inherent attributes of the POI, but also contains the check-in times of the POI;
POI popularity is calculated by TF-IDF algorithm based on check-in data, formula as follows:
Figure GDA0004104679780000021
sizeof(v i ) Denotes POIv i Number of check-ins of (Sv) i ) Is represented by the formula i Check-in times, sizeof (Cv), for all POIs of the same class i ) Is represented by the formula i The number of all POIs in the same category, n represents the number of interest points;
the method comprises the following steps of clustering POI (point of interest) by utilizing a position clustering algorithm according to position information of the POI to obtain a proper region block, and enabling each POI to be allocated with a position label of a corresponding region, wherein the processing steps are as follows:
s1, randomly selecting k POIs from a POI set as an initial clustering center;
s2, calculating the Euclidean distance rho from the residual POI to the clustering center, and putting the closest interest points into corresponding categories to form new categories, wherein the rho calculation method comprises the following steps:
Figure GDA0004104679780000031
lat i 、lat j respectively represent POIv i 、v j Latitude information of (lo) i 、lon j Respectively represent POIv i 、v j Longitude information of (a);
s3, taking the mean value of longitude and latitude of all POI in the current cluster as a new central point, and updating the POI closest to the cluster center;
s4, until the target function is converged or the clustering center is not changed, otherwise, transferring to the step S2;
s5, outputting a POI clustering result;
the user auxiliary information comprises a favorite category of the user and a high activity position of the user, wherein the favorite category of the user refers to a POI category which is frequently signed in by the user;
the auxiliary information based on the convolution attention mechanism is constructed as follows:
and carrying out one-hot coding on the user ID and the interest point ID, normalizing the numerical type variable in order to apply auxiliary attribute information of the user and the POI in the model, and carrying out one-hot coding on the category type variable. User potential feature vector U u POI latent feature vector V v User auxiliary information U a And POI auxiliary information V a Are fed into the embedding layer, U, respectively u 、V v 、U a And V a The potential feature vector of (a) is calculated as follows:
Figure GDA0004104679780000032
f. r, t and l are representational functions;
learning potential representations from user and POI assistance information using a convolutional neural network, which consists of convolutional and pooling layers that can extract deeper level features, and is performed as follows:
Figure GDA0004104679780000041
* Is a convolution operator, w is a filter, b is the bias of w, g is a nonlinear activation function, and posing is a pooling function;
the importance of user and POI assistance information is distinguished by adding an attention mechanism that uses a softmax (Q) function to compute a score as a probability distribution of assistance information output, and the output is summed with
Figure GDA0004104679780000042
Figure GDA0004104679780000043
Combined by element multiplication>
Figure GDA0004104679780000044
The final output of the attention mechanism is obtained, and the attention mechanism is shown as follows: />
Figure GDA0004104679780000045
The softmax (Q) function is defined as:
Figure GDA0004104679780000046
u jc representing one of the attributes;
user feature vector
Figure GDA0004104679780000047
And a user auxiliary information feature vector->
Figure GDA0004104679780000048
Fusing to obtain a complete user feature vector H U Picking POI feature vector->
Figure GDA0004104679780000049
And POI assistance information feature vector &>
Figure GDA00041046797800000410
Fusing to obtain complete POI feature vector G V As shown in the following formula:
Figure GDA00041046797800000411
the neural matrix decomposition model is:
after obtaining the complete potential feature vectors of the user and the POI, the generalized matrix decomposition simulates the potential feature interaction between the user and the POI by using a linear kernel, the multi-layer perception machine learns the interaction function between the user and the POI by using a non-linear kernel, in order to effectively integrate the generalized matrix decomposition and the multi-layer perception machine, the generalized matrix decomposition and the multi-layer perception machine can be mutually enhanced, the complex interaction between the user and the POI is learned, the generalized matrix decomposition and the multi-layer perception machine are independently embedded, and the two models are combined by connecting a last hidden layer, wherein the formula is as follows:
Figure GDA0004104679780000051
W x 、b x and a x Respectively representing the weight matrix, the offset vector and the activation function of the x-th layer perceptron, h represents the edge weight of the output layer,
Figure GDA0004104679780000052
and &>
Figure GDA0004104679780000053
Respectively representing a generalized matrix decomposition and a complete user feature vector for a multi-level perceptron>
Figure GDA0004104679780000054
And
Figure GDA0004104679780000055
respectively representing the generalized matrix factorization and the complete POI eigenvector of the multi-layered perceptron, sigma represents a sigmoid activation function, phi represents the product of the elements of the vector MF-CAA And phi MLP-CAA Represents the predictor vector, <' > treated by the generalized matrix decomposition and the multi-layered perceptron, respectively>
Figure GDA0004104679780000056
Representing the predicted scores, the model combines the linear of the generalized matrix decomposition and the non-linear of the multi-layer perceptron to model the underlying structure between the user and the POI. POI inherent attribute information includes longitude, latitude, category and the likeHis information.
The user high activity location is the user's living area.
Compared with the prior art, the method has the advantages that the auxiliary information of the user and the POI is constructed from the sign-in data by using the position clustering and the TF-IDF algorithm, the potential representation is learned from the auxiliary information of the user and the POI by using the convolutional neural network, and the importance of the auxiliary information is distinguished by introducing an attention mechanism, so that the expression of the potential feature vectors of the user and the POI is enriched. And learning the interaction relation between the user and the potential feature vector of the POI by using a neural matrix decomposition model, and calculating the preference score of the user to the POI.
Drawings
FIG. 1 is a schematic diagram of a process framework of the present invention;
Detailed Description
The invention will now be further described with reference to the accompanying drawings.
Referring to fig. 1, the invention discloses a point of interest recommendation method considering the linear and nonlinear interaction of auxiliary information features, comprising the following steps:
as shown in fig. 1, firstly, the user and POI auxiliary information is constructed from the check-in data by using a position clustering and TF-IDF algorithm, potential representations are learned from the user and POI auxiliary information by using a convolutional neural network, and an attention mechanism is introduced to distinguish the importance of the auxiliary information, so that the expression of potential feature vectors of the user and the POI is enriched. And learning the interaction relation between the user and the potential feature vector of the POI by using a neural matrix decomposition model, and calculating the preference score of the user to the POI.
The method mainly comprises POI auxiliary information, user auxiliary information, auxiliary information construction based on a convolution attention mechanism and a neural matrix decomposition model;
the POI auxiliary information comprises a POI category, a POI popularity and a POI belonging area;
the POI category is inherent information;
the POI popularity is the popularity of the POI to the user, and the check-in data in the POI is evaluated through check-in data set, wherein the check-in data not only contains information about inherent attributes of the POI, but also contains the check-in times of the POI;
it is not sufficient to use the number of visitors to the POI directly for calculation of the popularity of the POI. Research indicates that the category information of the POI plays an important role in the point of interest recommendation process, because the category information of the POI can be used to characterize the POI.
POI popularity is calculated by TF-IDF algorithm based on check-in data, formula as follows:
Figure GDA0004104679780000071
sizeof(v i ) Denotes POIv i Number of check-ins of (Sv) i ) Is represented by the formula i Check-in times, sizeof (Cv), for all POIs of the same class i ) Is represented by the formula i The number of all POIs in the same category, n represents the number of the points of interest;
the method comprises the following steps that position information of an area to which a POI belongs is basic attributes of the POI and is also a key factor which must be considered by a POI recommendation algorithm, the POI is clustered according to the position information of the POI by utilizing a position clustering algorithm to obtain a proper area block, so that each POI is assigned with a position label of the corresponding area, and the processing steps are as follows:
s1, randomly selecting k POIs from a POI set as an initial clustering center;
s2, calculating the Euclidean distance rho from the residual POI to the clustering center, and putting the closest interest points into corresponding categories to form new categories, wherein the rho calculation method comprises the following steps:
Figure GDA0004104679780000072
lat i 、lat j respectively represent POIv i 、v j Latitude information of (lo) i 、lon j Respectively represent POIv i 、v j Longitude information of (a);
s3, taking the mean value of the longitude and latitude of all POI in the current cluster as a new central point, and updating the POI closest to the cluster center;
s4, until the target function is converged or the clustering center is not changed, otherwise, transferring to the step S2;
s5, outputting a POI clustering result;
the user assistance information includes a category of user preference, and a user high activity position (area) constitutes the user assistance information. Where the user-preferred categories refer to the POI categories that the user most frequently signs in. In the real world, the user's high activity location may be the user's living area. Thus, the POIs that the user most frequently signs in are used to infer a high activity location for the user.
The auxiliary information based on the convolution attention mechanism is constructed as follows:
carrying out one-hot coding on a user ID and an interest point ID, normalizing numerical variables in order to apply auxiliary attribute information of a user and a POI in a model, carrying out one-hot coding on category variables, and carrying out a user potential feature vector U u POI latent feature vector V v User auxiliary information U a And POI auxiliary information V a Are fed into the embedding layer, U, respectively u 、V v 、U a And V a The potential feature vector of (a) is calculated as follows:
Figure GDA0004104679780000081
f. r, t and l are representational functions;
because of the widespread use and good performance of convolutional neural networks, learning potential representations from user and POI assistance information using convolutional neural networks, which consist of convolutional and pooling layers that can extract deeper level features, is performed as follows:
Figure GDA0004104679780000082
* For convolution operators, w is the filter, b is the bias of w, g is the nonlinear activation function, posing is the pooling function (e.g., maximum pooling or average pooling)
The importance of user and POI assistance information is distinguished by adding an attention mechanism that uses a softmax (Q) function to compute a score as a probability distribution of assistance information output, and the output is summed with
Figure GDA0004104679780000083
Figure GDA0004104679780000084
Combined by element multiplication>
Figure GDA0004104679780000085
The final output of the attention mechanism is obtained, and the attention mechanism is shown as the following formula:
Figure GDA0004104679780000086
the softmax (Q) function is defined as:
Figure GDA0004104679780000087
u jc representing one of the attributes;
user feature vector
Figure GDA0004104679780000091
And a user auxiliary information feature vector->
Figure GDA0004104679780000092
Fusing to obtain complete user characteristic vector H U Pick up the POI feature vector->
Figure GDA0004104679780000093
And POI assistance information feature vector ≥>
Figure GDA0004104679780000094
Fusing to obtain complete POI characteristic vector G V As shown in the following formula: />
Figure GDA0004104679780000095
The neural matrix decomposition model is:
after obtaining the complete potential feature vectors of the user and the POI, the generalized matrix decomposition simulates the potential feature interaction between the user and the POI by using a linear kernel, the multi-layer perception machine learns the interaction function between the user and the POI by using a non-linear kernel, in order to effectively integrate the generalized matrix decomposition and the multi-layer perception machine, the generalized matrix decomposition and the multi-layer perception machine can be mutually enhanced, the complex interaction between the user and the POI is learned, the generalized matrix decomposition and the multi-layer perception machine are independently embedded, and the two models are combined by connecting a last hidden layer, wherein the formula is as follows:
Figure GDA0004104679780000096
W x 、b x and a x Respectively representing the weight matrix, the offset vector and the activation function of the x-th layer perceptron, h represents the edge weight of the output layer,
Figure GDA0004104679780000097
and &>
Figure GDA0004104679780000098
Represents the generalized matrix factorization and the complete user eigenvector, <' > or>
Figure GDA0004104679780000099
And
Figure GDA00041046797800000910
respectively representing the generalized matrix factorization and the complete POI eigenvector of the multi-layered perceptron, sigma represents a sigmoid activation function, phi represents the product of the elements of the vector MF-CAA And phi MLP-CAA Represents the predictor vector, <' > treated by the generalized matrix decomposition and the multi-layered perceptron, respectively>
Figure GDA00041046797800000911
Representing the predicted scores, the model combines the linear of the generalized matrix decomposition and the non-linear of the multi-layer perceptron to model the underlying structure between the user and the POI. The POI inherent attribute information includes longitude, latitude, category, and other information.
The user high activity location is the user's living area.
The above is only a preferred embodiment of the present invention, and is only used to help understand the method and the core idea of the present invention, and the protection scope of the present invention is not limited to the above embodiments, and all technical solutions belonging to the idea of the present invention belong to the protection scope of the present invention. It should be noted that modifications and embellishments within the scope of the invention may occur to those skilled in the art without departing from the principle of the invention, and are considered to be within the scope of the invention. The method integrally relieves the problems of data sparseness and implicit feedback in interest point recommendation, the auxiliary information of the user and the POI is mined and constructed from the user sign-in data by utilizing position clustering and a TF-IDF algorithm, the potential representation is learned from the auxiliary information of the user and the POI by utilizing a convolutional neural network, and the importance of the constructed auxiliary information of the user and the POI is distinguished by introducing an attention mechanism so as to relieve the problem of data sparseness; a neural matrix decomposition model is constructed, linear interaction of the user and the POI characteristic vectors is captured by utilizing generalized matrix decomposition, nonlinear interaction of the user and the POI characteristic vectors is captured by utilizing a multilayer perceptron, and the two parts are fused by connecting the last hidden layer so as to relieve the problem of implicit feedback.

Claims (3)

1. The interest point recommendation method considering the linear and nonlinear interaction of the auxiliary information features is characterized by mainly comprising POI auxiliary information, user auxiliary information, auxiliary information construction based on a convolution attention mechanism and a neural matrix decomposition model;
the POI auxiliary information comprises a POI category, a POI popularity and a POI belonging area;
the POI category is intrinsic information;
the POI popularity is the popularity of the POI to the user, and is evaluated through check-in data set data of the user checking in at the POI, wherein the check-in data not only contains information about inherent attributes of the POI, but also contains check-in times of the POI;
the POI popularity is calculated through a TF-IDF algorithm based on the check-in data, and the formula is as follows:
Figure FDA0004104679770000011
the sizeof (v) i ) Denotes POIv i Number of check-ins of (Sv) i ) Is represented by the formula i Check-in times, sizeof (Cv), for all POIs of the same class i ) Is represented by the formula i The number of all POIs in the same category, n represents the number of the points of interest;
the method comprises the following steps of clustering POI (point of interest) by utilizing a position clustering algorithm according to position information of the POI to obtain a proper region block, and enabling each POI to be allocated with a position label of a corresponding region, wherein the processing steps are as follows:
s1, randomly selecting k POIs from a POI set as an initial clustering center;
s2, calculating the Euclidean distance rho from the residual POI to the clustering center, and putting the closest interest points into corresponding categories to form new categories, wherein the rho calculation method comprises the following steps:
Figure FDA0004104679770000012
the lat i 、lat j Respectively represent POIv i 、v j Latitude information of (lo) i 、lon j Respectively represent POIv i 、v j Longitude information of (a);
s3, taking the mean value of longitude and latitude of all POI in the current cluster as a new central point, and updating the POI closest to the cluster center;
s4, until the target function is converged or the clustering center is not changed, otherwise, transferring to the step S2;
s5, outputting a POI clustering result;
the user auxiliary information comprises a user preferred category and a user high activity position, wherein the user preferred category refers to a POI category which is frequently signed in by a user;
the auxiliary information based on the convolution attention mechanism is constructed as follows:
carrying out unique hot coding on the user ID and the interest point ID, carrying out normalization processing on numerical variables in order to apply auxiliary attribute information of the user and the POI in a model, carrying out unique hot coding processing on category variables, and carrying out unique hot coding on the user potential feature vector U u POI latent feature vector V v User auxiliary information U a And POI auxiliary information V a Are fed into the embedding layer, U, respectively u 、V v 、U a And V a The potential feature vector of (a) is calculated as follows:
Figure FDA0004104679770000021
the f, r, t and l are representational functions;
learning potential representations from user and POI assistance information using a convolutional neural network, which consists of convolutional and pooling layers that can extract deeper level features, and is performed as follows:
Figure FDA0004104679770000022
the x is a convolution operator, w is a filter, b is the bias of w, g is a nonlinear activation function, and posing is a pooling function;
the importance of user and POI assistance information is distinguished by adding an attention mechanism that uses a softmax (Q) function to compute a score as a probability distribution of assistance information output, and the output is summed with
Figure FDA0004104679770000023
Figure FDA0004104679770000031
In combination, by element multiplication
Figure FDA0004104679770000032
The final output of the attention mechanism is obtained, and the attention mechanism is shown as the following formula:
Figure FDA0004104679770000033
the softmax (Q) function is defined as:
Figure FDA0004104679770000034
u jc representing one of the attributes;
feature vector of user
Figure FDA0004104679770000035
And a user auxiliary information feature vector->
Figure FDA0004104679770000036
Fusing to obtain a complete user feature vector H U Pick up the POI feature vector->
Figure FDA0004104679770000037
And POI assistance information feature vector ≥>
Figure FDA0004104679770000038
Fusing to obtain complete POI characteristic vector G V As shown in the following formula:
Figure FDA0004104679770000039
the neural matrix decomposition model is as follows:
after obtaining the complete potential feature vectors of the user and the POI, the generalized matrix decomposition simulates the potential feature interaction between the user and the POI by using a linear kernel, the multi-layer perception machine learns the interaction function between the user and the POI by using a non-linear kernel, in order to effectively integrate the generalized matrix decomposition and the multi-layer perception machine, the generalized matrix decomposition and the multi-layer perception machine can be mutually enhanced, the complex interaction between the user and the POI is learned, the generalized matrix decomposition and the multi-layer perception machine are independently embedded, and the two models are combined by connecting a last hidden layer, wherein the formula is as follows:
Figure FDA0004104679770000041
the W is x 、b x And a x Respectively representing the weight matrix, the offset vector and the activation function of the x-th layer perceptron, h representing the edge weight of the output layer,
Figure FDA0004104679770000042
and &>
Figure FDA0004104679770000043
Represents the generalized matrix factorization and the complete user eigenvector, <' > or>
Figure FDA0004104679770000044
And
Figure FDA0004104679770000045
respectively representing the generalized matrix factorization and the complete POI eigenvector of the multi-layered perceptron, sigma represents a sigmoid activation function, phi represents the product of the elements of the vector MF-CAA And phi MLP-CAA Represents the predictor vector, <' > treated by the generalized matrix decomposition and the multi-layered perceptron, respectively>
Figure FDA0004104679770000046
Representing predicted scores, the model combining the linear of generalized matrix factorization and the non-linear of multi-layer perceptrons to model the distance between the user and the POIA potential structure.
2. The method of claim 1, wherein the POI intrinsic attribute information includes longitude, latitude, category, and other information.
3. The method of point of interest recommendation accounting for ancillary information feature linear and non-linear interactions as claimed in claim 1 wherein the user high activity location is a user's living area.
CN202210968531.4A 2022-08-12 2022-08-12 Interest point recommendation method considering auxiliary information characteristic linear and nonlinear interaction Active CN115408621B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210968531.4A CN115408621B (en) 2022-08-12 2022-08-12 Interest point recommendation method considering auxiliary information characteristic linear and nonlinear interaction

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210968531.4A CN115408621B (en) 2022-08-12 2022-08-12 Interest point recommendation method considering auxiliary information characteristic linear and nonlinear interaction

Publications (2)

Publication Number Publication Date
CN115408621A CN115408621A (en) 2022-11-29
CN115408621B true CN115408621B (en) 2023-04-18

Family

ID=84158971

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210968531.4A Active CN115408621B (en) 2022-08-12 2022-08-12 Interest point recommendation method considering auxiliary information characteristic linear and nonlinear interaction

Country Status (1)

Country Link
CN (1) CN115408621B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117057809A (en) * 2023-10-12 2023-11-14 四川钱坤云智能科技有限公司 User behavior analysis method, system, electronic equipment and medium

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111061961A (en) * 2019-11-19 2020-04-24 江西财经大学 Multi-feature-fused matrix decomposition interest point recommendation method and implementation system thereof

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110008402B (en) * 2019-02-22 2021-09-03 苏州大学 Interest point recommendation method based on decentralized matrix decomposition of social network
CN109935326A (en) * 2019-02-28 2019-06-25 生活空间(沈阳)数据技术服务有限公司 A kind of probability of illness prediction meanss and storage medium
CN111680228B (en) * 2020-06-11 2022-03-18 浙江工商大学 Interest point recommendation method based on geographic position fusion and category popularity
CN111931067B (en) * 2020-09-14 2023-09-29 北京百度网讯科技有限公司 Point of interest recommendation method, device, equipment and medium
CN113536109B (en) * 2021-06-01 2022-10-21 重庆大学 Interest point recommendation method based on neural network and mobile context
CN114662015A (en) * 2022-02-25 2022-06-24 武汉大学 Interest point recommendation method and system based on deep reinforcement learning
CN114625971B (en) * 2022-05-12 2022-09-09 湖南工商大学 Interest point recommendation method and device based on user sign-in

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111061961A (en) * 2019-11-19 2020-04-24 江西财经大学 Multi-feature-fused matrix decomposition interest point recommendation method and implementation system thereof

Also Published As

Publication number Publication date
CN115408621A (en) 2022-11-29

Similar Documents

Publication Publication Date Title
CN110555112B (en) Interest point recommendation method based on user positive and negative preference learning
WO2021098372A1 (en) Method and apparatus for training graph neural network model for representing knowledge graph
CN111259133B (en) Personalized recommendation method integrating multiple information
CN110795571B (en) Cultural travel resource recommendation method based on deep learning and knowledge graph
CN111125530B (en) Information flow recommendation method based on multi-type feature deep learning
CN109871504B (en) Course recommendation system based on heterogeneous information network and deep learning
CN113807422B (en) Weighted graph convolutional neural network scoring prediction model integrating multi-feature information
CN110008377B (en) Method for recommending movies by using user attributes
CN114817663A (en) Service modeling and recommendation method based on class perception graph neural network
CN113407864A (en) Group recommendation method based on mixed attention network
CN115408621B (en) Interest point recommendation method considering auxiliary information characteristic linear and nonlinear interaction
CN111125540A (en) Recommendation method integrating knowledge graph representation learning and bias matrix decomposition
CN112784163A (en) RS 24-based personalized POIs recommendation method
CN117495511A (en) Commodity recommendation system and method based on contrast learning and community perception
CN112084418B (en) Microblog user community discovery method based on neighbor information and attribute network characterization learning
Wang et al. Online course recommendation algorithm based on multilevel fusion of user features and item features
CN113128526A (en) Image recognition method and device, electronic equipment and computer-readable storage medium
Tang et al. Accurately predicting quality of services in ioT via using self-attention representation and deep factorization machines
CN116910375A (en) Cross-domain recommendation method and system based on user preference diversity
Sun et al. Tcsa-net: a temporal-context-based self-attention network for next location prediction
CN116186385A (en) Service recommendation method based on decoupling characterization learning and graph neural network
Sun Music Individualization Recommendation System Based on Big Data Analysis
Christoforidis et al. Recommending points of interest in LBSNs using deep learning techniques
CN113962748A (en) Method for aligning users of heterogeneous e-commerce platform by using holomorphic information representation based on meta-path
Xingxing et al. Poi recommendation based on graph enhanced attention gnn

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