CN115269977A - Recommendation method for fusion knowledge and collaborative information based on graph neural network - Google Patents

Recommendation method for fusion knowledge and collaborative information based on graph neural network Download PDF

Info

Publication number
CN115269977A
CN115269977A CN202210870240.1A CN202210870240A CN115269977A CN 115269977 A CN115269977 A CN 115269977A CN 202210870240 A CN202210870240 A CN 202210870240A CN 115269977 A CN115269977 A CN 115269977A
Authority
CN
China
Prior art keywords
information
user
calculating
knowledge
neural network
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.)
Pending
Application number
CN202210870240.1A
Other languages
Chinese (zh)
Inventor
宫继兵
杨凯伦
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.)
Yanshan University
Original Assignee
Yanshan University
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 Yanshan University filed Critical Yanshan University
Priority to CN202210870240.1A priority Critical patent/CN115269977A/en
Publication of CN115269977A publication Critical patent/CN115269977A/en
Pending legal-status Critical Current

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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
    • G06F16/367—Ontology
    • 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
    • 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
    • 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)
  • Theoretical Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Animal Behavior & Ethology (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention discloses a recommendation method for fusion knowledge and collaborative information based on a graph neural network, which belongs to the technical field of recommendation algorithms and comprises the following steps: initializing a vector; acquiring an article embedding vector and a user embedding vector containing knowledge information through a knowledge information graph neural network on an information spreading layer; obtaining an article embedding vector and a user embedding vector containing the collaborative information through a collaborative information graph neural network on an information transmission layer; calculating at an information fusion layer to obtain fused user and article embedding vectors; calculating to obtain the prediction preference scores of all positive and negative samples; constructing a loss function of the recommendation method, and calculating loss; updating model parameters by using an Adam optimization algorithm and the set learning rate; calculating the prediction preference scores of all non-interactive articles of the user, and outputting the first K articles in sequence; and evaluating the recommendation method through the indexes. The method and the device make up the problems of sparse user behavior data and cold start, and further improve the recommendation precision.

Description

Recommendation method for fusion knowledge and collaborative information based on graph neural network
Technical Field
The invention relates to the technical field of recommendation algorithms, in particular to a recommendation method for fusion knowledge and collaborative information based on a graph neural network.
Background
With the rapid development of internet application, information presentation is increasing explosively, and the information search method cannot meet the requirements of users on information screening. Various information service platforms face a great challenge of screening huge and complex information according to different characteristics, preferences, scenes and other factors of users. The recommendation system takes an artificial intelligence algorithm as a kernel to mine the information of the user articles, and becomes an effective scheme for solving the challenge. The richness of recommendation scenes provides different challenges for a recommendation system, but the primary problem solved by the recommendation system is the accuracy of recommendation. Recommendation systems need to accurately predict the user's liking for items, information, etc. In the traditional recommendation system, the collaborative filtering algorithm achieves good effect of recommending users by collaboratively integrating other user behaviors. But in many application scenarios the user's behavioral data is sparse and there is little to no interaction data for new items to be interacted with by the user (the so-called cold start problem). For example, some inactive users in social media have only little interactive information, and new items updated in e-commerce platforms have only little purchasing and browsing information, and these problems become a major obstacle for improving recommendation precision by the collaborative filtering recommendation algorithm.
Therefore, a recommendation method is needed to solve the problems of data sparseness and cold start faced by a recommendation algorithm and improve recommendation accuracy.
Disclosure of Invention
The technical problem to be solved by the invention is to provide a recommendation method based on fusion knowledge and collaborative information of a graph neural network, auxiliary information is introduced into a recommendation system, and the recommendation method carries out recommendation by using similar auxiliary information, so that the problems of sparseness of user behavior data and cold start are solved, and the recommendation precision is further improved.
In order to solve the technical problems, the technical scheme adopted by the invention is as follows:
a recommendation method for fusion knowledge and collaborative information based on a graph neural network comprises the following steps:
s1, initializing a user embedding vector, an article embedding vector, a non-article entity embedding vector in a knowledge graph and a relation embedding vector in the knowledge graph;
s2, obtaining an article embedding vector and a user embedding vector containing knowledge information through a knowledge information graph neural network on an information spreading layer;
s3, obtaining an article embedding vector and a user embedding vector containing the collaborative information through a collaborative information graph neural network on an information transmission layer;
s4, calculating at an information fusion layer to obtain fused user and article embedding vectors;
s5, calculating to obtain the prediction preference scores of all positive and negative samples;
s6, constructing a loss function of the recommendation method, and calculating loss;
s7, updating model parameters by using an Adam optimization algorithm and the set learning rate delta;
s8, calculating the prediction preference scores of all the non-interacted articles of the user, and outputting the first K articles in sequence;
and S9, evaluating the recommendation method through two indexes of the recall rate and the normalized breaking and loss accumulation gain of the first K items.
The technical scheme of the invention is further improved as follows: s2 specifically comprises the following steps:
s21, obtaining an article embedding vector e containing knowledge information through a knowledge information graph neural network on an information spreading layeriThe formula is as follows:
Figure BDA0003751851850000021
wherein, l is a Hadamard product,
Figure BDA0003751851850000022
is a set of user neighbors of an item, evIs a non-article entity embedded vector in the knowledge graph, erIs a relation embedding vector in the knowledge graph; r represents a relationship;KG represents knowledge information; k represents the order of the graph neural network; i represents an item; v represents an entity;
Figure BDA0003751851850000023
an item representation vector representing the k +1 order graph neural network output containing knowledge information;
s22, obtaining a user embedded vector e containing knowledge information through a knowledge information graph neural network on an information spreading layeruThe formula is as follows:
Figure BDA0003751851850000031
wherein the content of the first and second substances,
Figure BDA0003751851850000032
is a user's set of item neighbors; u represents a user.
The technical scheme of the invention is further improved as follows: s3 specifically comprises the following steps:
s31, obtaining an article embedding vector containing the collaborative information through a collaborative information graph neural network on an information propagation layer, wherein the formula is as follows:
Figure BDA0003751851850000033
wherein, UI represents collaboration information;
Figure BDA0003751851850000034
representing a k +1 order item representation vector containing collaborative information;
s32, obtaining an article embedding vector containing the collaborative information through a collaborative information graph neural network on an information propagation layer, wherein the formula is as follows:
Figure BDA0003751851850000035
wherein the content of the first and second substances,
Figure BDA0003751851850000036
representing a k +1 order user representation vector containing the collaborative information.
The technical scheme of the invention is further improved as follows: s4 specifically comprises the following steps:
s41, calculating at an information fusion layer to obtain fused user embedded vectors of each layer;
s41 specifically includes the following steps:
s411, calculating the attention weight of the knowledge information, wherein the formula is as follows:
Figure BDA0003751851850000037
wherein α () represents an attention function;
Figure BDA0003751851850000038
a user representation vector representing the output of the k +1 order graph neural network; exp () stands for exponential function operation;
s412, calculating the attention weight of the collaborative information, wherein the formula is as follows:
Figure BDA0003751851850000041
s413, calculating the user embedded vectors of each layer after fusion, wherein the formula is as follows:
Figure BDA0003751851850000042
s42, calculating at the information fusion layer to obtain fusion-processed article embedding vectors of each layer;
s42 specifically comprises the following steps:
s421, calculating the attention weight of the knowledge information, wherein the formula is as follows:
Figure BDA0003751851850000043
wherein the content of the first and second substances,
Figure BDA0003751851850000044
an item representation vector representing the output of the k +1 order graph neural network.
S422, calculating the attention weight of the collaborative information, wherein the formula is as follows:
Figure BDA0003751851850000045
s423, calculating the embedded vectors of the fused articles in each layer, wherein the formula is as follows:
Figure BDA0003751851850000046
s43, calculating to obtain user embedded vectors added in each layer in the information fusion layer, wherein the formula is as follows:
Figure BDA0003751851850000047
wherein the content of the first and second substances,
Figure BDA0003751851850000048
representing the final user representation vector;
s44, calculating to obtain the added article embedding vector of each layer in the information fusion layer, wherein the formula is as follows:
Figure BDA0003751851850000049
wherein the content of the first and second substances,
Figure BDA0003751851850000051
representing the final item representation vector.
The technical scheme of the invention is further improved as follows: s5 specifically comprises the following steps:
s51, calculating the preference score of the positive sample, wherein the formula is as follows:
Figure BDA0003751851850000052
wherein, T represents a vector transposition operation;
Figure BDA0003751851850000053
representing the preference score of the user u output by the model to the item i;
s52, calculating the preference score of the negative sample;
s52 specifically includes the following steps:
s521, randomly selecting non-interactive articles of the user to form a negative sample;
s522, calculating the preference score of the negative sample, wherein the formula is as follows:
Figure BDA0003751851850000054
the technical scheme of the invention is further improved as follows: s6 specifically comprises the following steps:
s61, constructing and calculating the BPR loss, wherein the formula is as follows:
Figure BDA0003751851850000055
wherein, the first and the second end of the pipe are connected with each other,
Figure BDA0003751851850000056
represents BPR loss; o represents a sample set; σ is a sigmoid function; ln is a logarithmic function with e as the base;
s62, constructing and calculating the overall loss, wherein the formula is as follows:
Figure BDA0003751851850000057
wherein the content of the first and second substances,
Figure BDA0003751851850000058
is the final loss; lambda [ alpha ]1Is a regularization coefficient; Θ is all trainable parameters of the model.
The technical scheme of the invention is further improved as follows: s9 specifically comprises the following steps:
s91, calculating a recall rate Recall @ K, wherein the formula is as follows:
Figure BDA0003751851850000059
wherein, TP is recommended for the article, and the user is interacted at the same time, which indicates that the recommended article is exactly the favorite of the user; the FN is not recommended for the item, but the user interacts to indicate that the user likes are not identified;
s92, calculating a normalized breaking cumulative gain NDCG @ K, wherein the formula is as follows:
Figure BDA0003751851850000061
Figure BDA0003751851850000062
wherein r (i) refers to the relevance score of the item at position i in the recommendation list, and idcg @ k is an ideal discount accumulation gain, that is, the items in the recommendation list are all the items actually interacted by the user.
The technical scheme of the invention is further improved as follows: k takes the value of 20.
Due to the adoption of the technical scheme, the invention has the technical progress that: (invention points)
1. According to the invention, the knowledge information and the cooperation information in the recommendation data are respectively learned by setting the dual-channel graph neural network, so that the technical effects of separating the interference between the two information and generating more reasonable user article representation are achieved.
2. According to the invention, the information fusion mechanism based on the attention mechanism is arranged to fuse the user article representation containing the knowledge information and the collaborative information, so that the real interest and the article attribute of the user can be more accurately represented, and the technical effect of improving the recommendation accuracy is achieved.
Drawings
Fig. 1 is an overall frame diagram of the present invention.
Detailed Description
The invention is further described in detail below with reference to the drawings and examples:
as shown in fig. 1, a method for recommending knowledge and collaborative information based on a graph neural network includes the following steps:
s1, initializing a user embedding vector, an article embedding vector, a non-article entity embedding vector in a knowledge graph and a relation embedding vector in the knowledge graph;
s2, obtaining an article embedding vector and a user embedding vector which contain knowledge information through a knowledge information graph neural network on an information spreading layer;
the specific process of S2 is as follows:
s21, obtaining an article embedding vector e containing knowledge information through a knowledge information graph neural network on an information spreading layeriThe formula is as follows:
Figure BDA0003751851850000071
wherein, an is a Hadamard product,
Figure BDA0003751851850000072
is a set of user neighbors of an item, evIs a non-article entity embedded vector in the knowledge graph, erIs a relation embedding vector in the knowledge graph; r represents a relationship; KG represents knowledge information; k represents the order of the graph neural network; i represents an item; v represents an entity;
Figure BDA0003751851850000073
an item representation vector representing the k +1 order graph neural network output containing knowledge information;
s22, passing through the information transmission layerObtaining user embedded vector e containing knowledge information by identifying information graph neural networkuThe formula is as follows:
Figure BDA0003751851850000074
wherein the content of the first and second substances,
Figure BDA0003751851850000075
is a user's set of item neighbors; u represents a user;
s3, obtaining an article embedding vector and a user embedding vector containing the collaborative information through a collaborative information graph neural network on an information transmission layer;
the specific process of S3 is as follows:
s31, obtaining an article embedding vector containing the collaborative information through a collaborative information graph neural network on an information propagation layer, wherein the formula is as follows:
Figure BDA0003751851850000076
wherein, UI represents collaboration information;
Figure BDA0003751851850000077
representing a k +1 order item representation vector containing collaborative information;
s32, obtaining an article embedding vector containing the collaborative information through a collaborative information graph neural network on an information propagation layer, wherein the formula is as follows:
Figure BDA0003751851850000081
wherein, the first and the second end of the pipe are connected with each other,
Figure BDA0003751851850000082
representing a k +1 order user representation vector containing the collaborative information;
s4, calculating at an information fusion layer to obtain fused user and article embedding vectors;
the specific process of S4 is as follows:
s41, calculating at an information fusion layer to obtain fused user embedded vectors of each layer;
the specific process of S41 is as follows:
s411, calculating the attention weight of the knowledge information, wherein the formula is as follows:
Figure BDA0003751851850000083
wherein α () represents an attention function;
Figure BDA0003751851850000084
a user representation vector representing the output of the k +1 order graph neural network; exp () represents an exponential function operation;
s412, calculating the attention weight of the collaborative information, wherein the formula is as follows:
Figure BDA0003751851850000085
s413, calculating the user embedded vectors of each layer after fusion, wherein the formula is as follows:
Figure BDA0003751851850000086
s42, calculating at the information fusion layer to obtain fusion-processed article embedding vectors of each layer;
the specific process of S42 is as follows:
s421, calculating the attention weight of the knowledge information, wherein the formula is as follows:
Figure BDA0003751851850000087
wherein the content of the first and second substances,
Figure BDA0003751851850000088
representing the k +1 order graph neural network outputThe output items represent vectors.
S422, calculating the attention weight of the collaborative information, wherein the formula is as follows:
Figure BDA0003751851850000091
s423, calculating the embedded vectors of the fused articles in each layer, wherein the formula is as follows:
Figure BDA0003751851850000092
s43, calculating to obtain user embedded vectors added in each layer in the information fusion layer, wherein the formula is as follows:
Figure BDA0003751851850000093
wherein the content of the first and second substances,
Figure BDA0003751851850000094
representing the final user representation vector;
s44, calculating to obtain the added article embedding vector of each layer in the information fusion layer, wherein the formula is as follows:
Figure BDA0003751851850000095
wherein the content of the first and second substances,
Figure BDA0003751851850000096
representing a final item representation vector;
s5, calculating to obtain the prediction preference scores of all positive and negative samples;
the specific process of S5 is as follows:
s51, calculating the preference score of the positive sample, wherein the formula is as follows:
Figure BDA0003751851850000097
wherein, T represents a vector transposition operation;
Figure BDA0003751851850000098
representing the preference score of the user u output by the model to the item i;
s52, calculating the preference score of the negative sample;
the specific process of S52 is as follows:
s521, randomly selecting non-interactive articles of the user to form a negative sample;
s522, calculating the preference score of the negative sample, wherein the formula is as follows:
Figure BDA0003751851850000099
s6, constructing a loss function of the recommendation method, and calculating loss;
the specific process of S6 is as follows:
s61, constructing and calculating the BPR loss, wherein the formula is as follows:
Figure BDA0003751851850000101
wherein the content of the first and second substances,
Figure BDA0003751851850000102
represents BPR loss; o represents a sample set; σ is a sigmoid function; ln is a logarithmic function with e as the base;
s62, constructing and calculating the overall loss, wherein the formula is as follows:
Figure BDA0003751851850000103
wherein the content of the first and second substances,
Figure BDA0003751851850000104
is the final loss; lambda [ alpha ]1Is a regularization coefficient; theta is all of the modelsTraining parameters;
s7, updating model parameters by using an Adam optimization algorithm and the set learning rate delta;
s8, calculating the prediction preference scores of all the non-interacted articles of the user, and outputting the first K articles in sequence;
s9, evaluating the recommendation method through two indexes of the recall rate and the normalized breaking and loss accumulation gain of the first K items;
the specific process of S9 is as follows:
s91, calculating a recall rate Recall @ K, wherein the formula is as follows:
Figure BDA0003751851850000105
wherein, TP is recommended for the article, and the user also interacts at the same time, which indicates that the recommended is exactly the favorite of the user; the FN is not recommended for the item, but the user interacts to indicate that the user likes are not identified;
s92, calculating a normalized breaking cumulative gain NDCG @ K, wherein the formula is as follows:
Figure BDA0003751851850000106
Figure BDA0003751851850000107
wherein r (i) refers to the relevance score of the item at position i in the recommendation list, and idcg @ k is an ideal discount accumulation gain, that is, the items in the recommendation list are all the items actually interacted by the user.
Typically K takes the value 20.
In the context of figure 1 of the drawings,
Figure BDA0003751851850000111
representing the vector inner product.
In conclusion, the method and the device make up the problems of sparse user behavior data and cold start, and further improve the recommendation precision.

Claims (8)

1. A recommendation method for fusion knowledge and collaborative information based on a graph neural network is characterized in that: the method comprises the following steps:
s1, initializing a user embedding vector, an article embedding vector, a non-article entity embedding vector in a knowledge graph and a relation embedding vector in the knowledge graph;
s2, obtaining an article embedding vector and a user embedding vector containing knowledge information through a knowledge information graph neural network on an information spreading layer;
s3, obtaining an article embedding vector and a user embedding vector containing the collaborative information through a collaborative information graph neural network on an information transmission layer;
s4, calculating at an information fusion layer to obtain fusion user and article embedding vectors;
s5, calculating to obtain the prediction preference scores of all positive and negative samples;
s6, constructing a loss function of the recommendation method, and calculating loss;
s7, updating model parameters by using an Adam optimization algorithm and the set learning rate delta;
s8, calculating the prediction preference scores of all the non-interacted articles of the user, and outputting the first K articles in sequence;
and S9, evaluating the recommendation method through two indexes of the recall rate and the normalized breaking and loss accumulation gain of the first K items.
2. The method for recommending knowledge and collaborative information based on neural network of figure as claimed in claim 1, wherein: s2 specifically comprises the following steps:
s21, obtaining an article embedding vector e containing knowledge information through a knowledge information graph neural network on an information spreading layeriThe formula is as follows:
Figure FDA0003751851840000011
therein, anIs a product of the Hadamard multiplication,
Figure FDA0003751851840000012
is a set of user neighbors of an item, evIs a non-article entity embedded vector in the knowledge graph, erIs a relation embedding vector in the knowledge graph; r represents a relationship; KG represents knowledge information; k represents the order of the graph neural network; i represents an item; v represents an entity;
Figure FDA0003751851840000013
an item representation vector representing the k +1 order graph neural network output containing knowledge information;
s22, obtaining a user embedded vector e containing knowledge information through a knowledge information graph neural network on an information spreading layeruThe formula is as follows:
Figure FDA0003751851840000021
wherein the content of the first and second substances,
Figure FDA0003751851840000022
is a user's set of item neighbors; u represents a user.
3. The method for recommending knowledge and collaborative information based on neural network of figure as claimed in claim 1, wherein: s3 specifically comprises the following steps:
s31, obtaining an article embedding vector containing the collaborative information through a collaborative information graph neural network on an information propagation layer, wherein the formula is as follows:
Figure FDA0003751851840000023
wherein, UI represents collaboration information;
Figure FDA0003751851840000024
representing a k +1 order item representation vector containing collaborative information;
s32, obtaining an article embedding vector containing the collaborative information through a collaborative information graph neural network on an information propagation layer, wherein the formula is as follows:
Figure FDA0003751851840000025
wherein the content of the first and second substances,
Figure FDA0003751851840000026
representing a k +1 order user representation vector containing the collaborative information.
4. The method for recommending knowledge and collaborative information based on neural network of figure as claimed in claim 1, wherein: s4 specifically comprises the following steps:
s41, calculating at an information fusion layer to obtain fused user embedded vectors of each layer;
s41 specifically includes the following steps:
s411, calculating the attention weight of the knowledge information, wherein the formula is as follows:
Figure FDA0003751851840000031
wherein α () represents an attention function;
Figure FDA0003751851840000032
a user representation vector representing the output of the k +1 order graph neural network; exp () represents an exponential function operation;
s412, calculating the attention weight of the collaborative information, wherein the formula is as follows:
Figure FDA0003751851840000033
s413, calculating the user embedded vectors of each layer after fusion, wherein the formula is as follows:
Figure FDA0003751851840000034
s42, calculating at the information fusion layer to obtain fusion-processed article embedding vectors of each layer;
s42 specifically comprises the following steps:
s421, calculating the attention weight of the knowledge information, wherein the formula is as follows:
Figure FDA0003751851840000035
wherein the content of the first and second substances,
Figure FDA0003751851840000036
an item representation vector representing the k +1 order graph neural network output;
s422, calculating the attention weight of the collaborative information, wherein the formula is as follows:
Figure FDA0003751851840000037
s423, calculating the embedded vectors of the fused articles in each layer, wherein the formula is as follows:
Figure FDA0003751851840000038
s43, calculating to obtain user embedded vectors added in each layer in the information fusion layer, wherein the formula is as follows:
Figure FDA0003751851840000039
wherein the content of the first and second substances,
Figure FDA0003751851840000041
representing the final user representation vector;
s44, calculating to obtain the added article embedding vector of each layer in the information fusion layer, wherein the formula is as follows:
Figure FDA0003751851840000042
wherein, the first and the second end of the pipe are connected with each other,
Figure FDA0003751851840000043
representing the final item representation vector.
5. The method for recommending knowledge and collaborative information based on neural network of figure as claimed in claim 1, wherein: s5 specifically comprises the following steps:
s51, calculating the preference score of the positive sample, wherein the formula is as follows:
Figure FDA0003751851840000044
wherein, T represents vector transposition operation;
Figure FDA0003751851840000045
representing the preference score of the user u output by the model to the item i;
s52, calculating the preference score of the negative sample;
s52 specifically includes the following steps:
s521, randomly selecting non-interactive articles of the user to form a negative sample;
s522, calculating the preference score of the negative sample, wherein the formula is as follows:
Figure FDA0003751851840000046
6. the method for recommending knowledge and collaborative information based on neural network of figure as claimed in claim 1, wherein: s6 specifically comprises the following steps:
s61, constructing and calculating the BPR loss, wherein the formula is as follows:
Figure FDA0003751851840000047
wherein the content of the first and second substances,
Figure FDA0003751851840000048
represents BPR loss; o represents a sample set; σ is a sigmoid function; ln is a logarithmic function with e as the base;
s62, constructing and calculating the overall loss, wherein the formula is as follows:
Figure FDA0003751851840000049
wherein the content of the first and second substances,
Figure FDA0003751851840000051
is the final loss; lambda [ alpha ]1Is a regularization coefficient; Θ is all trainable parameters of the model.
7. The method for recommending knowledge and collaborative information based on neural network of figure as claimed in claim 1, wherein: s9 specifically comprises the following steps:
s91, calculating a recall rate Recall @ K, wherein the formula is as follows:
Figure FDA0003751851840000052
wherein, TP is recommended for the article, and the user is interacted at the same time, which indicates that the recommended article is exactly the favorite of the user; the FN is not recommended for the item, but the user interacts to indicate that the user likes are not identified;
s92, calculating a normalized breaking cumulative gain NDCG @ K, wherein the formula is as follows:
Figure FDA0003751851840000053
Figure FDA0003751851840000054
wherein r (i) refers to the relevance score of the item at position i in the recommendation list, and idcg @ k is an ideal discount accumulation gain, that is, the items in the recommendation list are all the items actually interacted by the user.
8. The method for recommending knowledge and collaboration information based on neural network of figure according to any one of claims 1 or 7, wherein: k takes the value of 20.
CN202210870240.1A 2022-07-18 2022-07-18 Recommendation method for fusion knowledge and collaborative information based on graph neural network Pending CN115269977A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210870240.1A CN115269977A (en) 2022-07-18 2022-07-18 Recommendation method for fusion knowledge and collaborative information based on graph neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210870240.1A CN115269977A (en) 2022-07-18 2022-07-18 Recommendation method for fusion knowledge and collaborative information based on graph neural network

Publications (1)

Publication Number Publication Date
CN115269977A true CN115269977A (en) 2022-11-01

Family

ID=83769615

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210870240.1A Pending CN115269977A (en) 2022-07-18 2022-07-18 Recommendation method for fusion knowledge and collaborative information based on graph neural network

Country Status (1)

Country Link
CN (1) CN115269977A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116342228A (en) * 2023-05-18 2023-06-27 云筑信息科技(成都)有限公司 Related recommendation method based on directed graph neural network
CN117171449A (en) * 2023-09-21 2023-12-05 西南石油大学 Recommendation method based on graph neural network
CN117708421A (en) * 2023-12-16 2024-03-15 辽宁工业大学 Dynamic recommendation method and system based on modularized neural network

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116342228A (en) * 2023-05-18 2023-06-27 云筑信息科技(成都)有限公司 Related recommendation method based on directed graph neural network
CN116342228B (en) * 2023-05-18 2023-10-20 云筑信息科技(成都)有限公司 Related recommendation method based on directed graph neural network
CN117171449A (en) * 2023-09-21 2023-12-05 西南石油大学 Recommendation method based on graph neural network
CN117171449B (en) * 2023-09-21 2024-03-19 西南石油大学 Recommendation method based on graph neural network
CN117708421A (en) * 2023-12-16 2024-03-15 辽宁工业大学 Dynamic recommendation method and system based on modularized neural network

Similar Documents

Publication Publication Date Title
CN111339415B (en) Click rate prediction method and device based on multi-interactive attention network
CN111241311B (en) Media information recommendation method and device, electronic equipment and storage medium
CN111460130B (en) Information recommendation method, device, equipment and readable storage medium
TWI754033B (en) Generating document for a point of interest
CN108460619B (en) Method for providing collaborative recommendation model fusing explicit and implicit feedback
EP4181026A1 (en) Recommendation model training method and apparatus, recommendation method and apparatus, and computer-readable medium
CN110879864B (en) Context recommendation method based on graph neural network and attention mechanism
CN115269977A (en) Recommendation method for fusion knowledge and collaborative information based on graph neural network
CN111310063A (en) Neural network-based article recommendation method for memory perception gated factorization machine
CN112115377A (en) Graph neural network link prediction recommendation method based on social relationship
CN112884552B (en) Lightweight multi-mode recommendation method based on generation countermeasure and knowledge distillation
CN111949886B (en) Sample data generation method and related device for information recommendation
CN109902229B (en) Comment-based interpretable recommendation method
CN108595533A (en) A kind of item recommendation method, storage medium and server based on collaborative filtering
CN112819575A (en) Session recommendation method considering repeated purchasing behavior
Hou et al. A deep reinforcement learning real-time recommendation model based on long and short-term preference
CN112948696A (en) Cross-domain medical care equipment recommendation method and system with privacy protection function
Li et al. Intelligent Recommendation Algorithm of Consumer Electronics Products with Graph Embedding and Multi-Head Self-Attention in IoE
CN115391555A (en) User-perceived knowledge map recommendation system and method
CN115809339A (en) Cross-domain recommendation method, system, device and storage medium
Tang et al. Service recommendation based on dynamic user portrait: an integrated approach
CN111104552A (en) Method for predicting movie scoring category based on movie structural information and brief introduction
CN115545834B (en) Personalized service recommendation method based on graphic neural network and metadata
WO2024061073A1 (en) Multimedia information generation method and apparatus, and computer-readable storage medium
Alshammari A Restaurant Recommendation Engine Using Feature-based Explainable Matrix Factorization

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