CN117520665B - Social recommendation method based on generation of countermeasure network - Google Patents
Social recommendation method based on generation of countermeasure network Download PDFInfo
- Publication number
- CN117520665B CN117520665B CN202410016462.6A CN202410016462A CN117520665B CN 117520665 B CN117520665 B CN 117520665B CN 202410016462 A CN202410016462 A CN 202410016462A CN 117520665 B CN117520665 B CN 117520665B
- Authority
- CN
- China
- Prior art keywords
- vector
- user
- representing
- generator
- embedded vector
- 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
- 238000000034 method Methods 0.000 title claims abstract description 23
- 230000006870 function Effects 0.000 claims abstract description 37
- 238000012549 training Methods 0.000 claims abstract description 20
- 238000013528 artificial neural network Methods 0.000 claims abstract description 13
- 230000008569 process Effects 0.000 claims abstract description 8
- 230000003993 interaction Effects 0.000 claims description 7
- 238000006243 chemical reaction Methods 0.000 claims description 4
- 239000011159 matrix material Substances 0.000 claims description 4
- 238000005295 random walk Methods 0.000 claims description 4
- 230000004069 differentiation Effects 0.000 claims description 2
- 230000001902 propagating effect Effects 0.000 claims description 2
- 230000008485 antagonism Effects 0.000 claims 1
- 230000000630 rising effect Effects 0.000 claims 1
- 239000000203 mixture Substances 0.000 description 4
- 239000002131 composite material Substances 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 230000002452 interceptive effect Effects 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000001174 ascending effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000003068 static effect Effects 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
- G06F16/9536—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
- G06N3/042—Knowledge-based neural networks; Logical representations of neural 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
- G06N3/094—Adversarial learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Databases & Information Systems (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)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a social recommendation method based on generating an countermeasure network, which is characterized in that a complex heterogeneous graph is decomposed into two simple homogeneous subgraphs through a meta-path, the structure and semantic information contained in a user group without social labels can be fully explored and utilized, so that the quality of social recommendation results is improved, then the joint embedding vector is obtained through pre-training of a graph neural network, then the joint embedding vector is converted into a user independent heat vector and a project independent heat vector through a straight-through Gumbil-Softmax, the fact that the social recommendation can be trained in the counter propagation process can be ensured, after loss functions of a generator and a discriminator in the countermeasure network are well determined, the user independent heat vector, the project independent heat vector, the user embedding vector and the project embedding vector are respectively input into the generator and the discriminator, and when balance is realized between the generator and the discriminator in the countermeasure network is generated, a high-quality recommendation result is obtained.
Description
Technical Field
The invention relates to the technical field of data processing, in particular to a social recommendation method based on generation of an countermeasure network.
Background
The recommendation system is widely focused and studied as one of important tools for acquiring information in the big data age. In a social recommendation system, an isograph is a core data structure, and the isograph can properly express social relations among users and interactive relations among users and projects. The social and interactive information of the user group with the social relation in the different composition can be fully mined, so that the accuracy and precision of the social recommendation system result can be remarkably improved.
However, users with clear social relations in the real world only occupy a small number, and a large number of user groups have no clear social labels, but interact with the same items, so that the social recommendation system based on different composition cannot feed back the information, and the recommendation result is poor in quality.
Disclosure of Invention
Therefore, the embodiment of the invention provides a social recommendation method based on generating an countermeasure network so as to improve the quality of recommendation results.
According to one embodiment of the invention, the social recommendation method based on the generation of the countermeasure network comprises the following steps:
step 1, splitting a heterogeneous graph based on social contact between users and interaction between the users and items into a first homogeneous sub graph and a second homogeneous sub graph through a first meta path and a second meta path, wherein the first homogeneous sub graph is a user relationship graph for describing social link relationships among the users, and the second homogeneous sub graph is a user relationship graph for describing no social link relationship among the users but sharing the same item;
step 2, pre-training is carried out through a graph neural network based on the heterogram to obtain a user embedded vector and a project embedded vector, pre-training is carried out through the graph neural network based on the first homogeneous subgraph to obtain a first uniform path embedded vector, pre-training is carried out through the graph neural network based on the second homogeneous subgraph to obtain a second uniform path embedded vector, and then a joint embedded vector is obtained based on the first uniform path embedded vector and the second uniform path embedded vector;
step 3, in a generator for generating a reactance network, the joint embedded vector is converted into a user independent heat vector and a project independent heat vector through a straight-through Gumbel-Softmax;
step 4, determining to generate a loss function of a generator and a loss function of a discriminator in the countermeasure network;
and 5, respectively inputting the user independent heat vector, the project independent heat vector, the user embedded vector and the project embedded vector into a generator and a discriminator, generating an objective function of the countermeasure network based on the loss function of the generator and the loss function of the discriminator by iterative updating, and finally outputting a final recommended project list through the generation of the countermeasure network.
According to the social recommendation method based on the generation of the countermeasure network, a complex heterogeneous graph is decomposed into two simple homogeneous subgraphs through a meta-path, the two simple homogeneous subgraphs are respectively a user relationship graph used for describing social link relationships among users, and a user relationship graph used for describing no social link relationship among users but sharing the same item, so that structure and semantic information contained in a user group without social labels can be fully explored and utilized, the result quality of social recommendation is improved, each homogeneous subgraph is associated with the structure and semantic information of a specific user social connection, then a joint embedding vector is obtained through pre-training of the graph neural network, the joint embedding vector is converted into a user independent heat vector and a project independent heat vector through a straight-through Gumbel-Softmax, the fact that the social recommendation is trainable in a back propagation process can be ensured, after loss functions of a generator and a discriminator in the countermeasure network are determined, the user independent heat vector, the project independent heat vector, the user embedding vector and the project embedding vector are respectively input into the generator and the discriminator, and when balance between the generator and the discriminator in the countermeasure network is generated, the result is high in quality is obtained.
Drawings
The foregoing and/or additional aspects and advantages of embodiments of the invention will be apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings, in which:
FIG. 1 is a flow diagram of a social recommendation method based on generating an countermeasure network in accordance with an embodiment of the invention;
fig. 2 is a schematic diagram of an exemplary splitting of an isograph into a first homograph and a second homograph.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Referring to fig. 1, an embodiment of the present invention provides a social recommendation method based on generating an countermeasure network, where the method includes steps 1 to 5:
step 1, splitting a heterogeneous graph based on social contact between users and interaction between the users and the items into a first homogeneous sub-graph and a second homogeneous sub-graph through a first meta-path and a second meta-path, wherein the first homogeneous sub-graph is a user relationship graph for describing social link relationships among the users, and the second homogeneous sub-graph is a user relationship graph for describing that the users do not have social link relationships but share the same items.
In this embodiment, a different composition is constructed in the real world of user-to-user social connection and user-to-item interaction, please refer to fig. 2, wherein different vertices in the different composition represent users or items, and different edges represent social connection and interaction information between users and between users and items.
The meta-path is a pattern-guided path in the iso-graph that represents a composite relationship between two entities. According to the concept of meta-paths, the present embodiment disassembles the heterogeneous graph in fig. 2 into a first homogeneous sub-graph and a second homogeneous sub-graph. Specifically, a first homography built around a target user u through UTU (i.e., user-social-user) meta-paths is used to describe the social link relationship of user u. A second homogeneous subgraph is constructed through the UIU (i.e., user-no-social-user) meta-path to describe a user relationship graph that has no social links with the target user u but shares the same item. The two homogeneous subgraphs provide rich social network structural and semantic information from two different angles.
And 2, pre-training through a graph neural network based on the heterogram to obtain a user embedded vector and a project embedded vector, pre-training through the graph neural network based on the first homogeneous subgraph to obtain a first uniform path embedded vector, pre-training through the graph neural network based on the second homogeneous subgraph to obtain a second uniform path embedded vector, and obtaining a joint embedded vector based on the first uniform path embedded vector and the second uniform path embedded vector.
The invention adopts the graph neural network as a pre-training process, takes the iso-graph and two homography as inputs, embeds the iso-graph and the homography into a low-dimensional vector space, and the structural information and the relation between nodes in the graph are reserved in the space.
The step 2 specifically includes:
random walk sequences with the same step sizeRespectively applied on the heterograph, the first homozygotic and the second homozygotic, wherein +.>，/>Is the step of walk, +.>、/>、/>Respectively represent 1 st, 2 nd and +.>Nodes in the process of individual wander;
for each sequence in the heterograms, the first homograph and the second homograph, taking each user and item as a central node, taking nodes within the length and size range of the random walk sequence as positive samples, and randomly selecting some nodes from the nodes in the heterograms, the first homograph and the second homograph as negative samples outside the positive samples;
training each sequence in the heterogram, the first homogeneous subgraph and the second homogeneous subgraph by using a Skip model so as to obtain a user embedded vector, a project embedded vector, a first meta-path embedded vector and a second meta-path embedded vector, wherein the aim of the Skip model (Skip-Gram model) is to maximize a log-likelihood function;
and fusing the first element path embedded vector and the second element path embedded vector through Hadamard products to obtain a joint embedded vector.
And 3, in a generator for generating the reactance network, the joint embedded vector is converted into a user independent heat vector and a project independent heat vector through a straight-through Gumbel-Softmax.
The step 3 specifically includes:
the joint embedding vector is firstly converted into probability distribution, and the expression is as follows:
；
wherein,is the i-th element in the joint embedded vector, < >>Is the j-th element in the joint embedded vector, is->Is the total number of elements in the joint embedded vector, +.>Representing the conversion vector by softmax function +.>Values in the probability distribution obtained;
then isAdding noise, normalizing to obtain a probability distribution vector +.>The expression is as follows:
wherein,is added to->Gumbel noise of (2); />Is a temperature parameter for controlling the degree to which the output user variable approaches the independent heat vector, in particular, < + >>The closer to 0, the closer to the independent heat vector the output; />Representing the conversion vector by softmax function +.>Values in the probability distribution obtained, +.>Is added to->Gumbel noise of (2);
then forward propagating, training to generate generators in the countermeasure network, definingThe index of the largest element in (a) isConstructing a user independent heat vector->The expression is:
；
then back-propagating, updating the generator in the generation countermeasure network, usingAs a gradient, this operation ensures continuity of gradient propagation, specifically satisfying the following conditional expression:
；
wherein,representing differentiation,/->Representing a loss function of the generator;
obtaining the user independent heat vectorThen, obtaining project independent heat vector according to the following steps, firstly establishing interaction matrix of user and project from the original data set>A trainable matrix is then created>Thereby obtaining the project independent heat vector +.>The expression is as follows:
wherein,representing a straight-through Gumbi-Softmax function, T representing a transpose operation, ++>Representing the hadamard product.
And 4, determining to generate a loss function of a generator and a loss function of a discriminator in the countermeasure network.
In step 4, the loss function of the generatorThe following conditional expression is satisfied:
；
wherein,is a mathematical desired operation, +.>Representing Sigmoid function->Representing the score of user u for item v,representing user u's one-hot vector for the project>Score of->Representing a collection of synthetic items, +.>Representing the conditional probability distribution of the synthesized item.
Capturing the generator relative to/>And the overall expected loss of its corresponding probability, minimizing +.>The resulting item may be considered similar to the item that the user has interacted with. Thus, the generator is used to forge a genuine item for the arbiter to learn.
Loss function of discriminatorThe following conditional expression is satisfied:
；
wherein,learnable parameters representing a discriminator, +.>Representing the score of user u for item j, +.>Is a regularization term. By minimizing +.>Personalized recommendations may be generated for each user.
And 5, respectively inputting the user independent heat vector, the project independent heat vector, the user embedded vector and the project embedded vector into a generator and a discriminator, generating an objective function of the countermeasure network based on the loss function of the generator and the loss function of the discriminator by iterative updating, and finally outputting a final recommended project list through the generation of the countermeasure network.
The invention iteratively updates the generator and the arbiter until their loss gap is acceptable or the number of iterations reaches a maximum. At this time, the discriminator outputs the final recommended item list.
The iterative update generator and arbiter reach an equilibrium state where the generator's ability to produce synthetic items that resemble real items and the arbiter's ability to distinguish between synthetic items and real items are balanced. In the equilibrium state, the ranking output of the arbiter serves as the final recommended output of the present invention.
The invention adopts an alternate iterative mode to optimize the generator and the discriminator, the optimization target for generating the countermeasure network is formulated into a very small and very large game, and in particular, in the step 5, the objective function of the countermeasure network is generatedThe expression of (2) is:
；
wherein,representing the discriminator->A representation generator.
In step 5, parameters of a generator and a discriminator for generating an countermeasure network are updated by random gradient ascending and descending respectively, and the expression is:
；
wherein,、/>the learnable parameters of the generators corresponding to the time t and the time t+1 are respectively represented by +.>Indicates learning rate (I/O)>Representing the gradient of the parameters in the loss function of the generator,/->、/>The learnable parameters of the discriminators corresponding to the time t and the time t+1 are respectively represented, and the +.>Representing the gradient of the parameters in the loss function of the arbiter,score representing user u interacting with item v in generator, < >>Representing user u in the generator together with the item one-hot vector +.>Score of interaction->Score indicating user u interacted with item v in the arbiter, ++>Representing user u in the arbiter with the item one-hot vector +.>Score of interaction->Representing the score of user u interacting with item j in the arbiter.
By iteration of the update operation, a loop is generated that counter the network into a self-evolving loop. If the quality of the recommended results cannot be significantly improved, the counter-propagating gradient will guide the generator to produce a higher quality user. At the same time, the higher the user quality generated, the greater the optimum pressure on the arbiter. With constant iteration of the countermeasure learning, a balance will occur between the generator and the arbiter. Specifically, the generator creates a composite user and composite item such that the arbiter no longer has room for further optimization, the item at this time being the best choice for the current user.
The training process of the present invention is not static, but rather a dynamic and adaptive evaluation mechanism. This mechanism is organically combined with model training, rather than an isolated training process. Through the steps, the method and the device have stronger robustness and flexibility.
In summary, according to the social recommendation method based on generating the countermeasure network in the embodiment of the invention, a complex heterogeneous graph is decomposed into two simple homogeneous subgraphs through a meta path, which are respectively user relation graphs for describing social link relations among users and user relation graphs for describing no social link relation among users but sharing the same item, so that the structure and semantic information contained in a user group without social labels can be fully explored and utilized, the result quality of social recommendation is improved, each homogeneous subgraph is associated with the structure and semantic information of a specific user social, then a joint embedding vector is obtained through pre-training of the graph neural network, and the joint embedding vector is converted into a user unique heat vector and a project unique heat vector through a straight-through gummel-Softmax, so that the social recommendation can be trained in a reverse propagation process, after a loss function of a generator and a discriminator in the countermeasure network is determined, the user unique heat vector, the project unique heat vector, the user vector and the project embedding vector are respectively input into the generator and the discriminator, and when the balance between the generator and the discriminator in the countermeasure network is realized, namely, a high quality result is obtained.
In the description of the present specification, a description referring to terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiments or examples. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
While embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that: many changes, modifications, substitutions and variations may be made to the embodiments without departing from the spirit and principles of the invention, the scope of which is defined by the claims and their equivalents.
Claims (2)
1. A method of generating social recommendations based on an antagonism network, the method comprising:
step 1, splitting a heterogeneous graph based on social contact between users and interaction between the users and items into a first homogeneous sub graph and a second homogeneous sub graph through a first meta path and a second meta path, wherein the first homogeneous sub graph is a user relationship graph for describing social link relationships among the users, and the second homogeneous sub graph is a user relationship graph for describing no social link relationship among the users but sharing the same item;
step 2, pre-training is carried out through a graph neural network based on the heterogram to obtain a user embedded vector and a project embedded vector, pre-training is carried out through the graph neural network based on the first homogeneous subgraph to obtain a first uniform path embedded vector, pre-training is carried out through the graph neural network based on the second homogeneous subgraph to obtain a second uniform path embedded vector, and then a joint embedded vector is obtained based on the first uniform path embedded vector and the second uniform path embedded vector;
step 3, in a generator for generating a reactance network, the joint embedded vector is converted into a user independent heat vector and a project independent heat vector through a straight-through Gumbel-Softmax;
step 4, determining to generate a loss function of a generator and a loss function of a discriminator in the countermeasure network;
step 5, inputting the user independent heat vector, the project independent heat vector, the user embedded vector and the project embedded vector into a generator and a discriminator respectively, generating an objective function of the countermeasure network based on the loss function of the generator and the loss function of the discriminator by iterative updating, and finally outputting a final recommended project list through the generation of the countermeasure network;
the step 2 specifically comprises the following steps:
random walk sequences with the same step sizeRespectively applied on the heterograph, the first homozygotic and the second homozygotic, wherein +.>，/>Is the step of walk, +.>、/>、/>Respectively represent 1 st, 2 nd and +.>Nodes in the process of individual wander;
for each sequence in the heterograms, the first homograph and the second homograph, taking each user and item as a central node, taking nodes within the length and size range of the random walk sequence as positive samples, and randomly selecting some nodes from the nodes in the heterograms, the first homograph and the second homograph as negative samples outside the positive samples;
training each sequence in the heterograms, the first homogeneous subgraph and the second homogeneous subgraph by using a jump sub model, so as to obtain a user embedded vector, a project embedded vector, a first meta-path embedded vector and a second meta-path embedded vector;
the first element path embedded vector and the second element path embedded vector are fused through Hadamard products, and a joint embedded vector is obtained;
the step 3 specifically comprises the following steps:
the joint embedding vector is firstly converted into probability distribution, and the expression is as follows:
；
wherein,is the i-th element in the joint embedded vector, < >>Is the j-th element in the joint embedded vector, is->Is the total number of elements in the joint embedded vector, +.>Representing the conversion vector by softmax function +.>Values in the probability distribution obtained;
then isAdding noise, normalizing to obtain a probability distribution vector +.>The expression is as follows:
；
wherein,is added to/>Gumbel noise of>Is a temperature parameter, ++>Representing the conversion vector by softmax function +.>Values in the probability distribution obtained, +.>Is added to->Gumbel noise of (2);
then forward propagating, training to generate generators in the countermeasure network, definingThe index of the largest element in (2) is +.>Constructing a user independent heat vector->The expression is:
；
then back-propagating, updating the generator in the generation countermeasure network, usingAs the gradient, the following conditional expression is satisfied:
；
wherein,representing differentiation,/->Representing a loss function of the generator;
obtaining the user independent heat vectorThen, obtaining project independent heat vector according to the following steps, firstly establishing interaction matrix of user and project from the original data set>A trainable matrix is then created>Thereby obtaining the project independent heat vector +.>The expression is as follows:
；
wherein,representing a straight-through Gumbi-Softmax function, T representing a transpose operation, ++>Representing the Hadamard product;
in step 4, the loss function of the generatorMeets the following requirementsThe following conditional expression:
；
wherein,is a mathematical desired operation, +.>Representing Sigmoid function->Representing the score of user u for item v, +.>Representing user u's one-hot vector for the project>Score of->Representing a collection of synthetic items, +.>A conditional probability distribution representing the synthesized item;
loss function of discriminatorThe following conditional expression is satisfied:
；
wherein,learnable parameters representing a discriminator, +.>Representing the score of user u for item j, +.>Is a regularization term;
in step 5, an objective function of the countermeasure network is generatedThe expression of (2) is:
；
wherein,representing the discriminator->A representation generator.
2. The social recommendation method based on generating an countermeasure network according to claim 1, wherein in step 5, parameters of a generator and a discriminator generating the countermeasure network are updated by random gradient rising and falling, respectively, expressed as:
；
wherein,、/>the learnable parameters of the generators corresponding to the time t and the time t+1 are respectively represented by +.>Indicates learning rate (I/O)>Representing the gradient of the parameters in the loss function of the generator,/->、/>The learnable parameters of the discriminators corresponding to the time t and the time t+1 are respectively represented, and the +.>Representing the gradient of the parameters in the loss function of the arbiter,/->Score representing user u interacting with item v in generator, < >>Representing user u in the generator together with the item one-hot vector +.>Score of interaction->Score indicating user u interacted with item v in the arbiter, ++>Representing user u in the arbiter with the item one-hot vector +.>Score of interaction->Representing the score of user u interacting with item j in the arbiter.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410016462.6A CN117520665B (en) | 2024-01-05 | 2024-01-05 | Social recommendation method based on generation of countermeasure network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410016462.6A CN117520665B (en) | 2024-01-05 | 2024-01-05 | Social recommendation method based on generation of countermeasure network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117520665A CN117520665A (en) | 2024-02-06 |
| CN117520665B true CN117520665B (en) | 2024-03-26 |
Family
ID=89751719
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410016462.6A Active CN117520665B (en) | 2024-01-05 | 2024-01-05 | Social recommendation method based on generation of countermeasure network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117520665B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110442781A (en) * | 2019-06-28 | 2019-11-12 | 武汉大学 | It is a kind of based on generate confrontation network to grade ranked items recommended method |
| CN113569906A (en) * | 2021-06-10 | 2021-10-29 | 重庆大学 | Heterogeneous graph information extraction method and device based on meta-path subgraph |
| CN113609398A (en) * | 2021-08-17 | 2021-11-05 | 石家庄铁道大学 | Social recommendation method based on heterogeneous graph neural network |
| CN115082142A (en) * | 2022-05-10 | 2022-09-20 | 华南理工大学 | Recommendation method, device and medium based on heterogeneous relational graph neural network |
| WO2023024017A1 (en) * | 2021-08-26 | 2023-03-02 | Ebay Inc. | Multi-modal hypergraph-based click prediction |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113222700B (en) * | 2021-05-17 | 2023-04-18 | 中国人民解放军国防科技大学 | Session-based recommendation method and device |
-
2024
- 2024-01-05 CN CN202410016462.6A patent/CN117520665B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110442781A (en) * | 2019-06-28 | 2019-11-12 | 武汉大学 | It is a kind of based on generate confrontation network to grade ranked items recommended method |
| CN113569906A (en) * | 2021-06-10 | 2021-10-29 | 重庆大学 | Heterogeneous graph information extraction method and device based on meta-path subgraph |
| CN113609398A (en) * | 2021-08-17 | 2021-11-05 | 石家庄铁道大学 | Social recommendation method based on heterogeneous graph neural network |
| WO2023024017A1 (en) * | 2021-08-26 | 2023-03-02 | Ebay Inc. | Multi-modal hypergraph-based click prediction |
| CN115082142A (en) * | 2022-05-10 | 2022-09-20 | 华南理工大学 | Recommendation method, device and medium based on heterogeneous relational graph neural network |
Non-Patent Citations (2)
| Title |
|---|
| 基于异质图神经网络的社交推荐方法研究;姜彦斌;《知网》;20230215;第2023年卷(第02期);全文 * |
| 基于生成对抗网络的长短兴趣推荐模型;康嘉钰;苏凡军;;计算机技术与发展;20200610(06);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117520665A (en) | 2024-02-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110837598B (en) | Information recommendation method, device, equipment and storage medium | |
| Ma et al. | Personalized scientific paper recommendation based on heterogeneous graph representation | |
| Li et al. | Restricted Boltzmann machine-based approaches for link prediction in dynamic networks | |
| CN110910218B (en) | Multi-behavior migration recommendation method based on deep learning | |
| Li et al. | Deep probabilistic matrix factorization framework for online collaborative filtering | |
| CN110677284B (en) | Heterogeneous network link prediction method based on meta path | |
| Hong et al. | Genetic-fuzzy data mining with divide-and-conquer strategy | |
| Chitty-Venkata et al. | Neural architecture search for transformers: A survey | |
| TW201915790A (en) | Generating document for a point of interest | |
| CN111143705B (en) | Recommendation method based on graph convolution network | |
| CN111949885B (en) | Personalized recommendation method for scenic spots | |
| Chen et al. | Neural feature-aware recommendation with signed hypergraph convolutional network | |
| CN113127737B (en) | Personalized search method and search system integrating attention mechanism | |
| CN114564596A (en) | Cross-language knowledge graph link prediction method based on graph attention machine mechanism | |
| Nistor et al. | IntelliSwAS: Optimizing deep neural network architectures using a particle swarm-based approach | |
| CN117216281A (en) | Knowledge graph-based user interest diffusion recommendation method and system | |
| CN115129807A (en) | Fine-grained classification method and system for social media topic comments based on self-attention | |
| CN113590976A (en) | Recommendation method of space self-adaptive graph convolution network | |
| Wu et al. | A Federated Deep Learning Framework for Privacy-Preserving Consumer Electronics Recommendations | |
| CN112559904A (en) | Conversational social recommendation method based on door mechanism and multi-modal graph network | |
| CN117520665B (en) | Social recommendation method based on generation of countermeasure network | |
| Han et al. | A generalization of recurrent neural networks for graph embedding | |
| Boudiba et al. | Tag-based embedding representations in neural collaborative filtering approaches. | |
| Ali et al. | Gradient-Based Neural Architecture Search: A Comprehensive Evaluation | |
| Cui et al. | RAKCR: Reviews sentiment-aware based knowledge graph convolutional networks for Personalized Recommendation |
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 |