CN117171449B - Recommendation method based on graph neural network - Google Patents
Recommendation method based on graph neural network Download PDFInfo
- Publication number
- CN117171449B CN117171449B CN202311224098.4A CN202311224098A CN117171449B CN 117171449 B CN117171449 B CN 117171449B CN 202311224098 A CN202311224098 A CN 202311224098A CN 117171449 B CN117171449 B CN 117171449B
- Authority
- CN
- China
- Prior art keywords
- user
- vector
- updated
- project
- item
- 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 44
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 19
- 239000013598 vector Substances 0.000 claims abstract description 157
- 230000003993 interaction Effects 0.000 claims abstract description 29
- 239000011159 matrix material Substances 0.000 claims description 27
- 230000008569 process Effects 0.000 claims description 13
- 230000009466 transformation Effects 0.000 claims description 13
- 238000010606 normalization Methods 0.000 claims description 5
- 238000010276 construction Methods 0.000 claims description 2
- 230000000875 corresponding effect Effects 0.000 description 11
- 230000006399 behavior Effects 0.000 description 8
- 230000000694 effects Effects 0.000 description 4
- 230000006872 improvement Effects 0.000 description 4
- 210000005036 nerve Anatomy 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000009825 accumulation Methods 0.000 description 2
- 230000001186 cumulative effect Effects 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000012163 sequencing technique Methods 0.000 description 2
- VYPSYNLAJGMNEJ-UHFFFAOYSA-N Silicium dioxide Chemical compound O=[Si]=O VYPSYNLAJGMNEJ-UHFFFAOYSA-N 0.000 description 1
- 230000003542 behavioural effect Effects 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 230000001276 controlling effect Effects 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000002085 persistent effect Effects 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a recommendation method based on a graph neural network, which comprises the following steps: respectively distributing an initial embedded vector to each user, item and relation; inputting the initialized user vector and project vector into the graphic neural network for processing to obtain a corresponding updated user vector and updated project vector; inputting the updated user vector, the updated project vector and the initialized relation vector into a graph neural network for secondary processing to obtain a corresponding secondary updated user vector and a secondary updated project vector; weighting all the obtained user vectors after secondary updating and project vectors after secondary updating respectively to obtain final user vectors and final project vectors; and performing dot product operation on the final user vector and the final item vector to obtain interaction probability of the user on the corresponding item, and generating a recommendation result. The recommendation method can accurately understand interests and demands of users and realize personalized recommendation.
Description
Technical Field
The invention belongs to the field of intelligent recommendation, and particularly relates to a recommendation method based on a graph neural network.
Background
With the advent of the information age, the total amount of information has exploded beyond the point where humans can handle or make efficient use of it in time. In the face of endlessly complex information, the recommendation system can overcome information overload and help users to acquire really needed information. In addition, the recommendation system is widely applied to the fields of news, movies, commodities and the like. Therefore, in the current big data context, intelligent systems that accurately and individually recommend user needs are becoming particularly important.
Most of the existing recommendation methods still have the problems of cold start of a recommendation system, insufficient data sparsity and recommendation interpretation and the like; the conventional recommending method has good recommending effect on hot articles, but has poor recommending effect on long-tail articles; the current recommendation method can only recommend articles similar to the historical behaviors of users, and differences among the articles are ignored, so that the recommended articles are too homogeneous.
The current recommendation algorithm based on the graph nerve also has three main problems: 1. ignoring the user's different personalized interests in the relationship; 2. focusing on extracting information from the knowledge graph, but ignoring interaction information contained in the user-project history interactions; 3. other complex models are introduced as auxiliary tools, which causes the problems of long training time, high consumption of computing resources and the like due to the rapid increase of parameters and computing complexity.
Therefore, there is a need to develop an intelligent recommendation method suitable for various scenes to provide personalized recommendation services, increase the interpretability of the recommendation results, and improve the quality and user experience of the recommendation results.
Disclosure of Invention
The invention aims to provide a recommendation method based on a graph neural network, so as to solve the problems in the prior art.
In order to achieve the above object, the present invention provides a recommendation method based on a graph neural network, comprising the following steps:
respectively distributing an initial embedded vector to each user, item and relation;
inputting the initialized user vector and project vector into the graphic neural network for processing to obtain a corresponding updated user vector and updated project vector;
inputting the updated user vector, the updated project vector and the initialized relation vector into a graph neural network for secondary processing to obtain a corresponding secondary updated user vector and a secondary updated project vector;
weighting all the obtained user vectors after secondary updating and project vectors after secondary updating respectively to obtain final user vectors and final project vectors;
and performing dot product operation on the final user vector and the final item vector to obtain interaction probability of the user on the corresponding item, and generating a recommendation result.
Optionally, the process of obtaining the corresponding updated user vector and updated item vector includes: constructing a user project bipartite graph based on the user project interaction matrix; acquiring the number of user surrounding project nodes to be updated and the number of user surrounding nodes to be updated based on the user project bipartite graph; based on the number of project nodes around the user to be updated, obtaining updated user vectors; and obtaining updated project vectors based on the number of user nodes around the project to be updated.
Optionally, the process of constructing the user item bipartite graph includes:
the process for constructing the user project bipartite graph comprises the following steps: creating user nodes and project nodes based on a user project interaction matrix, and adding an edge between the corresponding user nodes and project nodes based on interaction between the user and the project; and then, based on the interaction strength between the user and the project, distributing a weight for each side, and completing the construction of the user project bipartite graph.
Optionally, the formula for obtaining the updated user vector is:
the formula for obtaining the updated item vector is:
wherein,representing updated user vector->Representing updated item vector,/->Representing the number of item nodes around the user to be updated, < >>Representing the number of user nodes around the item to be updated, < >>Representing a set of item nodes surrounding the user to be updated, < + >>A set of user nodes surrounding the item to be updated.
Optionally, the process of obtaining the corresponding secondary updated user vector and the secondary updated item vector includes: acquiring a knowledge graph vector, inputting the knowledge graph vector and the updated user vector into a graph neural network, and respectively acquiring a corresponding first attention representation and a corresponding second attention representation; obtaining a second updated project vector based on the first attention representation; based on the second attention representation, a second updated user vector is obtained.
Optionally, the process of obtaining the first attention representation comprises: performing dot product operation on the updated user vector and the initialized relation vector to obtain a first operation result; performing dot product operation on the updated project vector and the initialized relation vector to obtain a second operation result; and carrying out weighting and normalization processing on the first operation result and the second operation result to obtain a first attention representation.
Optionally, the process of obtaining the second attention representation comprises: acquiring a first feature transformation matrix, a second feature transformation matrix and a third feature transformation matrix, and performing dot product operation on the updated user vector and the first feature transformation matrix, the second feature transformation matrix and the third feature transformation matrix respectively to acquire a corresponding first vector, a corresponding second vector and a corresponding third vector; and carrying out dot product operation on the first vector and the second vector, carrying out normalization processing on the operation result, and carrying out weighted summation on the obtained normalization score and the third vector to obtain a second attention representation.
Optionally, the process of obtaining the item vector after the second update includes: summing the updated project vector and the first attention representation to obtain a second updated project vector;
optionally, the process of obtaining the user vector after the second update includes: and summing the updated user vector and the second attention representation to obtain a second updated user vector.
The invention has the technical effects that:
1. according to the invention, through the application of the graph neural network, the interests and the demands of the user can be more accurately understood, more personalized recommendation is realized, and the unique demands of the user are met.
2. The invention can capture the complex relationship between the user and the project through the attention-introducing mechanism, so that the recommendation is more accurate, and the user is more likely to be interested in the recommended content.
3. According to the method and the device, the knowledge graph is introduced when the user vector and the project vector are updated secondarily, so that the interpretability of the improvement of the user trust and the user experience can be realized.
Drawings
The accompanying drawings, which are included to provide a further understanding of the application, illustrate and explain the application and are not to be construed as limiting the application. In the drawings:
FIG. 1 is a flowchart of a recommendation method according to an embodiment of the present invention;
fig. 2 is a schematic diagram of a network architecture based on a graph nerve recommendation method in a movie recommendation system according to an embodiment of the present invention;
fig. 3 is a schematic block diagram of a movie recommendation system according to an embodiment of the present invention.
Detailed Description
It should be noted that, in the case of no conflict, the embodiments and features in the embodiments may be combined with each other. The present application will be described in detail below with reference to the accompanying drawings in conjunction with embodiments.
It should be noted that the steps illustrated in the flowcharts of the figures may be performed in a computer system such as a set of computer executable instructions, and that although a logical order is illustrated in the flowcharts, in some cases the steps illustrated or described may be performed in an order other than that illustrated herein.
Example 1
As shown in fig. 1, in this embodiment, a recommendation method based on a graph neural network is provided, which specifically includes the following steps:
step 1: for each user e u ∈R d (entity e) i Relationship e r ,e i ,e r ∈R d ) Assigning an initial embedded vector:the initialization vector is 64-dimensional.
Step 2: the user initialized in the step 1And item vector->Inputting a lightweight graph convolutional network module (Light graph convolutional network, LGC) to obtain a user +.>And item->Vector.
Step 3: outputting the user output in the step 2And item->Vector, and initialized relationship in step 1And entity->Vector input attention and graph roll-up network module (knowledges-aware layer of attention and graph convolutional networks, KAGC) to get user +.>And item->Vector.
Step 4: repeating the steps 2 and 3 to obtain a plurality of usersAnd item->Vector, and respectively weighting and combining user and project vector to obtain user +.>And item->Vector.
Step 5: and (4) enabling the user obtained in the step (4) to performAnd item->The vector obtains the interaction probability of the user on the item through dot product operation, and a recommendation result is returned.
Further, step 2 specifically includes:
step 21: constructing a user project bipartite graph by using the user project interaction matrix;
further, step 21 specifically includes:
establishing a user item interaction matrix: the activities of the user on the platform, such as clicking, browsing, purchasing, scoring, commenting, sharing, favorites and the like, are monitored through methods of user behavior logs, user feedback surveys, user purchase histories, user social media data and the like, and various interaction behaviors or feedback data between the user and the project are tidied. And marking the positive interaction behaviors among the user items as 1, marking the non-positive interaction behaviors among the user items as 0, and forming a user item interaction matrix.
And constructing a user project bipartite graph according to the user project interaction matrix. Specifically, user nodes and project nodes are created based on a user project interaction matrix. Each user and item is represented in the figure as a node. Ensuring that each node has a unique identifier. Next, edges are added to the bipartite graph according to the user item interaction matrix. Typically, if there is an interaction between the user and the item, an edge is added between them. For example, if user A scored item X, an edge is added between user A node and item X node. For each edge, a weight may be assigned that reflects the strength or importance of the interaction.
In this figure, the user and the item form two non-overlapping node sets, respectively. Each user node corresponds to a row of the matrix and each item node corresponds to a column of the matrix. Then, for the user and the project with the interaction record of 1 in the user project interaction matrix, an edge connection with the weight of 1 is added between the user and the project.
Step 22: obtaining the number of project nodes around a certain user through a bipartite graphObtaining the number of user nodes around a certain project through a bipartite graph to obtain +.>
Step 23: using the results of step 22, the vector representations of the user and the item are updated.
Further, step 23 specifically includes:
by passing throughUpdating the vector representation of the user;
by passing throughUpdating the vector representation of the item;
further, the step 3 specifically includes:
step 31: inputting the knowledge graph vector into a graph neural network to obtain attention expression;
step 32: processing the user vector to obtain a self-attention representation;
step 33: the item vector is represented by the attention in step 31 for updating the item vector and the user vector is represented by the self-attention in step 32 for updating the user vector;
further, the step 31 specifically includes:
first, a desired knowledge graph is established, and relationships between entities, such as associations between movies and actors, are constructed by extracting related entities (e.g., movies, books, music, characters, etc.) and their attribute information (e.g., actors, directors, year of publication, etc.) from wikidada. Next, behavioral data of the user (e.g., browsing history, score, purchase record of the user) is correlated with the entities in the knowledge-graph to form a user-entity connection.
For each entity e in the knowledge graph i ∈R d Relationship e r ∈R d Assigning an initial embedded vector:the initialization vector is 64-dimensional.
Will userRelation to initialization->Dot product operation is performed to obtain +.> A score of attention between the user and the relationship;
project is put intoRelation to initialization->Dot product operation is performed to obtain +.> Referred to as an attention score between the user and the item;
the above-mentioned materials are mixedAnd->By->Weighting is achieved->A personalized score of attention to the item referred to by the user;
will beNormalized to obtain->A normalized personalized score attention score for the item referred to by the user;
by passing throughObtaining neighbor vector representations of the items;
further, the step 32 specifically includes:
by the userVector, three vectors are calculated using the following equation, W q ，W k ，W v The transformation matrix is characterized by:
through Q u And K is equal to u Is obtained by vector dot product of (2)Referred to as user self-attention score;
will beNormalized to obtain->A user self-attention score called normalized;
by passing throughA self-attention vector representation of the user is obtained.
Further, step 33 specifically includes:
by passing throughObtaining a vector representation of the item->
By passing throughGet vector representation of user->
Further, step 4 specifically includes:
obtaining a plurality of user vectors by utilizing the step 3
Obtaining a plurality of project vectors by utilizing the step 3
Passing multiple user vectors throughGet user vector +.>
Passing multiple project vectors throughGet user vector +.>
Comparison experiment:
TABLE 1
According to Table 1, the inventive method consistently yields the best recommended performance on the MovieLens-20M dataset, the Last-FM dataset, and the Book-cross dataset. On the MovieLens-20M dataset, the recall @20 and normalized break cumulative gain @20 were improved by 14.41% and 15.07%, respectively, as compared to the best performing baseline LightGCN method. 8.86% and 18.82% improvement on the Last-FM dataset and 20.90% and 22.80% improvement on the Book-cross dataset, respectively. Recall is a measure that evaluates how well a recommendation system can incorporate user interests into a recommendation list given those interests. The recall boost representation system is able to more fully cover items that the user may be interested in, reducing the instances of missing user interest, which is important to ensure that the user does not miss what may be liked.
The normalized damage accumulation gain is an index for evaluating the sorting quality of a recommendation system, and the ranking and the relevance of the articles in a recommendation list are considered. The normalized break-up cumulative gain boost representation system not only provides items with high relevance, but also ranks them higher, ensuring that the user more easily sees and interacts with the items, which helps to improve user satisfaction and click-through rate.
Therefore, the simultaneous improvement of the recall rate and the normalized damage accumulation gain shows that the recommended performance of the method of the invention achieves good balance: the method not only covers the interests of the user as much as possible, but also presents the interests in a more relevant way, and provides attractive personalized recommendation, which is helpful for improving the user experience and increasing the trust of the user to the recommendation system, thereby promoting the participation and interaction of the user.
This shows that the method of the present invention is good at incorporating complementary information in the knowledge graph of the item into the item vector representation and has an insight into the unique requirements and potential interests of different users; therefore, it can generate better recommendations, thereby improving the performance of the recommendation system. Meanwhile, the method adds the project knowledge graph and the attention mechanism, so that the problems of data sparsity and insufficient interpretation of a recommendation system can be well relieved.
Example two
As shown in fig. 2, the steps of performing movie recommendation based on the graph nerve recommendation method in the movie recommendation system include: 1. inputting user information, project information and knowledge graph information into an initialization vector layer, initializing a user, a project and an entity into a 64-dimensional vector representation, wherein the user vector isThe item vector is +.>2. Inputting an initial user vector and an item vector into a first layer in a double-layer picture convolution information propagation layer (BIKAGCN), and updating vector representations of the user and the item by utilizing information propagation rules of the layer, wherein the key function of the step is to extract collaborative information between the user and the item interaction matrix; 3. inputting the updated user and item vector representation of the first layer and the initialized knowledge-graph information into a second layer of a double-layer picture convolution information propagation layer (BIKAGCN) through mergingIdentifying pattern related information to update user and project vector representations; 4. the two steps 2.3 are repeated for a plurality of times to obtain a plurality of user and project vector representations, then user project vector representations of different layers are summarized to obtain final user and project vector representations, and the final user vector is e u The final item vector is e i The method comprises the steps of carrying out a first treatment on the surface of the 4. Obtaining the final recommended prediction +.>5. The model is trained and optimized through the loss function.
Fig. 3 is a schematic block diagram of a movie recommendation system based on the graph nerve recommendation method. Specifically, the structure of the movie recommendation system includes:
and the recall module is the first step of the recommendation flow, screens out part of the content in the whole data set and sends the part of the content to the sequencing module. This step can filter out most of the content that does not meet the user's interests, thereby reducing the number of candidate sets to be processed, and thus reducing the computational burden of the ranking module. During the recall phase, using the recommendation method of the present embodiment, it is possible to efficiently calculate a candidate set containing movies of interest to the user. Specifically, the recommendation method of the embodiment can learn the preference of the user on the movie and the characteristics of the movie, namely potential factors, and obtain the scoring predicted value of the user on the movie by calculating the inner product of the user and the potential factors of the movie. Then, a certain number of movies in the top ranking are selected as candidate sets, and sent to a ranking module for further processing.
The ranking module is responsible for classifying the candidate movie sets, and by using the recommendation method of the embodiment, the system inserts a user-item pair into each movie of the candidate set, obtains a prediction result, and finally ranks the movies according to the prediction result.
The business strategy module is an additional method introduced for manually controlling the recommendation list. It is responsible for implementing certain predefined measures of business policy in the output of the ranking module, such as increasing or decreasing the ranking of certain movies. The business policy may be adjusted by adjusting the priority of movies according to predefined rules, such as a predetermined ordering of platform products, or according to movies recommended by the viewer.
The user behavior collection module is used for the user to work in the following three cases: firstly, after receiving a recommendation list, a user can select to click or watch certain movies, which is a behavior for directly judging the accuracy of a model, and the behavior can be used as real-time training data to be sent to a sequencing model for updating; second, users actively score or comment on movies, which are all required by the recall model algorithm, and can be saved as offline training data.
The history information module is responsible for persisting the collected user information into a database, which is the training data required by the recall model algorithm.
The foregoing is merely a preferred embodiment of the present application, but the scope of the present application is not limited thereto, and any changes or substitutions easily contemplated by those skilled in the art within the technical scope of the present application should be covered by the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.
Claims (2)
1. The recommendation method based on the graph neural network is characterized by comprising the following steps of:
respectively distributing an initial embedded vector to each user, item and relation;
inputting the initialized user vector and project vector into the graphic neural network for processing to obtain a corresponding updated user vector and updated project vector;
inputting the updated user vector, the updated project vector and the initialized relation vector into a graph neural network for secondary processing to obtain a corresponding secondary updated user vector and a secondary updated project vector;
weighting all the obtained user vectors after secondary updating and project vectors after secondary updating respectively to obtain final user vectors and final project vectors;
performing dot product operation on the final user vector and the final item vector to obtain interaction probability of the user on the corresponding item, and generating a recommendation result;
the process of obtaining the corresponding updated user vector and updated project vector comprises: constructing a user project bipartite graph based on the user project interaction matrix; the process for constructing the user project bipartite graph comprises the following steps: creating user nodes and project nodes based on a user project interaction matrix, and adding an edge between the corresponding user nodes and project nodes based on interaction between the user and the project; then, based on the interaction strength between the user and the project, a weight is distributed to each side, and the construction of a user project bipartite graph is completed; then, based on the user project bipartite graph, the number of the project nodes around the user to be updated and the number of the user nodes around the user to be updated are obtained; based on the number of project nodes around the user to be updated, obtaining updated user vectors; based on the number of user nodes around the item to be updated, obtaining an updated item vector;
the process of obtaining the corresponding user vector after the secondary update and the project vector after the secondary update comprises the following steps: acquiring a knowledge graph vector, inputting the knowledge graph vector and the updated user vector into a graph neural network, and respectively acquiring a corresponding first attention representation and a corresponding second attention representation; summing the updated project vector and the first attention representation to obtain a second updated project vector; summing the updated user vector and the second attention representation to obtain a second updated user vector;
the process of obtaining the first attention representation includes: performing dot product operation on the updated user vector and the initialized relation vector to obtain a first operation result; performing dot product operation on the updated project vector and the initialized relation vector to obtain a second operation result; weighting and normalizing the first operation result and the second operation result to obtain a first attention representation;
the process of obtaining the second attention representation includes: acquiring a first feature transformation matrix, a second feature transformation matrix and a third feature transformation matrix, and performing dot product operation on the updated user vector and the first feature transformation matrix, the second feature transformation matrix and the third feature transformation matrix respectively to acquire a corresponding first vector, a corresponding second vector and a corresponding third vector; and carrying out dot product operation on the first vector and the second vector, carrying out normalization processing on the operation result, and carrying out weighted summation on the obtained normalization score and the third vector to obtain a second attention representation.
2. The method for recommending a data based on a graphic neural network according to claim 1, wherein,
the formula for obtaining updated user vectors is:
the formula for obtaining the updated item vector is:
wherein,representing updated user vector->Representing updated item vector,/->Representing the number of item nodes around the user to be updated, < >>Representing the number of user nodes around the item to be updated, < >>Representing a set of item nodes surrounding the user to be updated, < + >>A set of user nodes surrounding the item to be updated.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311224098.4A CN117171449B (en) | 2023-09-21 | 2023-09-21 | Recommendation method based on graph neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311224098.4A CN117171449B (en) | 2023-09-21 | 2023-09-21 | Recommendation method based on graph neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117171449A CN117171449A (en) | 2023-12-05 |
| CN117171449B true CN117171449B (en) | 2024-03-19 |
Family
ID=88933608
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311224098.4A Active CN117171449B (en) | 2023-09-21 | 2023-09-21 | Recommendation method based on graph neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117171449B (en) |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112989064A (en) * | 2021-03-16 | 2021-06-18 | 重庆理工大学 | Recommendation method for aggregating knowledge graph neural network and self-adaptive attention |
| CN113158071A (en) * | 2021-03-19 | 2021-07-23 | 广东工业大学 | Knowledge social contact recommendation method, system and equipment based on graph neural network |
| CN113987200A (en) * | 2021-10-19 | 2022-01-28 | 云南大学 | Recommendation method, system, terminal and medium combining neural network with knowledge graph |
| CN115269977A (en) * | 2022-07-18 | 2022-11-01 | 燕山大学 | Recommendation method for fusion knowledge and collaborative information based on graph neural network |
| CN115329215A (en) * | 2022-08-19 | 2022-11-11 | 北京建筑大学 | Recommendation method and system based on self-adaptive dynamic knowledge graph in heterogeneous network |
| CN115374347A (en) * | 2022-07-07 | 2022-11-22 | 江苏科技大学 | Social recommendation method based on knowledge graph attention network |
| CN115545300A (en) * | 2022-09-30 | 2022-12-30 | 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) | Method and device for predicting user behavior based on graph neural network |
| CN115795139A (en) * | 2021-09-08 | 2023-03-14 | 天翼电子商务有限公司 | Individualized recommendation method based on knowledge graph attribute mixing |
| CN115795022A (en) * | 2022-11-22 | 2023-03-14 | 广东工业大学 | Recommendation method, system, equipment and storage medium based on knowledge graph |
| CN116628328A (en) * | 2023-03-09 | 2023-08-22 | 浙江大学 | Web API recommendation method and device based on functional semantics and structural interaction |
| CN116662656A (en) * | 2023-05-29 | 2023-08-29 | 浙江工业大学 | Movie recommendation method based on collaborative enhancement and graph annotation intention neural network |
| CN116662601A (en) * | 2023-05-29 | 2023-08-29 | 浙江工业大学 | Song recommendation method based on graphic neural network and knowledge graph |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TW202318283A (en) * | 2021-10-25 | 2023-05-01 | 智影顧問股份有限公司 | Dispatching planning system of virtual foreman including a knowledge map unit, a matchmaking unit, and a recommendation unit |
| US20230177631A1 (en) * | 2021-12-08 | 2023-06-08 | International Business Machines Corporation | Automatically enhancing content items through identifying relationships |
-
2023
- 2023-09-21 CN CN202311224098.4A patent/CN117171449B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112989064A (en) * | 2021-03-16 | 2021-06-18 | 重庆理工大学 | Recommendation method for aggregating knowledge graph neural network and self-adaptive attention |
| CN113158071A (en) * | 2021-03-19 | 2021-07-23 | 广东工业大学 | Knowledge social contact recommendation method, system and equipment based on graph neural network |
| CN115795139A (en) * | 2021-09-08 | 2023-03-14 | 天翼电子商务有限公司 | Individualized recommendation method based on knowledge graph attribute mixing |
| CN113987200A (en) * | 2021-10-19 | 2022-01-28 | 云南大学 | Recommendation method, system, terminal and medium combining neural network with knowledge graph |
| CN115374347A (en) * | 2022-07-07 | 2022-11-22 | 江苏科技大学 | Social recommendation method based on knowledge graph attention network |
| CN115269977A (en) * | 2022-07-18 | 2022-11-01 | 燕山大学 | Recommendation method for fusion knowledge and collaborative information based on graph neural network |
| CN115329215A (en) * | 2022-08-19 | 2022-11-11 | 北京建筑大学 | Recommendation method and system based on self-adaptive dynamic knowledge graph in heterogeneous network |
| CN115545300A (en) * | 2022-09-30 | 2022-12-30 | 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) | Method and device for predicting user behavior based on graph neural network |
| CN115795022A (en) * | 2022-11-22 | 2023-03-14 | 广东工业大学 | Recommendation method, system, equipment and storage medium based on knowledge graph |
| CN116628328A (en) * | 2023-03-09 | 2023-08-22 | 浙江大学 | Web API recommendation method and device based on functional semantics and structural interaction |
| CN116662656A (en) * | 2023-05-29 | 2023-08-29 | 浙江工业大学 | Movie recommendation method based on collaborative enhancement and graph annotation intention neural network |
| CN116662601A (en) * | 2023-05-29 | 2023-08-29 | 浙江工业大学 | Song recommendation method based on graphic neural network and knowledge graph |
Non-Patent Citations (4)
| Title |
|---|
| Wenqi Fan等.A graph neural network Framework for social Recommendations.《IEEE Transactions on Knowledge and Data Engineering》.第34卷(第5期),2033-2-47. * |
| 图卷积算法的研究进展;郑睿刚 等;《中山大学学报(自然科学版)》;20200315(第02期);7-20 * |
| 基于双用户视角与知识图谱注意力网络的推荐模型;张素琪 等;《现代计算机》;20200505(第13期);4-10 * |
| 基于注意力机制增强图卷积神经网络的个性化新闻推荐;杨宝生;;《兰州文理学院学报(自然科学版)》;20200910(第05期);57-62 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117171449A (en) | 2023-12-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Cheng et al. | Learning to recommend accurate and diverse items | |
| Thorat et al. | Survey on collaborative filtering, content-based filtering and hybrid recommendation system | |
| Forouzandeh et al. | Presentation a Trust Walker for rating prediction in recommender system with Biased Random Walk: Effects of H-index centrality, similarity in items and friends | |
| Bauer et al. | Recommender systems based on quantitative implicit customer feedback | |
| Gogna et al. | DiABlO: Optimization based design for improving diversity in recommender system | |
| Wang | Integrating Kansei engineering with conjoint analysis to fulfil market segmentation and product customisation for digital cameras | |
| Wang et al. | HGNN: Hyperedge-based graph neural network for MOOC course recommendation | |
| CN113051468B (en) | Movie recommendation method and system based on knowledge graph and reinforcement learning | |
| Zarzour et al. | RecDNNing: a recommender system using deep neural network with user and item embeddings | |
| Symeonidis et al. | Counteracting the filter bubble in recommender systems: Novelty-aware matrix factorization | |
| Liu et al. | Collaborative prediction for multi-entity interaction with hierarchical representation | |
| Thomas et al. | Machine learning techniques for recommender systems–a comparative case analysis | |
| Kavitha Devi et al. | Kernel based collaborative recommender system for e-purchasing | |
| Sreepada et al. | Multi-criteria recommendations through preference learning | |
| Yazdi et al. | Improving recommender systems accuracy in social networks using popularity | |
| CN117171449B (en) | Recommendation method based on graph neural network | |
| Nassa et al. | A personalized recommender system | |
| Regi et al. | A survey on recommendation techniques in E-Commerce | |
| CN112559905B (en) | Conversation recommendation method based on dual-mode attention mechanism and social similarity | |
| Wangwatcharakul et al. | Improving dynamic recommender system based on item clustering for preference drifts | |
| Kanaujia et al. | Real time financial analysis using big data technologies | |
| Zarei et al. | An adaptive similarity measure to tune trust influence in memory-based collaborative filtering | |
| Zuva et al. | Virtual group movie recommendation system using social network information | |
| Saraswathi et al. | Survey: a hybrid approach to solve cold-start problem in online recommendation system | |
| Ghanwat et al. | Improved personalized recommendation system with better user experience |
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 |