CN112364242B - Graph convolution recommendation system for context awareness - Google Patents
Graph convolution recommendation system for context awareness Download PDFInfo
- Publication number
- CN112364242B CN112364242B CN202011249269.5A CN202011249269A CN112364242B CN 112364242 B CN112364242 B CN 112364242B CN 202011249269 A CN202011249269 A CN 202011249269A CN 112364242 B CN112364242 B CN 112364242B
- Authority
- CN
- China
- Prior art keywords
- user
- article
- information
- context information
- context
- 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
- 239000013598 vector Substances 0.000 claims abstract description 45
- 238000012512 characterization method Methods 0.000 claims abstract description 24
- 230000003993 interaction Effects 0.000 claims description 33
- 239000011159 matrix material Substances 0.000 claims description 9
- 238000005096 rolling process Methods 0.000 claims description 8
- 230000003068 static effect Effects 0.000 claims description 6
- 238000003475 lamination Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 230000004913 activation Effects 0.000 claims description 2
- 230000010354 integration Effects 0.000 claims description 2
- 238000001914 filtration Methods 0.000 abstract description 11
- 230000000694 effects Effects 0.000 abstract description 6
- 238000013528 artificial neural network Methods 0.000 description 8
- 230000008901 benefit Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 238000000034 method Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 230000002452 interceptive effect Effects 0.000 description 2
- 240000000233 Melia azedarach Species 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 239000000306 component Substances 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 239000008358 core component Substances 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
Abstract
The invention discloses a graph convolution recommendation system aiming at context awareness, which comprises the following steps: an encoder, a layer of graph convolution, and a decoder; the encoder associates a hidden space vector for each non-zero feature of the input user information, item information and context information and combines these hidden space vectors from the three fields of the user information, item information and context information; in the graph convolution layer, based on a user-object bipartite graph with attributes which is constructed in advance, and combining the output of an encoder to perform graph convolution operation, and obtaining the final characteristic representation of the user and the object through a plurality of graph convolution operations; the decoder predicts the user's preference for items under the context information based on the final characterization of the user and the items and the associated embedded set of context information. The system is a universal recommendation system framework which is suitable for online service, can combine various auxiliary information, can capture collaborative filtering effect, and improves model performance.
Description
Technical Field
The invention relates to the field of recommendation systems and graph data mining, in particular to a graph convolution recommendation system aiming at context awareness.
Background
Personalized recommendation systems have become an indispensable service in the current internet as an important tool for alleviating information overload and improving user experience. Collaborative filtering model is one of the most representative recommendation models that uses historical interaction records of users and items, such as clicks, purchases, etc., to map each user and item to a high-dimensional vector space for personalized recommendation by computing similarity between vectors. Recently, as Graphic Neural Networks (GNNs) have achieved tremendous success in the fields of image processing, natural language processing, and the like, more and more researchers have introduced GNNs into recommendation systems to model collaborative filtering signals into high-order connectivity of user-object bipartite graphs, thereby improving the performance of the model. While the collaborative filtering model provides a general solution, it also has some inherent drawbacks such as the inability to utilize interactively related context information, i.e., the user-item bipartite graph does not contain context information, which can often have a significant impact on the user's choice. For example, in a restaurant recommendation scenario, the time and place factors can effectively filter out unsuitable candidate sets, and in an e-commerce recommendation scenario, the purchasing tendency of a user is often highly similar to that of recent consumption behavior. Therefore, it is important to develop a set of Context-aware recommendation systems (Context-Aware Recommender System, CARS) that can comprehensively consider various auxiliary information.
Existing CARS models generally follow the paradigm of the decomposer model (FactorizationMachine, FM) to translate the problem into a standard supervised learning task. Specifically, it encodes all information related to a record of interactions into a feature vector by means of multiple-hot Encoding (Multi-hot Encoding), and then models interactions between features using different feature interaction modules to predict the user's preference for items in the record. With the rise of neural networks in recent years, feature interaction modules are replaced by neural networks of various structures to enhance the expression capacity thereof. Comprehensive analysis of recent CARS progress we found that they had the following disadvantages: 1) The method adopts a standard supervised learning strategy, omits linkage among data samples, and leads the learned model to not capture collaborative filtering effect well, because a plurality of interaction records need to be considered simultaneously for identifying the collaborative filtering effect; 2) They tend to be of high complexity because they employ well-designed network structures in order to be able to model complex feature interactions, and when serving online, requiring one forward propagation of the network for each user-candidate item pair, such an inefficient, time-consuming inference (reference) strategy is not suitable for online service.
Disclosure of Invention
The invention aims to provide a context-aware graph convolution recommendation system, which is a universal recommendation system framework suitable for online service, can combine various auxiliary information, can capture collaborative filtering effects, and improves model performance.
The invention aims at realizing the following technical scheme:
a graph convolution recommendation system for context awareness, comprising: an encoder, a layer of graph convolution, and a decoder;
the encoder associates a hidden space vector for each non-zero characteristic of the input user information, article information and context information, combines the hidden space vectors from three fields of the user information, the article information and the context information, and outputs an initial representation of the user and the article and an associated embedded set of the context information;
in the picture scroll lamination, based on a user-object bipartite graph with the attribute which is constructed in advance, carrying out picture scroll operation by combining the output of an encoder, and obtaining the final characteristic representation of the user and the object through a plurality of times of picture scroll operation;
the decoder predicts a user's preference for an item under the context information based on the user's final feature characterization of the item and the associated embedded set of context information.
Compared with the prior CARS model based on the neural network, the technical scheme provided by the invention has the following advantages: 1) The accuracy of the test is significantly improved. 2) The system parameters are small, and the model deducing speed is high. 3) The graph rolling operation can effectively improve the expression and generalization capability of the system.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the description of the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic diagram of a graph convolution recommendation system for context awareness according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a data structure and its conversion into a user-object bipartite graph with attributes according to an embodiment of the present invention;
FIG. 3 is a diagram of the number of layers of graph convolution and the impact of whether to model a context for two datasets provided by an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are only some, but not all embodiments of the 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 fall within the scope of the invention.
The embodiment of the invention provides a context-aware graph convolution recommendation system, which is a universal recommendation system framework suitable for online service, can combine various auxiliary information, can capture collaborative filtering effect, and improves model performance.
As shown in fig. 1, a schematic diagram of a context-aware graph convolution recommendation system mainly includes: an encoder, a layer of graph convolution, and a decoder;
the encoder associates a hidden space vector for each non-zero feature of the input user information, item information and context information, combines the hidden space vectors from three fields of the user information, the item information and the context information, and outputs an initial characterization of the user and the item and an associated embedding (embedding) set of the context information;
in the picture scroll lamination, based on a user-object bipartite graph with the attribute which is constructed in advance, carrying out picture scroll operation by combining the output of an encoder, and obtaining the final characteristic representation of the user and the object through a plurality of times of picture scroll operation;
the decoder predicts a user's preference for an item under the context information based on the user's final feature characterization of the item and the associated embedded set of context information.
For ease of understanding, the principles of the above system and its operation are described in detail below.
1. And (3) a data structure.
In order to efficiently organize various information related to interactions, the information is divided into four types: users and their static portraits, items and their static attributes, dynamic context information, and interaction records. In the embodiment of the invention, the context information is an abstract description of a real interactive scene, so the context information can also be called as a context scene. User static portrayal (i.e., user information as described above) refers to a property of the user itself, such as his age, occupation, etc. The static properties of an item (i.e., the item information described above) refer to the relevant property information of the item.
As shown in fig. 2, the above information forms a hybrid data structure, which has a greater descriptive capability than a conventional CARS data format, for example, it may describe that a user interacts with the same item in different context scenarios. Furthermore, the mixed data structure formed by the information is converted into a user-article bipartite graph with attributes, so that the user-article bipartite graph is applicable to a graph neural network. The right side of the figure 2 is a user-object bipartite graph, u and i are high-dimensional sparse feature vectors of a user and an object after multi-heat encoding respectively; c represents a feature vector of the context information after multi-thermal encoding, wherein the bipartite graph comprises two types of nodes, namely user nodes (represented by rectangles) and article nodes (represented by circles), and the features of the user nodes and the article nodes are the respective output features of the encoder; the conjoined edge of the user node and the item node represents the interaction record of the user node and the item node, and the conjoined edge is characterized by a contextual feature (i.e. c 1 、c 2 )。
2. The structure of the system model and the working process thereof.
The result of the system model is shown in fig. 1, and mainly comprises three parts of an encoder, a graph convolution layer and a decoder.
1. An encoder.
The input of the encoder is a sparse feature vector after multiple thermal encodings, which has only a few non-zero elements, i.e. "non-zero features". Taking the user domain as an example, one piece of user information may include: the invention relates to a hidden space vector for each non-zero characteristic in the characteristic vector, so that a user has a plurality of hidden space vectors to describe the hidden space vector. Then, the encoder combines the hidden space vectors to obtain the initial characterization of the user, and the encoder operates the object domain in the same way, and the hidden space vectors are expressed as follows:
in the above formula, u and i respectively represent indexes of a user and an article, P represents an embedding matrix associated with all user features (each feature is associated with one hidden space vector, all hidden space vectors are spliced into a matrix, and the kth row of the matrix represents the embedding vector of the kth user feature and is marked as P) k ) Q represents the embedding matrix associated with all item features (the first row of the matrix represents the embedding vector for the first item feature, denoted as Q l ) The terms u and i represent the number of non-zero features of user u and item i, respectively.
It should be mentioned that in the later picture volume layer we do not update the embedded vector of the context information and therefore the encoder does not pool it but only performs the association of the original feature with the hidden space vector. Thus, the output of the encoder includes the user u initial characterizationAnd item i initial characterization->And is related to the context information cThe combined hidden space vectors form a set +.>(i.e., the previously mentioned associated embedded set of context information), where s is a non-zero feature in c, v s Representing the hidden space vector associated with feature s.
In the embodiment of the invention, the user can comprise the characteristics of gender, age, occupation and the like; the features of branding, category, price and the like of the articles can be included, and the information can be different according to different scenes. The invention is not limited to the nature of the user features, the nature of the item features, and in theory, the invention can handle any type of information.
It will be appreciated by those skilled in the art that the dimension of the hidden space vector is a super-parameter of the model, and that the dimension may be adjustable for different scenarios.
2. And (5) a graph convolution layer.
The graph convolution layer is used for overcoming the defects of the prior CARS model based on the supervised learning strategy, and improves the characterization of the user and the article by utilizing all interactive data of the user-article and capturing the collaborative filtering effect explicitly. In pre-constructed attribute-bearing bipartite graphs, the edges between user nodes and item nodes carry context features that are important for understanding context-dependent interaction patterns, and therefore more accurate user and item characterizations can be learned if the context features are incorporated into the graph convolution operation. Based on this, a new graph convolution operation is proposed in the embodiment of the present invention, expressed as:
wherein,and->Respectively representing all the observed interaction record sets of the user u and the object i, (i, c) representing the object and the context information binary group in one interaction record, and (u, c) representing the user and the context information binary group in one interaction record; />The hidden space vectors associated with the context feature c representing the encoder output form a set; />And (3) withRepresentation regularization; />Each representing a characteristic representation of a user and an article obtained by the first layer of graph rolling operation;each representing a user, feature characterization of the item resulting from the layer 1 graph convolution operation.
Rationality analysis of graph convolution operations: from the point of view of the user,is regularization, which can avoid unreasonable increase of the characterization value caused by increase of the number of the convolution layers; introducing the contextual feature representation into the graph convolution operation in a mean manner and adding it to the user representation, this approach may achieve the following: if the user prefers to interact with the item in a context scene, the user and item and the characterization of the context will be similar; by stacking a plurality of such layers of graph volumes, the user node will collect multi-hop neighbors (neighborsThe node may be a user node or an item node) to improve the characterization of the user. Similar analysis results can also be obtained from the perspective of the article, and will not be described in detail here.
Considering that the semantics of different picture convolution layers are different, after the picture convolution operation of the L layers, the following mode is adopted for integration, and the final characteristic representation of a user and an object is obtained:
wherein alpha is l Representing the weight of the first layer, p u 、q i The final characteristics of the user u and the object i are respectively represented.
3. And a decoder.
The decoder functions to predict the user's preference for items under the context information using the representation obtained by the picture scroll layer. FM can be used as a core component because it is a linear model and has better interpretability than a multi-layer perceptron, formulated as follows:
wherein,preference degree of user u for article i under context information c predicted for decoder; /> Namely->Final characterization of user p including graph convolutional layer output u Final characterization q of the article i And the associated embedding set of context information +.>v k 、v t Are embedded collections->Is a component of the group.
It should be noted that the decoder in the embodiment of the present invention is slightly different from the original FM: previous encoders and picture stacking have mapped all features of a user (or item) into a vector, which has the advantage that the system model can focus more on inter-domain feature interactions while reducing interference from intra-domain feature interactions.
4. And (5) model optimization.
In order to optimize the model parameters, log loss (Log loss function) which is often employed in the recommendation system is selected as an optimization target of the model, and is formulated as follows:
wherein,and->Representing the observed interaction record set (i.e., positive sample set) and the set of multiple negative sample records (i.e., negative sample set) randomly matched for each observation record, respectively, σ (·) represents an S-type activation function, λ is L 2 Regularization coefficient,/->Is a trainable parameter of the model.
Compared with the prior CARS model based on the neural network, the system provided by the embodiment of the invention has the following advantages: 1) The accuracy of the test is significantly improved. 2) The model parameters are small, and the model deducing speed is high. 3) The proposed graph rolling operation can effectively improve the expression and generalization capability of the model.
For the above advantages, detailed experiments were performed on three data sets to demonstrate that table 1 is a data statistic for three data sets:
| data set | Yelp-NC | Yelp-OH | Amazon-Book |
| Number of users | 6,336 | 5,170 | 44,709 |
| Number of articles | 13,003 | 12,997 | 46,831 |
| Interaction number | 185,408 | 143,884 | 1,174,785 |
| User feature number | 24 | 24 | - |
| Item characteristic number | 68 | 213 | 24,816 |
| Contextual feature number | 13,209 | 13,347 | 46,900 |
Table 1 data statistics of the dataset
1. The test precision is obviously improved: the present invention (GCM) was found to improve the accuracy by an average of 13.1% over three real data sets compared to multiple reference models, involving reference models comprising: 1) The matrix factorization model MF (Koren et al IEEE computer Journal 2009) is a classical collaborative filtering model that uses only user and item interactions for characterization learning. 2) LightGCN (He et al, SIGIR 2020) is a graph-based collaborative filtering model, but does not take into account additional features and contextual information of users, items. 3) The decomposer model FM (ICDM 2010) converts all information about interactions into feature vectors and models the user's preferences with the second order interactions of features. 4) NFM (He et al, SIGIR 2017) utilizes a multi-layer perceptron to capture non-linearities and high-order interactions between features of user, item, and contextual information. 5) xDeepFM (Lian et al, SIGKDD 2018) is a recently proposed neural network-based decomposer model that combines explicit and implicit feature high-order interactions. 6) GIN (Li et al, SIGIR 2019) is a graph-based recommendation model that exploits the attention mechanism on a built commodity similarity graph to mine the user's intent. The experimental results are shown in table 2:
table 2 experimental results
It can be seen that all the indicators of the present invention are superior to the existing model over the three data sets, attributing its performance improvement to: 1) The invention carries out embedding (embedding) propagation in the bipartite graph with the attribute, extracts useful information from neighbor nodes to improve the expressive force of a system model; 2) Unlike GIN models with embedded items only propagated on the graph, the present invention integrates the characterization of the user, item and context into the graph for information propagation, thus allowing more uniform characterization to be learned; 3) After accurate characterization is obtained, the present invention further employs FM to explicitly model interactions between features, which can be verified from Table 3. The GCM-MLP in Table 3 shows that the decoder uses MLP and the GCM-MF decoder uses a matrix factorization model.
TABLE 3 validation results for different models
2. The system model has few parameters and the model deducing speed is high: all training parameters of the GCM come from the encoder, i.e. feature embedding of the user, item and context, i.e. P, Q and all v. Assuming that the feature numbers of the three fields of the user, the item and the context are U, I, C, respectively, and the embedded dimension is D, the total parameter amount is (u+i+c) ×d, which is the same as the simplest embedded-based model FM. Thanks to the three-layer architecture of the GCM, after the model is trained on line, the final characterization of all users and objects can be obtained by only performing forward propagation of the graph convolution layer once, and further, only a decoder part is required to be performed during on-line service, so that on-line time complexity similar to that of FM is obtained.
Table 4 gives the time consumption for online service of 1000 Yelp-OH subscribers, it can be seen that GCM is much faster than other CARS models based on neural networks. The model CFM in table 4 (Xin et al, IJCAI 2019) uses the outer product as a pairwise second order interaction between features, followed by extraction of interaction pattern in combination with convolutional neural network.
| Model | FM | GCM | GIN | xDeepFM | CFM |
| Time/second | 8.51 | 14.93 | 35.45 | 365.82 | 2354.25 |
Table 4 on-line service time consuming for different models
3. The graph convolution operation can effectively improve the expression and generalization capability of the model: the graph convolution layer is the core of the GCM, whose rationality and validity can be verified from two aspects: 1) The number of layers of the graph convolutions; 2) Modeling of contextual information. As can be seen from fig. 3, the GCM continuously collects effective information from multi-hop neighbors with increasing layer number, and its expression capability is enhanced, while if the modeling of the context is neglected, the generalization capability of the model is reduced, and an oversmooth phenomenon is easily generated. The model GCM-C in fig. 3 is also a variant of GCM, which in contrast to GCM does not use context information in the picture scroll overlay, but only user and item information.
From the description of the above embodiments, it will be apparent to those skilled in the art that the above embodiments may be implemented in software, or may be implemented by means of software plus a necessary general hardware platform. With such understanding, the technical solutions of the foregoing embodiments may be embodied in a software product, where the software product may be stored in a nonvolatile storage medium (may be a CD-ROM, a U-disk, a mobile hard disk, etc.), and include several instructions for causing a computer device (may be a personal computer, a server, or a network device, etc.) to perform the methods of the embodiments of the present invention.
The foregoing is only a preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions easily contemplated by those skilled in the art within the scope of the present invention should be included in the scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.
Claims (4)
1. A graph convolution recommendation system for context awareness, comprising: an encoder, a layer of graph convolution, and a decoder;
the encoder associates a hidden space vector for each non-zero characteristic of the input user information, article information and context information, combines the hidden space vectors from three fields of the user information, the article information and the context information, and outputs an initial representation of the user and the article and an associated embedded set of the context information;
in the picture scroll lamination, based on a user-object bipartite graph with the attribute which is constructed in advance, carrying out picture scroll operation by combining the output of an encoder, and obtaining the final characteristic representation of the user and the object through a plurality of times of picture scroll operation;
the decoder predicts the preference degree of the user on the article under the context information based on the final characteristic characterization of the user and the article and the associated embedded set of the context information;
the step of constructing a user-item bipartite graph containing user interactions with items includes:
acquiring a mixed data structure consisting of a user and a static portrait thereof, an article and a static attribute thereof, context information and an interaction record;
converting the hybrid data structure into a user-object bipartite graph with attributes; the user-object bipartite graph comprises two types of nodes, namely a user node and an object node, and the characteristics of the user node and the object node are the characteristics of the encoder output; the connection edge of the user node and the article node represents the interaction record of the user node and the article node, and the connection edge is characterized by context characteristics;
the graph rolling operation in the graph rolling layer is expressed as follows:
wherein,and->Respectively representing all the observed interaction record sets of the user u and the object i, (i, c) representing the object and the context information binary group in one interaction record, and (u, c) representing the user and the context information binary group in one interaction record; />The hidden space vectors associated with the context features c representing the encoder output constitute a set, s being the non-zero features in the context information c, v s Representing a hidden space vector associated with a non-zero feature s; />And->Representation regularization;each representing a characteristic representation of a user and an article obtained by the first layer of graph rolling operation; />Each representing a characteristic representation of a user and an article obtained by the layer 1 graph rolling operation;
after the L-layer graph rolling operation, the following modes are adopted for integration, and final characteristic characterization of a user and an object is obtained:
wherein alpha is l Representing the weight of the first layer, p u 、q i The final characteristics of the user u and the object i are respectively represented.
2. The context aware graph convolution recommendation system of claim 1,
the encoder associates one hidden space vector for each non-zero feature of the user information, thereby describing a single user using a plurality of hidden space vectors, after which the encoder combines the plurality of hidden space vectors to obtain an initial representation of the userThe encoder also operates in the same way for the item, giving the initial characterization of the item +.>The formula is:
in the above formula, u and i respectively represent indexes of a user and an article, P represents an embedding matrix which is associated with all user characteristics and consists of hidden space vectors, a kth row represents an embedding vector which is associated with a kth user characteristic, Q represents an embedding matrix which is associated with all article characteristics and consists of hidden space vectors, a first row represents an embedding vector which is associated with a first article characteristic, u and i are respectively characteristic vectors of the user u and the article i after multi-heat coding, and I, u|and I|respectively represent the number of non-zero characteristics of the user u and the article i;
for the context information c, only the hidden space vector is associated, so that an associated embedded set of the context information is obtained, namely, the hidden space vector associated with the context information c forms a setWhere s is a non-zero feature in the context information c, v s Representing the hidden space vector associated with the non-zero feature s.
3. The context aware picture convolution recommendation system according to claim 1, wherein said decoder predicts a user's preference for items in a context scene by the formula:
wherein,preference degree of user u for article i under context information c predicted for decoder; /> Namely->Final characterization of user p including graph convolutional layer output u Final characterization q of the article i And the associated embedding set of context information +.>v k 、v t Are all elements embedded in the set v.
4. A context aware graph convolution recommendation system according to claim 1, characterized in that a log-loss function is chosen as an optimization objective for the system model, formulated as follows:
wherein,preference degree of user u for article i under context information c predicted for decoder; />Representing a set of observed interaction records, and a set of negative sample records randomly matched for each observation record, respectively, σ (·) representing an S-type activation function, λ being L 2 Regularized coefficient, Θ is a trainable parameter.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011249269.5A CN112364242B (en) | 2020-11-10 | 2020-11-10 | Graph convolution recommendation system for context awareness |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011249269.5A CN112364242B (en) | 2020-11-10 | 2020-11-10 | Graph convolution recommendation system for context awareness |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112364242A CN112364242A (en) | 2021-02-12 |
| CN112364242B true CN112364242B (en) | 2024-04-02 |
Family
ID=74509529
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011249269.5A Active CN112364242B (en) | 2020-11-10 | 2020-11-10 | Graph convolution recommendation system for context awareness |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112364242B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114417161B (en) * | 2022-01-21 | 2023-04-07 | 杭州碧游信息技术有限公司 | Virtual article time sequence recommendation method, device, medium and equipment based on special-purpose map |
| CN117349333A (en) * | 2022-06-28 | 2024-01-05 | 华为技术有限公司 | Recommendation method and device and electronic equipment |
| CN115687794A (en) * | 2022-12-29 | 2023-02-03 | 中国科学技术大学 | Student model training method, device, equipment and medium for recommending articles |
| CN116186421B (en) * | 2023-05-04 | 2023-07-14 | 中国科学技术大学 | Recommendation method, system, equipment and storage medium for eliminating popularity deviation |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108874914A (en) * | 2018-05-29 | 2018-11-23 | 吉林大学 | A kind of information recommendation method based on the long-pending and neural collaborative filtering of picture scroll |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11373233B2 (en) * | 2019-02-01 | 2022-06-28 | Target Brands, Inc. | Item recommendations using convolutions on weighted graphs |
-
2020
- 2020-11-10 CN CN202011249269.5A patent/CN112364242B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108874914A (en) * | 2018-05-29 | 2018-11-23 | 吉林大学 | A kind of information recommendation method based on the long-pending and neural collaborative filtering of picture scroll |
Non-Patent Citations (1)
| Title |
|---|
| 杨辉 ; 王月海 ; 豆震泽 ; .一种基于文本卷积神经网络的推荐算法.计算机与现代化.2020,(10),全文. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112364242A (en) | 2021-02-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112364242B (en) | Graph convolution recommendation system for context awareness | |
| Xin et al. | CFM: Convolutional Factorization Machines for Context-Aware Recommendation. | |
| CN108563755A (en) | A kind of personalized recommendation system and method based on bidirectional circulating neural network | |
| CN113255895B (en) | Structure diagram alignment method and multi-diagram joint data mining method based on diagram neural network representation learning | |
| Chen et al. | IR-Rec: An interpretive rules-guided recommendation over knowledge graph | |
| CN112364976A (en) | User preference prediction method based on session recommendation system | |
| Xiao et al. | Link prediction based on feature representation and fusion | |
| Bogaert et al. | Evaluating the importance of different communication types in romantic tie prediction on social media | |
| Pham et al. | Unsupervised training of Bayesian networks for data clustering | |
| CN115964568A (en) | Personalized recommendation method based on edge cache | |
| CN114817712A (en) | Project recommendation method based on multitask learning and knowledge graph enhancement | |
| CN115718826A (en) | Method, system, device and medium for classifying target nodes in graph structure data | |
| Devi et al. | Smoothing approach to alleviate the meager rating problem in collaborative recommender systems | |
| CN111310996B (en) | User trust relation prediction method and system based on graph self-coding network | |
| CN116993513A (en) | Financial wind control model interpretation method and device and computer equipment | |
| CN113744023B (en) | Double-channel collaborative filtering recommendation method based on graph convolution network | |
| CN114461943B (en) | Deep learning-based multi-source POI semantic matching method and device and storage medium thereof | |
| CN114936901A (en) | Visual perception recommendation method and system based on cross-modal semantic reasoning and fusion | |
| CN115391638A (en) | Recommendation model training method and device based on social network | |
| Li et al. | SEDCN: an improved deep & cross network recommendation algorithm based on SENET | |
| CN114519600A (en) | Graph neural network CTR estimation algorithm fusing adjacent node variances | |
| Cheng et al. | BHONEM: Binary high-order network embedding methods for networked-guarantee loans | |
| Karoui et al. | Homologous mesh extraction via monocular systems | |
| Kamani et al. | Cross-domain Recommender Systems via Multimodal Domain Adaptation | |
| Guo et al. | Explainable recommendation systems by generalized additive models with manifest and latent interactions |
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 |