CN117421661A - Group recommendation method of graph convolution network based on inverse fact enhancement - Google Patents
Group recommendation method of graph convolution network based on inverse fact enhancement Download PDFInfo
- Publication number
- CN117421661A CN117421661A CN202311744970.8A CN202311744970A CN117421661A CN 117421661 A CN117421661 A CN 117421661A CN 202311744970 A CN202311744970 A CN 202311744970A CN 117421661 A CN117421661 A CN 117421661A
- Authority
- CN
- China
- Prior art keywords
- group
- representation
- user
- level
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 45
- 238000012549 training Methods 0.000 claims abstract description 25
- 238000005096 rolling process Methods 0.000 claims abstract description 14
- 238000005065 mining Methods 0.000 claims abstract description 11
- 239000011159 matrix material Substances 0.000 claims description 14
- 230000003993 interaction Effects 0.000 claims description 11
- 230000014509 gene expression Effects 0.000 claims description 7
- 230000007246 mechanism Effects 0.000 claims description 6
- 230000002776 aggregation Effects 0.000 claims description 5
- 238000004220 aggregation Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 230000003416 augmentation Effects 0.000 claims 1
- 230000003094 perturbing effect Effects 0.000 claims 1
- 230000001364 causal effect Effects 0.000 description 5
- 230000006870 function Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 3
- 230000004927 fusion Effects 0.000 description 3
- 238000012935 Averaging Methods 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001537 neural effect Effects 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
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
-
- 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
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/01—Social networking
-
- 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
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Business, Economics & Management (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to the technical field of group recommendation, and provides a group recommendation method of a graph rolling network based on inverse fact enhancement, which comprises the following steps: constructing a group user hypergraph and a group project bipartite graph; the hypergraph of the group users is mined and the high-order information between the groups and the users is obtained to obtain user-level group representation; disturbance and convolution are carried out on the hypergraph of the group user, and a counterfactual group representation is obtained; mining high-order information between the group and the project for the group project bipartite graph and the user-level group representation, obtaining group-level group representation and project representation, and weighting and fusing the user-level group representation and the group-level group representation to obtain final group representation; training the hierarchical graph convolutional network based on the inverse fact loss, the group loss and the user loss obtained by the representation layer-by-layer training, and obtaining a recommended item list through the obtained prediction model. The invention can accurately model the characteristics of the group, the user and the project explicitly, and also enhances the interpretability of group recommendation and the adaptability of the model.
Description
Technical Field
The invention relates to the technical field of group recommendation, in particular to a group recommendation method of a graph rolling network based on inverse fact enhancement.
Background
The recommendation system is widely applied to online information systems such as an electronic commerce platform, a social media website, a news portal and the like as a powerful tool for solving the information overload problem so as to help a user select from various options in daily life. The recommender system collects the past preferences of the user and generates appropriate project recommendations to provide to the user. An effective recommendation would not only increase the flow and profits for the service provider, but would also help the user more easily find items of interest.
With the popularity of social media, online group activities have become exceedingly common in current social networks. However, the conventional personalized recommendation algorithm is mainly designed according to the preference of the individual user, and is difficult to meet the requirements of the group user. In a group, each member's interests and preferences may have an impact on the final group decision. Therefore, there is an urgent need for a recommendation system specifically designed for group users, i.e., a group recommendation system. The system considers the interaction and influence of the members in the group, and can provide personalized recommendation service for the group more accurately.
In the group recommendation process, it is a critical task to effectively aggregate preferences of all members in a group to form group preferences. Conventional aggregation methods are typically based on predefined heuristic rules, including fairness policies, minimum pain policies, maximum satisfaction policies, and the like. While these data-independent static policies can meet the formation of group preferences to some extent, in practice group decision is a complex dynamic process that requires consideration of the weight of each member. The development of neural attention mechanisms provides a flexible way to solve this problem. Recently, complex graph neural networks and innovative hypercube structures, combined with techniques such as contrast learning, have made significant progress in group recommendation methods, successfully solving specific challenges and existing problems.
However, the current group recommendation method is generally established on a statistical framework, mainly considers the statistical relationship between groups and projects, ignores potential causal relationships, and is limited to result in inaccurate final recommendation results, while the inverse fact learning is used as a method in causal inference, and has a great application prospect in group recommendation in causal mining, but the performance of the existing method is poor along with the change of data sparsity, and the adaptability to data with different sparsity is also insufficient.
Disclosure of Invention
The present invention is directed to solving at least one of the technical problems existing in the related art. To this end, the invention provides a group recommendation method for a graph roll-up network based on inverse fact enhancement.
The invention provides a group recommendation method of a graph rolling network based on inverse fact enhancement, which comprises the following steps:
s1: preprocessing the public group data to obtain a negative example data set, and constructing a group user hypergraph and a group project bipartite graph;
s2: the hypergraph convolution of the group users is used for mining the high-order information between the group and the users, and user-level group representation is obtained; disturbing the group user hypergraph, and convolving the disturbed group user hypergraph to obtain a counterfactual group representation;
s3: the group project bipartite graph and the user-level group representation are subjected to convolution mining to obtain group-level group representation and group-level project representation, the user-level group representation and the group-level group representation are weighted and fused through a residual error gating mechanism, and final group representation is obtained;
s4: obtaining a back facts loss based on the back facts group representation, the user-level group representation and the initial item representation, obtaining a group loss based on the final group representation and the group-level item representation, obtaining a user loss based on the initial user representation and the initial item representation, and performing multiple training on the hierarchical graph convolutional network according to the back facts loss, the group loss and the user loss to obtain a group prediction model;
s5: and inputting the item list to be recommended into the prediction model, obtaining a prediction score, and sorting the recommended item list based on the prediction score.
According to the group recommendation method of the graph rolling network based on the inverse fact enhancement, in the step S1, the group user hypergraph is constructed according to the group user relationship data, and the group project bipartite graph is constructed according to the group project interaction data.
According to the group recommendation method of the graph rolling network based on the inverse fact enhancement provided by the invention, in step S1, the negative example data comprises an initial representation set constructed based on an ID embedded vector lookup table, and the initial representation set comprises an initial group representation, an initial user representation and an initial item representation.
According to the group recommendation method of the graph rolling network based on the inverse fact enhancement provided by the invention, in the step S2, the expression of the user-level group representation is as follows:
wherein,for user-level group representation, < >>For the total number of convolution layers of the hypergraph convolution network, < >>Index value of convolution layer for hypergraph convolution network, < ->Is->Aggregation information represented by nodes updated by the convolutional layer.
According to the group recommendation method of the graph rolling network based on the inverse fact enhancement provided by the invention, in the step S2, the calculation formula for disturbing the hypergraph of the group user is as follows:
wherein,hypergraph incidence matrix for disturbed group users,>hypergraph association matrix for group user, +.>As an index function->Is a trainable mask matrix +.>The threshold value specified for the entry is determined for the indicator function.
According to the group recommendation method of the graph rolling network based on the inverse fact enhancement provided by the invention, the group-level group representation and the group-level item representation obtained in the step S3 are as follows:
wherein,for group-level group representation, +.>For group level item representation, < >>Is->The node representation obtained by the convolution layer,to be the instituteThere is an average of node embedded representations obtained by the convolution layer.
According to the group recommendation method of the graph rolling network based on the inverse fact enhancement provided by the invention, the expression of the final group representation in the step S3 is as follows:
wherein,for the final group representation, +.>Representing the occupied weight for the user-level group, < >>For sigmoid function, +.>For the first gating factor,/>For the second gating factor, +>Is the third weight coefficient.
According to the group recommendation method of the graph convolution network based on the inverse fact enhancement provided by the invention, in the step S4, the loss for performing multiple training by the layered graph convolution network comprises the following steps:
wherein,to counter the loss of facts->For group loss, ++>For user loss->Training sets for group items->Data in training set for group item, +.>Training set for user item->Data in training set for user item, +.>For item index value, & lt + & gt>For the sampling negative case corresponding to the item interacted with by the group, +.>For the sample negative case corresponding to the item interacted with by the user, +.>Predicting a score for a fact->Predicting a score for a counterfactual->Representing predictive scores for final groups, +.>Representing a predictive score for a group level item, +.>The user initially represents the predictive score +.>The predictive score is initially represented for the item.
Aiming at the defects of the current group recommendation model in terms of causality mining and adaptation to different sparsity data, the group recommendation method based on the inverse fact enhanced graph convolution network provided by the invention is used for mining the causality relationship between group members and project recommendation, exploring the basic principle behind the group recommendation and improving the adaptation capability of the model to the different sparsity data.
The invention provides a group recommendation method CF-hGCN based on a reinforced anti-facts layered graph convolution network, which is mainly applied to group recommendation tasks, and the CF-hGCN can deeply mine causal relations between group members and project recommendations by integrating graph anti-facts learning technology and the super graph convolution network in an anti-facts super graph learning module.
The CF-hGCN adopts an innovative hierarchical network structure, and the representations of the group, the user and the project are calculated through the inverse fact hypergraph learning module and the two graph learning modules in sequence. The hypergraph learning module can effectively capture preferences of members according to the highly sparse data, and the two graph learning modules can better capture consensus of groups according to the medium-low sparse data. Subsequently, the group representations from the two levels are balanced in a self-adaptive manner by means of a carefully designed residual error gating mechanism, so that the model can adapt to data with different sparsities, the characteristics of the group, the user and the project can be modeled more accurately explicitly, the performance on a group recommendation task is improved obviously, and the adaptability of the model to the data with different sparsities is enhanced obviously.
Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.
Drawings
In order to more clearly illustrate the invention or the technical solutions of the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described, and it is obvious that the drawings in the description below are some embodiments of the invention, and other drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a flowchart of a group recommendation method of a graph rolling network based on inverse fact enhancement.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is apparent that the described embodiments are some embodiments of the present invention, not all embodiments. 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. The following examples are illustrative of the invention but are not intended to limit the scope of the invention.
In the description of the embodiments of the present invention, it should be noted that the terms "center", "longitudinal", "lateral", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings, are merely for convenience in describing the embodiments of the present invention and simplifying the description, and do not indicate or imply that the apparatus or elements referred to must have a specific orientation, be configured and operated in a specific orientation, and thus should not be construed as limiting the embodiments of the present invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.
In describing embodiments of the present invention, it should be noted that, unless explicitly stated and limited otherwise, the terms "coupled," "coupled," and "connected" should be construed broadly, and may be either a fixed connection, a removable connection, or an integral connection, for example; can be mechanically or electrically connected; can be directly connected or indirectly connected through an intermediate medium. The specific meaning of the above terms in embodiments of the present invention will be understood in detail by those of ordinary skill in the art.
In embodiments of the invention, unless expressly specified and limited otherwise, a first feature "up" or "down" on a second feature may be that the first and second features are in direct contact, or that the first and second features are in indirect contact via an intervening medium. Moreover, a first feature being "above," "over" and "on" a second feature may be a first feature being directly above or obliquely above the second feature, or simply indicating that the first feature is level higher than the second feature. The first feature being "under", "below" and "beneath" the second feature may be the first feature being directly under or obliquely below the second feature, or simply indicating that the first feature is less level than the second feature.
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 embodiments of the present invention. In this specification, schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, the different embodiments or examples described in this specification and the features of the different embodiments or examples may be combined and combined by those skilled in the art without contradiction.
An embodiment of the present invention is described below with reference to fig. 1.
The invention provides a group recommendation method of a graph rolling network based on inverse fact enhancement, which comprises the following steps:
s1: preprocessing the public group data to obtain negative example data, and constructing a group user hypergraph and a group project bipartite graph;
in step S1, the group user hypergraph is constructed according to the group user relationship data, and the group project bipartite graph is constructed according to the group project interaction data.
In step S1, the negative example data includes an initial representation set constructed based on an ID embedded vector lookup table, where the initial representation set includes an initial group representation, an initial user representation, and an initial item representation.
Further, at this stage, the group recommendation data set includes user history interaction data, group history interaction data, and group user relationship data. The goal of the stage is to construct a group user hypergraph according to the group user relationship data, and simultaneously construct a group project bipartite graph according to the group project interaction data; constructing an embedded vector lookup table of users, groups and items; non-interacted user (group) negative examples are randomly sampled from the dataset, and the dataset for training is constructed.
The method specifically comprises the following steps: training data is first constructed, and since the loss function uses a pair of BPR losses, multiple negative examples need to be randomly sampled from the non-interacted items for each user (group) interaction pair with the item to form the training data required by the model of the present invention.
And secondly, constructing a group user hypergraph and a group project bipartite graph, and constructing an association matrix corresponding to the group user hypergraph according to the group user relationship in the data set in order to efficiently mine high-order information among the group, the users and the projects, wherein a value of 1 indicates that the group contains the user, and a value of 0 indicates that the group does not contain the user. And constructing an adjacent matrix corresponding to the bipartite graph of the group item according to the interaction of the group item in the training data, wherein a value of 1 indicates that the group has interaction with the item, and a value of 0 indicates that no interaction exists.
Finally, an embedded vector lookup table is constructed, based on the number of users, groups, and items, an ID embedded vector lookup table is constructed as its respective initial representation.
S2: the hypergraph convolution of the group users is used for mining high-order information among the group and a plurality of users among the users, and user-level group representation is obtained; disturbing the group user hypergraph, and convolving the disturbed group user hypergraph to obtain a counterfactual group representation;
further, the user-level group representation is a member-level group representation, and the goal of this stage is to construct a member-preference anti-facts hypergraph learning module, extract high-order information between the group and the user by using a hypergraph convolution operation to obtain the member-level group representation, and combine the hypergraph neural network with the anti-facts learning by the member-preference anti-facts hypergraph learning module to aim at mining causal relationships between group members and project recommendations. By minimizing the anti-facts losses, the fact prediction score calculated from the fact and anti-fact group representations is made as large as possible compared to the anti-fact prediction score.
In step S2, the expression of the user-level group representation is:
wherein,for user-level group representation, < >>For the total number of convolution layers of the hypergraph convolution network, < >>Index value of convolution layer for hypergraph convolution network, < ->Is->Aggregation information represented by nodes updated by the convolutional layer.
In step S2, the calculation formula for disturbing the hypergraph of the group user is as follows:
wherein,hypergraph incidence matrix for disturbed group users,>hypergraph association matrix for group user, +.>As an index function->Is a trainable mask matrix +.>The threshold value specified for the entry is determined for the indicator function.
The method specifically comprises the following steps: firstly, obtaining member-level group representation by supergraph convolution of constructed group user supergraph and initial representation of users and groups, then for enhancing expressive force, averaging embedded representations obtained by each layer to obtain user representation of the level, and similarly, averaging aggregation information of nodes updated by the previous layer in the message transmission process on each convolution layer to obtain group representation of the level.
And secondly, obtaining the inverse fact group representation, disturbing the original hypergraph correlation matrix through a trainable mask matrix, and obtaining the group representation in the inverse fact scene by the hypergraph convolution operation with the same steps based on the obtained disturbance hypergraph correlation matrix.
Finally, calculating the loss of the facts and optimizing, and passing through a three-layer multi-layer perceptron based on different group representations and initial item representations in the facts and facts scenesCalculating facts and anti-facts prediction scores, and then calculating anti-facts loss +.>。
S3: the group project bipartite graph and the user-level group representation are subjected to convolution mining to obtain group-level group representation and group-level project representation, the user-level group representation and the group-level group representation are weighted and fused through a residual error gating mechanism, and final group representation is obtained;
further, the stage mainly comprises a group consensus diagram learning module and a residual fusion module, wherein the goal of the stage is to construct the group consensus diagram learning module and the residual fusion module, extract high-order information between groups and projects by using a diagram rolling operation to obtain a group-level group representation, and enhance the representation of a model when more data is available by using a supervision signal between the group projects. And fusing the group representations of different levels into a final group representation through a residual fusion module. By minimizing the group loss, the predictive score of the group for positive samples is made as greater as possible than for negative samples.
Wherein the expressions of the group-level group representation and the group-level item representation obtained in step S3 are:
wherein,for group-level group representation, +.>For group level item representation, < >>Is->The node representation obtained by the convolution layer,the average of the node embedded representations obtained for all convolution layers.
Wherein the expression of the final group representation in step S3 is:
wherein,for the final group representation, +.>Representing the occupied weight for the user-level group, < >>For sigmoid function, +.>For the first gating factor,/>For the second gating factor, +>Is the third weight coefficient.
The method specifically comprises the following steps: firstly, a group-level group representation is obtained through graph convolution of a constructed group project bipartite graph, a calculated group representation and an initial representation of the project, and then the obtained embedment of each layer is averaged to obtain a final embedment.
Secondly, calculating the weights of all parts by using a gating mechanism through the calculated member-level group representation and the calculated group-level group representation, and taking up the weights of the member-level group representationMultiplied by itself and added with the weight of the group-level group representation and its occupation>The result of the multiplication results in a final group representation.
Finally, inputting the final group representation and the group level item representation into a multi-layer perceptron to calculate a prediction score, and then calculating the group loss。
S4: obtaining a back facts loss based on the back facts group representation, the user-level group representation and the initial item representation, obtaining a group loss based on the final group representation and the group-level item representation, obtaining a user loss based on the initial user representation and the initial item representation, and performing multiple training on the hierarchical graph convolutional network according to the back facts loss, the group loss and the user loss to obtain a group prediction model;
in step S4, the loss for performing multiple training by the hierarchical graph convolutional network includes:
wherein,to counter the loss of facts->For group loss, ++>For user loss->Training sets for group items->Data in training set for group item, +.>Training set for user item->Data in training set for user item, +.>For item index value, & lt + & gt>For the sampling negative case corresponding to the item interacted with by the group, +.>For the sample negative case corresponding to the item interacted with by the user, +.>Predicting a score for a fact->Predicting a score for a counterfactual->Representing predictive scores for final groups, +.>Representing a predictive score for a group level item, +.>The user initially represents the predictive score +.>The predictive score is initially represented for the item.
S5: and inputting group data to be recommended into the prediction model to obtain a prediction score, and sorting to obtain a recommended item list based on the prediction score.
Further, after multiple rounds of combined training, a trained model is obtained, any group and item representation needing to be predicted are taken as input, corresponding prediction scores are output, and the prediction scores of the group on all candidate items are ranked from high to low, and the top K are taken to obtain a final recommended item list.
Further, it also includes obtaining an initial representation of the user and the item, by minimizing the user loss such that the user's predictive score for the positive sample is as greater as possible than the predictive score for the negative sample, specifically, inputting the initial representation of the user and the item into the multi-layer perceptron to calculate the predictive score, and then calculating the user lossAnd optimized.
In some embodiments, the present invention has been tested using two widely used real world group recommendation data sets and one semi-synthetic group recommendation data set, wherein the basic statistical properties of the data sets used are shown in Table 1.
Table 1 basic statistical properties of the data set used in the examples of the present invention
The experiment is divided into two aspects: item recommendation is performed on a group of users, and item recommendation is performed on individual users, respectively referred to as a group recommendation task and a user recommendation task. In both experiments, we used two commonly used evaluation indices: results of HR (hit rate) and NDCG (normalized loss cumulative gain) obtained for the method provided by the present invention and other methods in the group recommendation field on each task of each dataset are shown in tables 2 and 3.
Table 2 comparison of the method of the present invention with other methods in the group recommendation field in the first set of experimental results of the dataset
Table 3 comparison of the method of the present invention with other methods in the group recommendation field in the second set of experimental results of the dataset
The experimental results in tables 2 and 3 show that the proposed method CF-hGCN performs better than the previous method, and the performance is improved to different degrees, and the maximum improvement amplitude is 12.7%. The comparison result fully proves that the method provided by the user has excellent effects on group recommendation and user recommendation tasks.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (8)
1. A method for group recommendation of a graph roll-up network based on counterfactual augmentation, comprising:
s1: preprocessing the public group data to obtain negative example data, and constructing a group user hypergraph and a group project bipartite graph;
s2: the hypergraph convolution of the group users is used for mining the high-order information between the group and the users, and user-level group representation is obtained; disturbing the group user hypergraph, and convolving the disturbed group user hypergraph to obtain a counterfactual group representation;
s3: the group project bipartite graph and the user-level group representation are subjected to convolution mining to obtain group-level group representation and group-level project representation, the user-level group representation and the group-level group representation are weighted and fused through a residual error gating mechanism, and final group representation is obtained;
s4: obtaining a back facts loss based on the back facts group representation, the user-level group representation and the initial item representation, obtaining a group loss based on the final group representation and the group-level item representation, obtaining a user loss based on the initial user representation and the initial item representation, and performing multiple training on the hierarchical graph convolutional network according to the back facts loss, the group loss and the user loss to obtain a group prediction model;
s5: and inputting the item list to be recommended into the prediction model, obtaining a prediction score, and sorting the recommended item list based on the prediction score.
2. The method for group recommendation in a graph rolling network based on inverse fact enhancement according to claim 1, wherein in step S1, the group user hypergraph is constructed according to group user relationship data, and the group project bipartite graph is constructed according to group project interaction data.
3. The method of claim 1, wherein in step S1, the negative example data includes an initial representation set constructed based on an ID-embedded vector lookup table, the initial representation set including an initial group representation, an initial user representation, and an initial item representation.
4. The method of claim 1, wherein in step S2, the expression of the user-level group representation is:
wherein,for user-level group representation, < >>For the total number of convolution layers of the hypergraph convolution network, < >>Index value of convolution layer for hypergraph convolution network, < ->Is->Aggregation information represented by nodes updated by the convolutional layer.
5. The method for group recommendation in a graph rolling network based on inverse fact enhancement according to claim 4, wherein in step S2, the calculation formula for perturbing the hypergraph of the group user is:
wherein,hypergraph incidence matrix for disturbed group users,>hypergraph association matrix for group user, +.>As a function of the index(s),is a trainable mask matrix +.>The threshold value specified for the entry is determined for the indicator function.
6. The method of claim 5, wherein the expressions of the group-level group representation and the group-level item representation obtained in step S3 are:
wherein,for group-level group representation, +.>For group level item representation, < >>Is->Node representation obtained by convolution layer, < >>The average of the node embedded representations obtained for all convolution layers.
7. The method of claim 6, wherein the final group representation in step S3 is expressed as:
wherein,for the final group representation, +.>Representing the occupied weight for the user-level group, < >>For sigmoid function, +.>For the first gating factor,/>For the second gating factor, +>Is the third weight coefficient.
8. The method of claim 7, wherein in step S4, the penalty for multiple training passes of the hierarchical graph convolutional network comprises:
wherein,to counter the loss of facts->For group loss, ++>For user loss->Training sets for group items->Data in training set for group item, +.>Training set for user item->Data in training set for user item, +.>For item index value, & lt + & gt>For the sampling negative case corresponding to the item interacted with by the group, +.>For the sample negative case corresponding to the item interacted with by the user, +.>Predicting a score for a fact->Predicting a score for a counterfactual->Representing predictive scores for final groups, +.>Representing a predictive score for a group level item, +.>The user initially represents the predictive score +.>The predictive score is initially represented for the item.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311744970.8A CN117421661B (en) | 2023-12-19 | 2023-12-19 | Group recommendation method of graph convolution network based on inverse fact enhancement |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311744970.8A CN117421661B (en) | 2023-12-19 | 2023-12-19 | Group recommendation method of graph convolution network based on inverse fact enhancement |
Publications (2)
Publication Number | Publication Date |
---|---|
CN117421661A true CN117421661A (en) | 2024-01-19 |
CN117421661B CN117421661B (en) | 2024-02-13 |
Family
ID=89525184
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202311744970.8A Active CN117421661B (en) | 2023-12-19 | 2023-12-19 | Group recommendation method of graph convolution network based on inverse fact enhancement |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117421661B (en) |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114090890A (en) * | 2021-11-23 | 2022-02-25 | 电子科技大学 | Counterfactual project recommendation method based on graph convolution network |
US20220083853A1 (en) * | 2020-09-15 | 2022-03-17 | Microsoft Technology Licensing, Llc | Recommending edges via importance aware machine learned model |
US20220253721A1 (en) * | 2021-01-30 | 2022-08-11 | Walmart Apollo, Llc | Generating recommendations using adversarial counterfactual learning and evaluation |
CN114936890A (en) * | 2022-03-31 | 2022-08-23 | 合肥工业大学 | Counter-fact fairness recommendation method based on inverse tendency weighting method |
CN116506302A (en) * | 2023-04-27 | 2023-07-28 | 河南科技大学 | Network alignment method based on inverse fact inference |
CN116888602A (en) * | 2020-12-17 | 2023-10-13 | 乌姆奈有限公司 | Interpretable transducer |
CN117112905A (en) * | 2023-09-01 | 2023-11-24 | 华中科技大学 | Sensitive attribute filtering fairness recommendation method and device based on bilateral countermeasure learning |
-
2023
- 2023-12-19 CN CN202311744970.8A patent/CN117421661B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20220083853A1 (en) * | 2020-09-15 | 2022-03-17 | Microsoft Technology Licensing, Llc | Recommending edges via importance aware machine learned model |
CN116888602A (en) * | 2020-12-17 | 2023-10-13 | 乌姆奈有限公司 | Interpretable transducer |
US20220253721A1 (en) * | 2021-01-30 | 2022-08-11 | Walmart Apollo, Llc | Generating recommendations using adversarial counterfactual learning and evaluation |
CN114090890A (en) * | 2021-11-23 | 2022-02-25 | 电子科技大学 | Counterfactual project recommendation method based on graph convolution network |
CN114936890A (en) * | 2022-03-31 | 2022-08-23 | 合肥工业大学 | Counter-fact fairness recommendation method based on inverse tendency weighting method |
CN116506302A (en) * | 2023-04-27 | 2023-07-28 | 河南科技大学 | Network alignment method based on inverse fact inference |
CN117112905A (en) * | 2023-09-01 | 2023-11-24 | 华中科技大学 | Sensitive attribute filtering fairness recommendation method and device based on bilateral countermeasure learning |
Non-Patent Citations (2)
Title |
---|
杨梦月;何洪波;王闰强;: "基于反事实学习及混淆因子建模的文章个性化推荐", 计算机系统应用, no. 10, 13 October 2020 (2020-10-13), pages 53 - 60 * |
郭文雅,张莹等: "一种面向指代短语理解的关系聚合网络", 《计算机研究与发展》, vol. 60, no. 11, 7 March 2023 (2023-03-07), pages 2611 * |
Also Published As
Publication number | Publication date |
---|---|
CN117421661B (en) | 2024-02-13 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Parvin et al. | TCFACO: Trust-aware collaborative filtering method based on ant colony optimization | |
CN111127142B (en) | Article recommendation method based on generalized nerve attention | |
CN112084427A (en) | Interest point recommendation method based on graph neural network | |
CN112990972B (en) | Recommendation method based on heterogeneous graph neural network | |
CN109190030B (en) | Implicit feedback recommendation method fusing node2vec and deep neural network | |
CN112989064A (en) | Recommendation method for aggregating knowledge graph neural network and self-adaptive attention | |
CN112507246B (en) | Social recommendation method fusing global and local social interest influence | |
CN110955829A (en) | Interest point recommendation method and system fusing credibility and measurement factor matrix decomposition | |
CN113343119B (en) | Group recommendation method based on hierarchical attention mechanism | |
Wang et al. | Deep user modeling for content-based event recommendation in event-based social networks | |
Alhamdani et al. | Recommender system for global terrorist database based on deep learning | |
CN112819024B (en) | Model processing method, user data processing method and device and computer equipment | |
CN113918832A (en) | Graph convolution collaborative filtering recommendation system based on social relationship | |
Linda et al. | Effective context-aware recommendations based on context weighting using genetic algorithm and alleviating data sparsity | |
CN111475744A (en) | Personalized position recommendation method based on ensemble learning | |
Leng et al. | Dynamically aggregating individuals’ social influence and interest evolution for group recommendations | |
CN113590976A (en) | Recommendation method of space self-adaptive graph convolution network | |
CN112364245B (en) | Top-K movie recommendation method based on heterogeneous information network embedding | |
CN113868537A (en) | Recommendation method based on multi-behavior session graph fusion | |
Wang et al. | Social dual-effect driven group modeling for neural group recommendation | |
Yang et al. | Gated graph convolutional network based on spatio-temporal semi-variogram for link prediction in dynamic complex network | |
CN117421661B (en) | Group recommendation method of graph convolution network based on inverse fact enhancement | |
CN116306834A (en) | Link prediction method based on global path perception graph neural network model | |
Dai et al. | DAS-GNN: Denoising autoencoder integrated with self-supervised learning in graph neural network-based recommendations | |
CN113256024B (en) | User behavior prediction method fusing group behaviors |
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 |